<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" version="2.0">
  <channel>
    <title>AKKA Blogs</title>
    <link>https://akka.io/blog</link>
    <description />
    <language>en</language>
    <pubDate>Wed, 04 Nov 2026 20:08:31 GMT</pubDate>
    <dc:date>2026-11-04T20:08:31Z</dc:date>
    <dc:language>en</dc:language>
    <item>
      <title>14 Human-in-the-Loop Patterns for AI Agents</title>
      <link>https://akka.io/blog/human-in-the-loop-ai-agents</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/human-in-the-loop-ai-agents" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/blog-band-human-oversight.jpg" alt="14 Human Oversight Patterns for AI Systems" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="body col"&gt; 
 &lt;div class="enote"&gt; 
  &lt;strong&gt;TL;DR&lt;/strong&gt; 
  &lt;ul&gt; 
   &lt;li&gt;&lt;span&gt;Human-in-the-loop AI agents hand some decisions to a person, such as approving a payment before it is sent.&lt;/span&gt;&lt;/li&gt; 
   &lt;li&gt;&lt;span&gt;We studied AI regulations, our customers, our own systems, and published applications, and found 206 human controls across 225 regulations.&lt;/span&gt;&lt;/li&gt; 
   &lt;li&gt;&lt;span&gt;Almost all of them fit one of 14 patterns, each set by who decides, the deadline, and what happens if nobody decides.&lt;/span&gt;&lt;/li&gt; 
  &lt;/ul&gt; 
 &lt;/div&gt;  
 &lt;p class="toc-label"&gt;Sections&lt;/p&gt; 
 &lt;ul&gt; 
  &lt;li&gt;&lt;a href="#rules-beyond-the-wait"&gt;&lt;span class="num"&gt;I&lt;/span&gt;&lt;span class="ttl"&gt;Rules beyond the wait&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#sources-and-method"&gt;&lt;span class="num"&gt;II&lt;/span&gt;&lt;span class="ttl"&gt;Sources and method&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#the-oversight-event-and-its-settings"&gt;&lt;span class="num"&gt;III&lt;/span&gt;&lt;span class="ttl"&gt;The oversight event and its settings&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-1-action-approval"&gt;&lt;span class="num"&gt;IV&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 1 - Action approval&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-2-decision-review"&gt;&lt;span class="num"&gt;V&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 2 - Decision review&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-3-change-authorization"&gt;&lt;span class="num"&gt;VI&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 3 - Change authorization&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-4-exception"&gt;&lt;span class="num"&gt;VII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 4 - Exception&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-5-match-and-adjudication"&gt;&lt;span class="num"&gt;VIII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 5 - Match and adjudication&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-6-outcome-reconciliation"&gt;&lt;span class="num"&gt;IX&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 6 - Outcome reconciliation&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-7-contest-and-appeal"&gt;&lt;span class="num"&gt;X&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 7 - Contest and appeal&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-8-override"&gt;&lt;span class="num"&gt;XI&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 8 - Override&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-9-flag-and-feedback"&gt;&lt;span class="num"&gt;XII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 9 - Flag and feedback&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-10-triage"&gt;&lt;span class="num"&gt;XIII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 10 - Triage&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-11-hand-off"&gt;&lt;span class="num"&gt;XIV&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 11 - Hand-off&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-12-halt-and-restart"&gt;&lt;span class="num"&gt;XV&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 12 - Halt and restart&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-13-attestation"&gt;&lt;span class="num"&gt;XVI&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 13 - Attestation&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-14-oversight-quality"&gt;&lt;span class="num"&gt;XVII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 14 - Oversight quality&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#reviewer-screens"&gt;&lt;span class="num"&gt;XVIII&lt;/span&gt;&lt;span class="ttl"&gt;Reviewer screens&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#implementation"&gt;&lt;span class="num"&gt;XIX&lt;/span&gt;&lt;span class="ttl"&gt;Implementation&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#lessons-learned"&gt;&lt;span class="num"&gt;XX&lt;/span&gt;&lt;span class="ttl"&gt;Lessons learned&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#limits-of-the-evidence"&gt;&lt;span class="num"&gt;XXI&lt;/span&gt;&lt;span class="ttl"&gt;Limits of the evidence&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#conclusion"&gt;&lt;span class="num"&gt;XXII&lt;/span&gt;&lt;span class="ttl"&gt;Conclusion&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#appendix-a-human-controls-in-the-corpus-by-pattern"&gt;&lt;span class="num"&gt;XXIII&lt;/span&gt;&lt;span class="ttl"&gt;Appendix A: Human controls in the corpus by pattern&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#appendix-b-the-decision-vocabulary"&gt;&lt;span class="num"&gt;XXIV&lt;/span&gt;&lt;span class="ttl"&gt;Appendix B: The decision vocabulary&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
 &lt;/ul&gt;  
 &lt;p class="lede"&gt;Every AI system we have helped build reaches a point where a person has to decide. A payments agent matches an invoice to a purchase order and asks to pay the supplier 4,200 EUR. A notices agent drafts an adverse decision letter to a loan applicant. A model team proposes sending a new model 10% of production traffic. The payment, the letter and the rollout wait for a person to decide before the work continues.&lt;/p&gt; 
 &lt;p&gt;Our customers build each human decision with the Akka SDK. A workflow step pauses for the person and resumes when the decision arrives. A timeout on the pause can send a reminder or move the decision to someone more senior. An autonomous agent can hand a task to a person and wait for the result. The workflow and the task are durable, so a wait that lasts days survives a restart of the service.&lt;/p&gt; 
 &lt;p&gt;Until now, each of these human processes was designed individually, as its own Akka workflow. The team building the application decided who may answer, how long the answer may take, what happens when nobody answers, and what record the answer leaves. Here are some examples in real-world AI systems we have helped build where a human-in-the-loop control existed:&lt;/p&gt; 
 &lt;ol&gt; 
  &lt;li&gt;A beverage manufacturer's data-use governance system sends every recommendation to a dataset owner or the governance lead, who confirms, overrides or escalates it with a mandatory reason. The reminder after three days and the escalation after seven are application code around a workflow pause.&lt;/li&gt; 
  &lt;li&gt;A medical-supply company's voice agent routes a call to a person at 21 points in its four call flows.&lt;/li&gt; 
 &lt;/ol&gt; 
 &lt;p&gt;We did the same thing in our own services. A loan origination service, a trace ledger and a context layer, all built at Akka, each implemented the same oversight contract by hand. The loan origination service and the trace ledger marked a decision at risk when 75% of its time had passed, and the context layer marked it at 80%.&lt;/p&gt; 
 &lt;p&gt;When we looked in our own regulatory corpus in &lt;a href="https://akka.io/platform/verify"&gt;Akka Verify&lt;/a&gt;, we found 206 controls that require a person to act. A control states one obligation from a regulation or framework and names the mechanism that enforces it. Our own schema had no way to describe what that person decides.&lt;/p&gt; 
 &lt;p&gt;We also read the documentation of 34 products in AI governance, LLM operations and cloud agent services. Only three AI governance platforms among them turn a monitoring signal into a work item that someone owns and closes with a recorded outcome.&lt;/p&gt; 
 &lt;p&gt;Across our customers' systems, our own services and the regulations, the same human processes kept recurring. We found fourteen recurring patterns, and each one is described below with the obligation behind it, its settings and a working example. Each pattern raises an oversight event, the record of one decision a person must make, and every oversight event moves through one set of states. The patterns differ in who decides, the deadline, the outcome when nobody decides, the allowed decisions and the record kept.&lt;/p&gt; 
 &lt;p&gt;The fourteen patterns will soon be available as templates in the Akka Agentic AI Platform.&lt;/p&gt; 
 &lt;table class="rtbl"&gt; 
  &lt;colgroup&gt; 
   &lt;col style="width: 24%;"&gt; 
   &lt;col style="width: 20%;"&gt; 
   &lt;col style="width: 56%;"&gt; 
  &lt;/colgroup&gt; 
  &lt;thead&gt; 
   &lt;tr&gt; 
    &lt;th&gt;Pattern&lt;/th&gt; 
    &lt;th&gt;When the person acts&lt;/th&gt; 
    &lt;th&gt;What happens&lt;/th&gt; 
   &lt;/tr&gt; 
  &lt;/thead&gt; 
  &lt;tbody&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;1. Action approval&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person approves an action before the system takes it.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;2. Decision review&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person reviews a decision about someone before it takes effect.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;3. Change authorization&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;Two people who neither proposed nor built a change approve it before it reaches production.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;4. Exception&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person grants a time-limited exception to a rule, with a reason.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;5. Match and adjudication&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person decides whether two records describe the same thing.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;6. Outcome reconciliation&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person establishes what happened when the system cannot tell.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;7. Contest and appeal&lt;/td&gt; 
    &lt;td class="prop"&gt;After the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person affected by a decision appeals it to someone independent of it.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;8. Override&lt;/td&gt; 
    &lt;td class="prop"&gt;After the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person sets aside or replaces an output the system produced, and says why.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;9. Flag and feedback&lt;/td&gt; 
    &lt;td class="prop"&gt;After the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;Anyone can flag an output as wrong, and a triager decides whether it becomes an evaluation case.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;10. Triage&lt;/td&gt; 
    &lt;td class="prop"&gt;While running&lt;/td&gt; 
    &lt;td class="prop"&gt;A monitor raises a signal, and a person decides whether it is a problem while the work carries on.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;11. Hand-off&lt;/td&gt; 
    &lt;td class="prop"&gt;While running&lt;/td&gt; 
    &lt;td class="prop"&gt;The agent hands a conversation to a person, who takes it over.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;12. Halt and restart&lt;/td&gt; 
    &lt;td class="prop"&gt;While running&lt;/td&gt; 
    &lt;td class="prop"&gt;An authorized person stops a capability at once, and restarting it takes two people and a safe state.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;13. Attestation&lt;/td&gt; 
    &lt;td class="prop"&gt;On a schedule&lt;/td&gt; 
    &lt;td class="prop"&gt;A person confirms on a schedule that a system is still fit for use.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;14. Oversight quality&lt;/td&gt; 
    &lt;td class="prop"&gt;Across many decisions&lt;/td&gt; 
    &lt;td class="prop"&gt;The oversight itself is watched, and its accountable owner acts when people review too quickly.&lt;/td&gt; 
   &lt;/tr&gt; 
  &lt;/tbody&gt; 
 &lt;/table&gt; 
 &lt;h2&gt;Rules beyond the wait&lt;/h2&gt; 
 &lt;p&gt;Teams that build human decisions with the Akka SDK start from its durable wait. A workflow in the Akka SDK pauses a step with &lt;code&gt;thenPause&lt;/code&gt; and continues when a command arrives. An operator can also suspend a workflow from outside with &lt;code&gt;suspend&lt;/code&gt; and continue it with &lt;code&gt;resume&lt;/code&gt;. An autonomous agent can create a task that a person completes instead of an agent. The workflow state and the task both persist, so a wait survives a restart of the service.&lt;/p&gt; 
 &lt;p&gt;The rules of the decision itself are application code. The person who answers needs authority: a role, and for some decisions a qualification such as expertise in the relevant field of medicine. Some people are excluded from answering, such as the agent or person who made the request, or the author of a change. The answer has a deadline, and the control declares the outcome when the deadline passes with no answer. A minimum review time rejects an approval made faster than the control allows. The answer leaves a record of who decided, what they saw, how long they looked and why.&lt;/p&gt; 
&lt;/div&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/human-in-the-loop-ai-agents" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/blog-band-human-oversight.jpg" alt="14 Human Oversight Patterns for AI Systems" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="body col"&gt; 
 &lt;div class="enote"&gt; 
  &lt;strong&gt;TL;DR&lt;/strong&gt; 
  &lt;ul&gt; 
   &lt;li&gt;&lt;span&gt;Human-in-the-loop AI agents hand some decisions to a person, such as approving a payment before it is sent.&lt;/span&gt;&lt;/li&gt; 
   &lt;li&gt;&lt;span&gt;We studied AI regulations, our customers, our own systems, and published applications, and found 206 human controls across 225 regulations.&lt;/span&gt;&lt;/li&gt; 
   &lt;li&gt;&lt;span&gt;Almost all of them fit one of 14 patterns, each set by who decides, the deadline, and what happens if nobody decides.&lt;/span&gt;&lt;/li&gt; 
  &lt;/ul&gt; 
 &lt;/div&gt;  
 &lt;p class="toc-label"&gt;Sections&lt;/p&gt; 
 &lt;ul&gt; 
  &lt;li&gt;&lt;a href="#rules-beyond-the-wait"&gt;&lt;span class="num"&gt;I&lt;/span&gt;&lt;span class="ttl"&gt;Rules beyond the wait&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#sources-and-method"&gt;&lt;span class="num"&gt;II&lt;/span&gt;&lt;span class="ttl"&gt;Sources and method&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#the-oversight-event-and-its-settings"&gt;&lt;span class="num"&gt;III&lt;/span&gt;&lt;span class="ttl"&gt;The oversight event and its settings&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-1-action-approval"&gt;&lt;span class="num"&gt;IV&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 1 - Action approval&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-2-decision-review"&gt;&lt;span class="num"&gt;V&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 2 - Decision review&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-3-change-authorization"&gt;&lt;span class="num"&gt;VI&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 3 - Change authorization&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-4-exception"&gt;&lt;span class="num"&gt;VII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 4 - Exception&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-5-match-and-adjudication"&gt;&lt;span class="num"&gt;VIII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 5 - Match and adjudication&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-6-outcome-reconciliation"&gt;&lt;span class="num"&gt;IX&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 6 - Outcome reconciliation&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-7-contest-and-appeal"&gt;&lt;span class="num"&gt;X&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 7 - Contest and appeal&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-8-override"&gt;&lt;span class="num"&gt;XI&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 8 - Override&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-9-flag-and-feedback"&gt;&lt;span class="num"&gt;XII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 9 - Flag and feedback&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-10-triage"&gt;&lt;span class="num"&gt;XIII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 10 - Triage&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-11-hand-off"&gt;&lt;span class="num"&gt;XIV&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 11 - Hand-off&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-12-halt-and-restart"&gt;&lt;span class="num"&gt;XV&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 12 - Halt and restart&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-13-attestation"&gt;&lt;span class="num"&gt;XVI&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 13 - Attestation&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#pattern-14-oversight-quality"&gt;&lt;span class="num"&gt;XVII&lt;/span&gt;&lt;span class="ttl"&gt;Pattern 14 - Oversight quality&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#reviewer-screens"&gt;&lt;span class="num"&gt;XVIII&lt;/span&gt;&lt;span class="ttl"&gt;Reviewer screens&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#implementation"&gt;&lt;span class="num"&gt;XIX&lt;/span&gt;&lt;span class="ttl"&gt;Implementation&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#lessons-learned"&gt;&lt;span class="num"&gt;XX&lt;/span&gt;&lt;span class="ttl"&gt;Lessons learned&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#limits-of-the-evidence"&gt;&lt;span class="num"&gt;XXI&lt;/span&gt;&lt;span class="ttl"&gt;Limits of the evidence&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#conclusion"&gt;&lt;span class="num"&gt;XXII&lt;/span&gt;&lt;span class="ttl"&gt;Conclusion&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#appendix-a-human-controls-in-the-corpus-by-pattern"&gt;&lt;span class="num"&gt;XXIII&lt;/span&gt;&lt;span class="ttl"&gt;Appendix A: Human controls in the corpus by pattern&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#appendix-b-the-decision-vocabulary"&gt;&lt;span class="num"&gt;XXIV&lt;/span&gt;&lt;span class="ttl"&gt;Appendix B: The decision vocabulary&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
 &lt;/ul&gt;  
 &lt;p class="lede"&gt;Every AI system we have helped build reaches a point where a person has to decide. A payments agent matches an invoice to a purchase order and asks to pay the supplier 4,200 EUR. A notices agent drafts an adverse decision letter to a loan applicant. A model team proposes sending a new model 10% of production traffic. The payment, the letter and the rollout wait for a person to decide before the work continues.&lt;/p&gt; 
 &lt;p&gt;Our customers build each human decision with the Akka SDK. A workflow step pauses for the person and resumes when the decision arrives. A timeout on the pause can send a reminder or move the decision to someone more senior. An autonomous agent can hand a task to a person and wait for the result. The workflow and the task are durable, so a wait that lasts days survives a restart of the service.&lt;/p&gt; 
 &lt;p&gt;Until now, each of these human processes was designed individually, as its own Akka workflow. The team building the application decided who may answer, how long the answer may take, what happens when nobody answers, and what record the answer leaves. Here are some examples in real-world AI systems we have helped build where a human-in-the-loop control existed:&lt;/p&gt; 
 &lt;ol&gt; 
  &lt;li&gt;A beverage manufacturer's data-use governance system sends every recommendation to a dataset owner or the governance lead, who confirms, overrides or escalates it with a mandatory reason. The reminder after three days and the escalation after seven are application code around a workflow pause.&lt;/li&gt; 
  &lt;li&gt;A medical-supply company's voice agent routes a call to a person at 21 points in its four call flows.&lt;/li&gt; 
 &lt;/ol&gt; 
 &lt;p&gt;We did the same thing in our own services. A loan origination service, a trace ledger and a context layer, all built at Akka, each implemented the same oversight contract by hand. The loan origination service and the trace ledger marked a decision at risk when 75% of its time had passed, and the context layer marked it at 80%.&lt;/p&gt; 
 &lt;p&gt;When we looked in our own regulatory corpus in &lt;a href="https://akka.io/platform/verify"&gt;Akka Verify&lt;/a&gt;, we found 206 controls that require a person to act. A control states one obligation from a regulation or framework and names the mechanism that enforces it. Our own schema had no way to describe what that person decides.&lt;/p&gt; 
 &lt;p&gt;We also read the documentation of 34 products in AI governance, LLM operations and cloud agent services. Only three AI governance platforms among them turn a monitoring signal into a work item that someone owns and closes with a recorded outcome.&lt;/p&gt; 
 &lt;p&gt;Across our customers' systems, our own services and the regulations, the same human processes kept recurring. We found fourteen recurring patterns, and each one is described below with the obligation behind it, its settings and a working example. Each pattern raises an oversight event, the record of one decision a person must make, and every oversight event moves through one set of states. The patterns differ in who decides, the deadline, the outcome when nobody decides, the allowed decisions and the record kept.&lt;/p&gt; 
 &lt;p&gt;The fourteen patterns will soon be available as templates in the Akka Agentic AI Platform.&lt;/p&gt; 
 &lt;table class="rtbl"&gt; 
  &lt;colgroup&gt; 
   &lt;col style="width: 24%;"&gt; 
   &lt;col style="width: 20%;"&gt; 
   &lt;col style="width: 56%;"&gt; 
  &lt;/colgroup&gt; 
  &lt;thead&gt; 
   &lt;tr&gt; 
    &lt;th&gt;Pattern&lt;/th&gt; 
    &lt;th&gt;When the person acts&lt;/th&gt; 
    &lt;th&gt;What happens&lt;/th&gt; 
   &lt;/tr&gt; 
  &lt;/thead&gt; 
  &lt;tbody&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;1. Action approval&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person approves an action before the system takes it.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;2. Decision review&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person reviews a decision about someone before it takes effect.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;3. Change authorization&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;Two people who neither proposed nor built a change approve it before it reaches production.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;4. Exception&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person grants a time-limited exception to a rule, with a reason.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;5. Match and adjudication&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person decides whether two records describe the same thing.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;6. Outcome reconciliation&lt;/td&gt; 
    &lt;td class="prop"&gt;Before the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person establishes what happened when the system cannot tell.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;7. Contest and appeal&lt;/td&gt; 
    &lt;td class="prop"&gt;After the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person affected by a decision appeals it to someone independent of it.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;8. Override&lt;/td&gt; 
    &lt;td class="prop"&gt;After the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;A person sets aside or replaces an output the system produced, and says why.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;9. Flag and feedback&lt;/td&gt; 
    &lt;td class="prop"&gt;After the system acts&lt;/td&gt; 
    &lt;td class="prop"&gt;Anyone can flag an output as wrong, and a triager decides whether it becomes an evaluation case.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;10. Triage&lt;/td&gt; 
    &lt;td class="prop"&gt;While running&lt;/td&gt; 
    &lt;td class="prop"&gt;A monitor raises a signal, and a person decides whether it is a problem while the work carries on.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;11. Hand-off&lt;/td&gt; 
    &lt;td class="prop"&gt;While running&lt;/td&gt; 
    &lt;td class="prop"&gt;The agent hands a conversation to a person, who takes it over.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;12. Halt and restart&lt;/td&gt; 
    &lt;td class="prop"&gt;While running&lt;/td&gt; 
    &lt;td class="prop"&gt;An authorized person stops a capability at once, and restarting it takes two people and a safe state.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;13. Attestation&lt;/td&gt; 
    &lt;td class="prop"&gt;On a schedule&lt;/td&gt; 
    &lt;td class="prop"&gt;A person confirms on a schedule that a system is still fit for use.&lt;/td&gt; 
   &lt;/tr&gt; 
   &lt;tr&gt; 
    &lt;td class="prop"&gt;14. Oversight quality&lt;/td&gt; 
    &lt;td class="prop"&gt;Across many decisions&lt;/td&gt; 
    &lt;td class="prop"&gt;The oversight itself is watched, and its accountable owner acts when people review too quickly.&lt;/td&gt; 
   &lt;/tr&gt; 
  &lt;/tbody&gt; 
 &lt;/table&gt; 
 &lt;h2&gt;Rules beyond the wait&lt;/h2&gt; 
 &lt;p&gt;Teams that build human decisions with the Akka SDK start from its durable wait. A workflow in the Akka SDK pauses a step with &lt;code&gt;thenPause&lt;/code&gt; and continues when a command arrives. An operator can also suspend a workflow from outside with &lt;code&gt;suspend&lt;/code&gt; and continue it with &lt;code&gt;resume&lt;/code&gt;. An autonomous agent can create a task that a person completes instead of an agent. The workflow state and the task both persist, so a wait survives a restart of the service.&lt;/p&gt; 
 &lt;p&gt;The rules of the decision itself are application code. The person who answers needs authority: a role, and for some decisions a qualification such as expertise in the relevant field of medicine. Some people are excluded from answering, such as the agent or person who made the request, or the author of a change. The answer has a deadline, and the control declares the outcome when the deadline passes with no answer. A minimum review time rejects an approval made faster than the control allows. The answer leaves a record of who decided, what they saw, how long they looked and why.&lt;/p&gt; 
&lt;/div&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fhuman-in-the-loop-ai-agents&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Architect / CTO</category>
      <category>Governance</category>
      <category>CAIO / CIO</category>
      <pubDate>Wed, 04 Nov 2026 19:00:34 GMT</pubDate>
      <guid>https://akka.io/blog/human-in-the-loop-ai-agents</guid>
      <dc:date>2026-11-04T19:00:34Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>Fast, Cheap Agent Decisions</title>
      <link>https://akka.io/blog/fast-cheap-agent-decisions</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/fast-cheap-agent-decisions" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/blog-band-jev.jpg" alt="Fast, Cheap Agent Decisions" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="body col"&gt; 
 &lt;p class="lede"&gt;Jev has only been available to the public for about a week (unveiled September 15th, 2026), but it’s already created a ton of excitement and discussion within the AI community. &lt;a href="https://typesafe.ai"&gt;TypeSafe&lt;/a&gt; made Jev available through limited, early access at the same time they disclosed their $40 million seed round.&lt;/p&gt; 
 &lt;p&gt;Let’s strip away the hype and talk about what Jev is, and what it isn’t. When our agentic code submits a prompt to an LLM, it can include a description of available tools, context, and conversation history. The typical pattern is to stream text into a model and we get the answer streamed back out.&lt;/p&gt; 
 &lt;p&gt;If we want more structured replies, we can give these LLMs a schema and tell them that their output must conform to that schema. This structured output can be anything from data queried from a customer’s account to a set of product recommendations or an itinerary for a vacation based on weather forecasts and traveler preferences.&lt;/p&gt; 
 &lt;p&gt;Sometimes this output schema is extremely focused, like asking specific questions. Is there PII exposed anywhere in this text? Is the sentiment in the text negative or positive? Does the supplied context refer to a tech support issue or is it an account query? Narrowing down the potential output of an LLM to these extremely focused questions dramatically increases accuracy and reduces hallucinations.&lt;/p&gt; 
 &lt;p&gt;However, using a general-purpose LLM to answer these questions is extremely inefficient. While you could technically use a jumbo jet at runway taxi speed to commute from home to the office, it’s slow, inefficient, and costly. Overkill has a cost and many of us are using all-purpose LLMs when we could be using something more focused and efficient.&lt;/p&gt; 
 &lt;p&gt;This is where Jev comes in. It’s a special kind of model with its own interface style. It doesn’t stream text bi-directionally like LLMs. Instead, you ask it questions based on some state and it gives you answers and their probability scores.&lt;/p&gt; 
 &lt;h2&gt;Performance and cost figures TypeSafe published for Jev&lt;/h2&gt; 
 &lt;p&gt;TypeSafe published these figures with the Jev launch on September 15, 2026. All of them come from TypeSafe’s own evaluations and demos.&lt;/p&gt; 
 &lt;div class="table-scroll"&gt; 
  &lt;table class="claims"&gt; 
   &lt;thead&gt; 
    &lt;tr&gt; 
     &lt;th&gt;Claim&lt;/th&gt; 
     &lt;th&gt;Jev&lt;/th&gt; 
     &lt;th&gt;Comparison, as TypeSafe states it&lt;/th&gt; 
    &lt;/tr&gt; 
   &lt;/thead&gt; 
   &lt;tbody&gt; 
    &lt;tr&gt; 
     &lt;td&gt;End-to-end response time&lt;/td&gt; 
     &lt;td&gt;70 ms to 500 ms&lt;/td&gt; 
     &lt;td&gt;3 s to 329 s for frontier models&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Speed on System One queries&lt;/td&gt; 
     &lt;td&gt;40x to 200x faster&lt;/td&gt; 
     &lt;td&gt;The same level of frontier intelligence&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Workflow evaluation, time per task&lt;/td&gt; 
     &lt;td&gt;0.114 s&lt;/td&gt; 
     &lt;td&gt;8.566 s for LLMs, 193.6x slower&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Workflow evaluation, cost per task&lt;/td&gt; 
     &lt;td&gt;$0.000081&lt;/td&gt; 
     &lt;td&gt;$0.013880 for LLMs, 444.6x more expensive&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Accuracy, average of four workflows&lt;/td&gt; 
     &lt;td&gt;67.8% at $0.0004 per workflow&lt;/td&gt; 
     &lt;td&gt;GPT-5.6 Terra 67.9% at $0.0304; Claude Sonnet 5 67.8% at $0.1174&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Input token price&lt;/td&gt; 
     &lt;td&gt;$0.042 per million ($42 per billion)&lt;/td&gt; 
     &lt;td&gt;$0.20 to $10 per million&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Input price against Claude Fable 5.1&lt;/td&gt; 
     &lt;td&gt;238x lower&lt;/td&gt; 
     &lt;td&gt;$10 per million&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Output token price&lt;/td&gt; 
     &lt;td&gt;Free&lt;/td&gt; 
     &lt;td&gt;About 5x the input price&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Hallucinations and type errors&lt;/td&gt; 
     &lt;td&gt;Zero, guaranteed by the output schema&lt;/td&gt; 
     &lt;td&gt;LLMs hallucinate and make type errors&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Confidence&lt;/td&gt; 
     &lt;td&gt;A calibrated probability with every answer&lt;/td&gt; 
     &lt;td&gt;Overconfident and inconsistent, even when prompted for confidence&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Doom demo&lt;/td&gt; 
     &lt;td&gt;10 queries per second for about $7 per hour&lt;/td&gt; 
     &lt;td&gt;None given&lt;/td&gt; 
    &lt;/tr&gt; 
   &lt;/tbody&gt; 
  &lt;/table&gt; 
 &lt;/div&gt; 
 &lt;p&gt;TypeSafe states that the 193.6x and 444.6x figures are on the higher end of real-world gains. The zero-hallucination figure follows from the output schema and was not measured. The LLM error rates TypeSafe compares against come from OpenRouter traffic. Sources: &lt;a href="https://typesafe.ai/blog/introducing-system-one-models-and-jev"&gt;TypeSafe launch post&lt;/a&gt;, &lt;a href="https://typesafe.ai"&gt;typesafe.ai&lt;/a&gt;, &lt;a href="https://evals.typesafe.ai/"&gt;evals.typesafe.ai&lt;/a&gt;.&lt;/p&gt; 
&lt;/div&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/fast-cheap-agent-decisions" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/blog-band-jev.jpg" alt="Fast, Cheap Agent Decisions" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="body col"&gt; 
 &lt;p class="lede"&gt;Jev has only been available to the public for about a week (unveiled September 15th, 2026), but it’s already created a ton of excitement and discussion within the AI community. &lt;a href="https://typesafe.ai"&gt;TypeSafe&lt;/a&gt; made Jev available through limited, early access at the same time they disclosed their $40 million seed round.&lt;/p&gt; 
 &lt;p&gt;Let’s strip away the hype and talk about what Jev is, and what it isn’t. When our agentic code submits a prompt to an LLM, it can include a description of available tools, context, and conversation history. The typical pattern is to stream text into a model and we get the answer streamed back out.&lt;/p&gt; 
 &lt;p&gt;If we want more structured replies, we can give these LLMs a schema and tell them that their output must conform to that schema. This structured output can be anything from data queried from a customer’s account to a set of product recommendations or an itinerary for a vacation based on weather forecasts and traveler preferences.&lt;/p&gt; 
 &lt;p&gt;Sometimes this output schema is extremely focused, like asking specific questions. Is there PII exposed anywhere in this text? Is the sentiment in the text negative or positive? Does the supplied context refer to a tech support issue or is it an account query? Narrowing down the potential output of an LLM to these extremely focused questions dramatically increases accuracy and reduces hallucinations.&lt;/p&gt; 
 &lt;p&gt;However, using a general-purpose LLM to answer these questions is extremely inefficient. While you could technically use a jumbo jet at runway taxi speed to commute from home to the office, it’s slow, inefficient, and costly. Overkill has a cost and many of us are using all-purpose LLMs when we could be using something more focused and efficient.&lt;/p&gt; 
 &lt;p&gt;This is where Jev comes in. It’s a special kind of model with its own interface style. It doesn’t stream text bi-directionally like LLMs. Instead, you ask it questions based on some state and it gives you answers and their probability scores.&lt;/p&gt; 
 &lt;h2&gt;Performance and cost figures TypeSafe published for Jev&lt;/h2&gt; 
 &lt;p&gt;TypeSafe published these figures with the Jev launch on September 15, 2026. All of them come from TypeSafe’s own evaluations and demos.&lt;/p&gt; 
 &lt;div class="table-scroll"&gt; 
  &lt;table class="claims"&gt; 
   &lt;thead&gt; 
    &lt;tr&gt; 
     &lt;th&gt;Claim&lt;/th&gt; 
     &lt;th&gt;Jev&lt;/th&gt; 
     &lt;th&gt;Comparison, as TypeSafe states it&lt;/th&gt; 
    &lt;/tr&gt; 
   &lt;/thead&gt; 
   &lt;tbody&gt; 
    &lt;tr&gt; 
     &lt;td&gt;End-to-end response time&lt;/td&gt; 
     &lt;td&gt;70 ms to 500 ms&lt;/td&gt; 
     &lt;td&gt;3 s to 329 s for frontier models&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Speed on System One queries&lt;/td&gt; 
     &lt;td&gt;40x to 200x faster&lt;/td&gt; 
     &lt;td&gt;The same level of frontier intelligence&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Workflow evaluation, time per task&lt;/td&gt; 
     &lt;td&gt;0.114 s&lt;/td&gt; 
     &lt;td&gt;8.566 s for LLMs, 193.6x slower&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Workflow evaluation, cost per task&lt;/td&gt; 
     &lt;td&gt;$0.000081&lt;/td&gt; 
     &lt;td&gt;$0.013880 for LLMs, 444.6x more expensive&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Accuracy, average of four workflows&lt;/td&gt; 
     &lt;td&gt;67.8% at $0.0004 per workflow&lt;/td&gt; 
     &lt;td&gt;GPT-5.6 Terra 67.9% at $0.0304; Claude Sonnet 5 67.8% at $0.1174&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Input token price&lt;/td&gt; 
     &lt;td&gt;$0.042 per million ($42 per billion)&lt;/td&gt; 
     &lt;td&gt;$0.20 to $10 per million&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Input price against Claude Fable 5.1&lt;/td&gt; 
     &lt;td&gt;238x lower&lt;/td&gt; 
     &lt;td&gt;$10 per million&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Output token price&lt;/td&gt; 
     &lt;td&gt;Free&lt;/td&gt; 
     &lt;td&gt;About 5x the input price&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Hallucinations and type errors&lt;/td&gt; 
     &lt;td&gt;Zero, guaranteed by the output schema&lt;/td&gt; 
     &lt;td&gt;LLMs hallucinate and make type errors&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Confidence&lt;/td&gt; 
     &lt;td&gt;A calibrated probability with every answer&lt;/td&gt; 
     &lt;td&gt;Overconfident and inconsistent, even when prompted for confidence&lt;/td&gt; 
    &lt;/tr&gt; 
    &lt;tr&gt; 
     &lt;td&gt;Doom demo&lt;/td&gt; 
     &lt;td&gt;10 queries per second for about $7 per hour&lt;/td&gt; 
     &lt;td&gt;None given&lt;/td&gt; 
    &lt;/tr&gt; 
   &lt;/tbody&gt; 
  &lt;/table&gt; 
 &lt;/div&gt; 
 &lt;p&gt;TypeSafe states that the 193.6x and 444.6x figures are on the higher end of real-world gains. The zero-hallucination figure follows from the output schema and was not measured. The LLM error rates TypeSafe compares against come from OpenRouter traffic. Sources: &lt;a href="https://typesafe.ai/blog/introducing-system-one-models-and-jev"&gt;TypeSafe launch post&lt;/a&gt;, &lt;a href="https://typesafe.ai"&gt;typesafe.ai&lt;/a&gt;, &lt;a href="https://evals.typesafe.ai/"&gt;evals.typesafe.ai&lt;/a&gt;.&lt;/p&gt; 
&lt;/div&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Ffast-cheap-agent-decisions&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Architect / CTO</category>
      <category>Agents</category>
      <pubDate>Wed, 23 Sep 2026 14:12:35 GMT</pubDate>
      <author>kevin.hoffman@akka.io (Kevin Hoffman)</author>
      <guid>https://akka.io/blog/fast-cheap-agent-decisions</guid>
      <dc:date>2026-09-23T14:12:35Z</dc:date>
    </item>
    <item>
      <title>15 Techniques to lower the cost of AI</title>
      <link>https://akka.io/blog/lowering-the-cost-of-ai</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/lowering-the-cost-of-ai" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/blog-band-lowering-cost-ai.jpg" alt="15 Techniques to lower the cost of AI" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt;  
&lt;div class="body col"&gt;  
 &lt;div class="kicker"&gt;
   AI · CAIO / CIO · ARCHITECT / CTO 
 &lt;/div&gt; 
 &lt;h1 class="title"&gt;15 Techniques to lower the cost of AI&lt;/h1&gt; 
 &lt;p class="standfirst"&gt;Incorporate smart harnesses, multi-model routing, self-hosting open-weight models, prompt caching, and SLM training to lower your AI spend.&lt;/p&gt; 
 &lt;div class="byline"&gt; 
  &lt;div&gt; 
   &lt;span class="au"&gt;Tyler Jewell&lt;/span&gt; 
  &lt;/div&gt; 
  &lt;div class="r"&gt; September 8, 2026 · 18 min read 
  &lt;/div&gt; 
 &lt;/div&gt;  
 &lt;p class="lede ai-lede"&gt;&lt;span class="dropcap"&gt;AI&lt;/span&gt; budgets approved in the last two years have grown exponentially. A16Z indicates that its grown 3x year-over-year on a per-employee basis.&lt;/p&gt;  
 &lt;p class="viz-title"&gt;Figure 1&lt;/p&gt; 
 &lt;h5&gt;Everyone is spending more on AI, but some a lot more than others&lt;/h5&gt;  AI spend per employee per month, split by percentile. Andreessen Horowitz, sourced from the Ramp AI Index, 12 August 2026. AI spend includes LLM subscriptions, coding agents, API access and GPU cloud spend.   
 &lt;p&gt;While many frontier model providers are subsidizing token spend for enterprises, this will not happen indefinitely and many enterprises are developing strategies for reducing their AI spend:&lt;/p&gt; 
 &lt;ol&gt; 
  &lt;li&gt;Self-hosting open-weight models.&lt;/li&gt; 
  &lt;li&gt;Multi-model routing with prompt compaction and low model leverage.&lt;/li&gt; 
  &lt;li&gt;Training SLMs against proprietary to create small LORA adapters.&lt;/li&gt; 
 &lt;/ol&gt; 
 &lt;div class="enote"&gt; 
  &lt;b&gt;TLDR;&lt;/b&gt; 
  &lt;ul&gt; 
   &lt;li&gt;Start with creating a system for observing, controlling, and routing all AI traffic across all providers.&lt;/li&gt; 
   &lt;li&gt;Assess your cost per user, per session, per task, and per AI provider.&lt;/li&gt; 
   &lt;li&gt;Execute over a dozen strategies that lessen your dependence upon expensive frontier models.&lt;/li&gt; 
  &lt;/ul&gt; 
 &lt;/div&gt;  
 &lt;p class="toc-label"&gt;Sections&lt;/p&gt; 
 &lt;ul&gt; 
  &lt;li&gt;&lt;a href="#unit-prices-are-falling-and-bills-are-rising"&gt;&lt;span class="num"&gt;I&lt;/span&gt;&lt;span class="ttl"&gt;Unit prices are falling and bills are rising&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#the-techniques-that-lower-ai-spend"&gt;&lt;span class="num"&gt;II&lt;/span&gt;&lt;span class="ttl"&gt;The techniques that lower AI spend&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#1-procurement-buying-more-tokens-for-less-money"&gt;&lt;span class="num"&gt;III&lt;/span&gt;&lt;span class="ttl"&gt;Technique 1 - Procurement: Buying more tokens for less money&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#2-semantic-response-caching"&gt;&lt;span class="num"&gt;IV&lt;/span&gt;&lt;span class="ttl"&gt;Technique 2 - Semantic response caching&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#3-multi-model-routing-cascade-and-escalation"&gt;&lt;span class="num"&gt;V&lt;/span&gt;&lt;span class="ttl"&gt;Technique 3 - Multi-model routing: cascade and escalation&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#4-planner-and-worker-role-binding"&gt;&lt;span class="num"&gt;VI&lt;/span&gt;&lt;span class="ttl"&gt;Technique 4 - Planner and worker role binding&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#5-sticky-sessions-and-the-prompt-cache-discount"&gt;&lt;span class="num"&gt;VII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 5 - Sticky sessions and the prompt cache discount&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#6-prompt-compression"&gt;&lt;span class="num"&gt;VIII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 6 - Prompt compression&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#7-memory-compaction"&gt;&lt;span class="num"&gt;IX&lt;/span&gt;&lt;span class="ttl"&gt;Technique 7 - Memory compaction&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#8-smart-harnesses-code-execution-instead-of-tool-calling-loops"&gt;&lt;span class="num"&gt;X&lt;/span&gt;&lt;span class="ttl"&gt;Technique 8 - Smart harnesses: code execution instead of tool-calling loops&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#9-guardrails-that-stop-requests"&gt;&lt;span class="num"&gt;XI&lt;/span&gt;&lt;span class="ttl"&gt;Technique 9 - Guardrails that stop requests&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#10-training-specialist-models-into-slms"&gt;&lt;span class="num"&gt;XII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 10 - Training specialist models into SLMs&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#11-self-hosting-open-weight-models"&gt;&lt;span class="num"&gt;XIII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 11 - Self-hosting open-weight models&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#12-durable-execution-not-paying-twice-for-a-failed-run"&gt;&lt;span class="num"&gt;XIV&lt;/span&gt;&lt;span class="ttl"&gt;Technique 12 - Durable execution: not paying twice for a failed run&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#13-collapsing-separately-metered-services"&gt;&lt;span class="num"&gt;XV&lt;/span&gt;&lt;span class="ttl"&gt;Technique 13 - Collapsing separately metered services&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#14-scale-to-zero-on-idle-capacity"&gt;&lt;span class="num"&gt;XVI&lt;/span&gt;&lt;span class="ttl"&gt;Technique 14 - Scale-to-zero on idle capacity&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#15-measuring-cost-per-verified-task"&gt;&lt;span class="num"&gt;XVII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 15 - Measuring cost per verified task&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#the-cost-of-agentic-rework"&gt;&lt;span class="num"&gt;XVIII&lt;/span&gt;&lt;span class="ttl"&gt;The cost of agentic rework&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#where-to-start"&gt;&lt;span class="num"&gt;XIX&lt;/span&gt;&lt;span class="ttl"&gt;Where to start&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#how-akka-can-help-lower-your-ai-bill"&gt;&lt;span class="num"&gt;XX&lt;/span&gt;&lt;span class="ttl"&gt;How Akka can help lower your AI bill&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#akka-ai-spend-calculators"&gt;&lt;span class="num"&gt;XXI&lt;/span&gt;&lt;span class="ttl"&gt;Akka AI spend calculators&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
 &lt;/ul&gt;  
&lt;/div&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/lowering-the-cost-of-ai" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/blog-band-lowering-cost-ai.jpg" alt="15 Techniques to lower the cost of AI" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt;  
&lt;div class="body col"&gt;  
 &lt;div class="kicker"&gt;
   AI · CAIO / CIO · ARCHITECT / CTO 
 &lt;/div&gt; 
 &lt;h1 class="title"&gt;15 Techniques to lower the cost of AI&lt;/h1&gt; 
 &lt;p class="standfirst"&gt;Incorporate smart harnesses, multi-model routing, self-hosting open-weight models, prompt caching, and SLM training to lower your AI spend.&lt;/p&gt; 
 &lt;div class="byline"&gt; 
  &lt;div&gt; 
   &lt;span class="au"&gt;Tyler Jewell&lt;/span&gt; 
  &lt;/div&gt; 
  &lt;div class="r"&gt; September 8, 2026 · 18 min read 
  &lt;/div&gt; 
 &lt;/div&gt;  
 &lt;p class="lede ai-lede"&gt;&lt;span class="dropcap"&gt;AI&lt;/span&gt; budgets approved in the last two years have grown exponentially. A16Z indicates that its grown 3x year-over-year on a per-employee basis.&lt;/p&gt;  
 &lt;p class="viz-title"&gt;Figure 1&lt;/p&gt; 
 &lt;h5&gt;Everyone is spending more on AI, but some a lot more than others&lt;/h5&gt;  AI spend per employee per month, split by percentile. Andreessen Horowitz, sourced from the Ramp AI Index, 12 August 2026. AI spend includes LLM subscriptions, coding agents, API access and GPU cloud spend.   
 &lt;p&gt;While many frontier model providers are subsidizing token spend for enterprises, this will not happen indefinitely and many enterprises are developing strategies for reducing their AI spend:&lt;/p&gt; 
 &lt;ol&gt; 
  &lt;li&gt;Self-hosting open-weight models.&lt;/li&gt; 
  &lt;li&gt;Multi-model routing with prompt compaction and low model leverage.&lt;/li&gt; 
  &lt;li&gt;Training SLMs against proprietary to create small LORA adapters.&lt;/li&gt; 
 &lt;/ol&gt; 
 &lt;div class="enote"&gt; 
  &lt;b&gt;TLDR;&lt;/b&gt; 
  &lt;ul&gt; 
   &lt;li&gt;Start with creating a system for observing, controlling, and routing all AI traffic across all providers.&lt;/li&gt; 
   &lt;li&gt;Assess your cost per user, per session, per task, and per AI provider.&lt;/li&gt; 
   &lt;li&gt;Execute over a dozen strategies that lessen your dependence upon expensive frontier models.&lt;/li&gt; 
  &lt;/ul&gt; 
 &lt;/div&gt;  
 &lt;p class="toc-label"&gt;Sections&lt;/p&gt; 
 &lt;ul&gt; 
  &lt;li&gt;&lt;a href="#unit-prices-are-falling-and-bills-are-rising"&gt;&lt;span class="num"&gt;I&lt;/span&gt;&lt;span class="ttl"&gt;Unit prices are falling and bills are rising&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#the-techniques-that-lower-ai-spend"&gt;&lt;span class="num"&gt;II&lt;/span&gt;&lt;span class="ttl"&gt;The techniques that lower AI spend&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#1-procurement-buying-more-tokens-for-less-money"&gt;&lt;span class="num"&gt;III&lt;/span&gt;&lt;span class="ttl"&gt;Technique 1 - Procurement: Buying more tokens for less money&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#2-semantic-response-caching"&gt;&lt;span class="num"&gt;IV&lt;/span&gt;&lt;span class="ttl"&gt;Technique 2 - Semantic response caching&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#3-multi-model-routing-cascade-and-escalation"&gt;&lt;span class="num"&gt;V&lt;/span&gt;&lt;span class="ttl"&gt;Technique 3 - Multi-model routing: cascade and escalation&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#4-planner-and-worker-role-binding"&gt;&lt;span class="num"&gt;VI&lt;/span&gt;&lt;span class="ttl"&gt;Technique 4 - Planner and worker role binding&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#5-sticky-sessions-and-the-prompt-cache-discount"&gt;&lt;span class="num"&gt;VII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 5 - Sticky sessions and the prompt cache discount&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#6-prompt-compression"&gt;&lt;span class="num"&gt;VIII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 6 - Prompt compression&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#7-memory-compaction"&gt;&lt;span class="num"&gt;IX&lt;/span&gt;&lt;span class="ttl"&gt;Technique 7 - Memory compaction&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#8-smart-harnesses-code-execution-instead-of-tool-calling-loops"&gt;&lt;span class="num"&gt;X&lt;/span&gt;&lt;span class="ttl"&gt;Technique 8 - Smart harnesses: code execution instead of tool-calling loops&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#9-guardrails-that-stop-requests"&gt;&lt;span class="num"&gt;XI&lt;/span&gt;&lt;span class="ttl"&gt;Technique 9 - Guardrails that stop requests&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#10-training-specialist-models-into-slms"&gt;&lt;span class="num"&gt;XII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 10 - Training specialist models into SLMs&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#11-self-hosting-open-weight-models"&gt;&lt;span class="num"&gt;XIII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 11 - Self-hosting open-weight models&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#12-durable-execution-not-paying-twice-for-a-failed-run"&gt;&lt;span class="num"&gt;XIV&lt;/span&gt;&lt;span class="ttl"&gt;Technique 12 - Durable execution: not paying twice for a failed run&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#13-collapsing-separately-metered-services"&gt;&lt;span class="num"&gt;XV&lt;/span&gt;&lt;span class="ttl"&gt;Technique 13 - Collapsing separately metered services&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#14-scale-to-zero-on-idle-capacity"&gt;&lt;span class="num"&gt;XVI&lt;/span&gt;&lt;span class="ttl"&gt;Technique 14 - Scale-to-zero on idle capacity&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#15-measuring-cost-per-verified-task"&gt;&lt;span class="num"&gt;XVII&lt;/span&gt;&lt;span class="ttl"&gt;Technique 15 - Measuring cost per verified task&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#the-cost-of-agentic-rework"&gt;&lt;span class="num"&gt;XVIII&lt;/span&gt;&lt;span class="ttl"&gt;The cost of agentic rework&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#where-to-start"&gt;&lt;span class="num"&gt;XIX&lt;/span&gt;&lt;span class="ttl"&gt;Where to start&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#how-akka-can-help-lower-your-ai-bill"&gt;&lt;span class="num"&gt;XX&lt;/span&gt;&lt;span class="ttl"&gt;How Akka can help lower your AI bill&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
  &lt;li&gt;&lt;a href="#akka-ai-spend-calculators"&gt;&lt;span class="num"&gt;XXI&lt;/span&gt;&lt;span class="ttl"&gt;Akka AI spend calculators&lt;/span&gt;&lt;span class="dots"&gt;&lt;/span&gt;&lt;/a&gt;&lt;/li&gt; 
 &lt;/ul&gt;  
&lt;/div&gt;   
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Flowering-the-cost-of-ai&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <pubDate>Wed, 16 Sep 2026 07:46:39 GMT</pubDate>
      <guid>https://akka.io/blog/lowering-the-cost-of-ai</guid>
      <dc:date>2026-09-16T07:46:39Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>We Ported 65 OSS Projects With AI</title>
      <link>https://akka.io/blog/we-ported-65-oss-projects</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/we-ported-65-oss-projects" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/ported-65-linkedin.png" alt="We Ported 65 OSS Projects With AI" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/we-ported-65-oss-projects" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/website/blog/images/ported-65-linkedin.png" alt="We Ported 65 OSS Projects With AI" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fwe-ported-65-oss-projects&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Architect / CTO</category>
      <category>CAIO / CIO</category>
      <pubDate>Thu, 03 Sep 2026 11:08:59 GMT</pubDate>
      <guid>https://akka.io/blog/we-ported-65-oss-projects</guid>
      <dc:date>2026-09-03T11:08:59Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>Building the Akka MCP Gateway: Governed AI Tool Access for the Whole Organization</title>
      <link>https://akka.io/blog/building-the-akka-mcp-gateway</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/building-the-akka-mcp-gateway" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/akka-mcp-gateway-featured-branded.png" alt="Building the Akka MCP Gateway: Governed AI Tool Access for the Whole Organization" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;p&gt;Becoming an &lt;strong&gt;AI-first organization at Akka&lt;/strong&gt; means more than giving employees access to a chat interface; it means connecting AI to the tools and data that drive our business, securely and responsibly. The Model Context Protocol (MCP) has emerged as the standard for doing exactly that, linking AI clients to external services. But wiring up each service individually, per user, per machine, introduces security gaps that put both employee and customer data at risk, and quickly becomes unmanageable to govern at scale.&lt;/p&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/building-the-akka-mcp-gateway" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/akka-mcp-gateway-featured-branded.png" alt="Building the Akka MCP Gateway: Governed AI Tool Access for the Whole Organization" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;p&gt;Becoming an &lt;strong&gt;AI-first organization at Akka&lt;/strong&gt; means more than giving employees access to a chat interface; it means connecting AI to the tools and data that drive our business, securely and responsibly. The Model Context Protocol (MCP) has emerged as the standard for doing exactly that, linking AI clients to external services. But wiring up each service individually, per user, per machine, introduces security gaps that put both employee and customer data at risk, and quickly becomes unmanageable to govern at scale.&lt;/p&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fbuilding-the-akka-mcp-gateway&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Architect / CTO</category>
      <category>Security</category>
      <category>Agents</category>
      <category>Distributed Systems</category>
      <category>Governance</category>
      <category>CAIO / CIO</category>
      <pubDate>Wed, 05 Aug 2026 21:26:04 GMT</pubDate>
      <guid>https://akka.io/blog/building-the-akka-mcp-gateway</guid>
      <dc:date>2026-08-05T21:26:04Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>Scaling Agentic AI with Governance at the Core</title>
      <link>https://akka.io/blog/scaling-agentic-ai-session-gartner-2026</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/scaling-agentic-ai-session-gartner-2026" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/akka-demo/images/featured-gartner-session.jpg" alt="Video thumbnail of Tyler Jewell's Gartner Application Innovation &amp;amp; Business Solutions Summit session — click to watch on Gartner" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="hs-embed-wrapper" style="position: relative; overflow: hidden; width: 100%; height: auto; padding: 0; max-width: 540px; min-width: 256px; display: block; margin: auto;"&gt; 
 &lt;a href="https://cn.gartner.com/stream/APN40/4887703" style="display: block; position: relative; overflow: hidden; max-width: 100%; padding-bottom: 56.25%; margin: 0px;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;p style="text-align: center; font-size: 13px; color: #666; margin-top: 8px;"&gt;Watch on Gartner — Application Innovation &amp;amp; Business Solutions Summit, June 3, 2026&lt;/p&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/scaling-agentic-ai-session-gartner-2026" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/akka-demo/images/featured-gartner-session.jpg" alt="Video thumbnail of Tyler Jewell's Gartner Application Innovation &amp;amp; Business Solutions Summit session — click to watch on Gartner" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="hs-embed-wrapper" style="position: relative; overflow: hidden; width: 100%; height: auto; padding: 0; max-width: 540px; min-width: 256px; display: block; margin: auto;"&gt; 
 &lt;a href="https://cn.gartner.com/stream/APN40/4887703" style="display: block; position: relative; overflow: hidden; max-width: 100%; padding-bottom: 56.25%; margin: 0px;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;p style="text-align: center; font-size: 13px; color: #666; margin-top: 8px;"&gt;Watch on Gartner — Application Innovation &amp;amp; Business Solutions Summit, June 3, 2026&lt;/p&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fscaling-agentic-ai-session-gartner-2026&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Architect / CTO</category>
      <category>Resources</category>
      <category>Governance</category>
      <category>CAIO / CIO</category>
      <pubDate>Wed, 01 Jul 2026 20:12:00 GMT</pubDate>
      <guid>https://akka.io/blog/scaling-agentic-ai-session-gartner-2026</guid>
      <dc:date>2026-07-01T20:12:00Z</dc:date>
      <dc:creator>Tyler Jewell</dc:creator>
    </item>
    <item>
      <title>The Autonomous Operating Environment</title>
      <link>https://akka.io/blog/the-autonomous-operating-environment</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/the-autonomous-operating-environment" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/autonomous-operating-environment/convergence-diagram-v2.png" alt="The combination of agent framework and agent harness leads to the autonomous operating environment." class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt;  
&lt;p&gt;Software is starting to operate itself.&lt;/p&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/the-autonomous-operating-environment" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/autonomous-operating-environment/convergence-diagram-v2.png" alt="The combination of agent framework and agent harness leads to the autonomous operating environment." class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt;  
&lt;p&gt;Software is starting to operate itself.&lt;/p&gt;   
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fthe-autonomous-operating-environment&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Operations</category>
      <category>Developer</category>
      <category>Orchestration</category>
      <category>Agents</category>
      <category>Governance</category>
      <pubDate>Wed, 27 May 2026 18:20:29 GMT</pubDate>
      <guid>https://akka.io/blog/the-autonomous-operating-environment</guid>
      <dc:date>2026-05-27T18:20:29Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>Go Slow to Go Fast</title>
      <link>https://akka.io/blog/go-slow-to-go-fast</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/go-slow-to-go-fast" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/hero_featured.jpg" alt="Manulife and Akka speak about how an agentic platform helps to create reliable and cost-optimized AI" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="hs-embed-wrapper" style="position: relative; overflow: hidden; width: 100%; height: auto; padding: 0; max-width: 540px; min-width: 256px; display: block; margin: auto;"&gt; 
 &lt;div class="hs-embed-content-wrapper"&gt; 
  &lt;div style="position: relative; overflow: hidden; max-width: 100%; padding-bottom: 56.25%; margin: 0px;"&gt;  
  &lt;/div&gt; 
 &lt;/div&gt; 
&lt;/div&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/go-slow-to-go-fast" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/hero_featured.jpg" alt="Manulife and Akka speak about how an agentic platform helps to create reliable and cost-optimized AI" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="hs-embed-wrapper" style="position: relative; overflow: hidden; width: 100%; height: auto; padding: 0; max-width: 540px; min-width: 256px; display: block; margin: auto;"&gt; 
 &lt;div class="hs-embed-content-wrapper"&gt; 
  &lt;div style="position: relative; overflow: hidden; max-width: 100%; padding-bottom: 56.25%; margin: 0px;"&gt; 
   &lt;iframe width="256" height="144.64" src="https://www.youtube.com/embed/VgebRXDhE34?feature=oembed" frameborder="0" allowfullscreen style="position: absolute; top: 0px; left: 0px; width: 100%; height: 100%; border: none;"&gt;&lt;/iframe&gt; 
  &lt;/div&gt; 
 &lt;/div&gt; 
&lt;/div&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fgo-slow-to-go-fast&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Resources</category>
      <category>Governance</category>
      <category>CAIO / CIO</category>
      <pubDate>Fri, 15 May 2026 17:26:45 GMT</pubDate>
      <guid>https://akka.io/blog/go-slow-to-go-fast</guid>
      <dc:date>2026-05-15T17:26:45Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>Introducing Akka Specify</title>
      <link>https://akka.io/blog/introducing-akka-specify</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/introducing-akka-specify" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/Spec-Driven%20Development%20Graphic.png" alt="Introducing Akka Specify" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="hs-embed-wrapper" style="position: relative; overflow: hidden; width: 100%; height: auto; padding: 0; max-width: 540px; min-width: 256px; display: block; margin: auto;"&gt; 
 &lt;div class="hs-embed-content-wrapper"&gt; 
  &lt;div style="position: relative; overflow: hidden; max-width: 100%; padding-bottom: 56.25%; margin: 0px;"&gt;  
  &lt;/div&gt; 
 &lt;/div&gt; 
&lt;/div&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/introducing-akka-specify" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/Spec-Driven%20Development%20Graphic.png" alt="Introducing Akka Specify" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;div class="hs-embed-wrapper" style="position: relative; overflow: hidden; width: 100%; height: auto; padding: 0; max-width: 540px; min-width: 256px; display: block; margin: auto;"&gt; 
 &lt;div class="hs-embed-content-wrapper"&gt; 
  &lt;div style="position: relative; overflow: hidden; max-width: 100%; padding-bottom: 56.25%; margin: 0px;"&gt; 
   &lt;iframe width="256" height="144.64" src="https://www.youtube.com/embed/IIl8bQMB06I?feature=oembed" frameborder="0" allowfullscreen style="position: absolute; top: 0px; left: 0px; width: 100%; height: 100%; border: none;"&gt;&lt;/iframe&gt; 
  &lt;/div&gt; 
 &lt;/div&gt; 
&lt;/div&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fintroducing-akka-specify&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Operations</category>
      <category>Resources</category>
      <category>Developer</category>
      <category>DevEx</category>
      <category>News</category>
      <category>Demo</category>
      <category>Agents</category>
      <category>CAIO / CIO</category>
      <pubDate>Mon, 16 Mar 2026 15:27:53 GMT</pubDate>
      <guid>https://akka.io/blog/introducing-akka-specify</guid>
      <dc:date>2026-03-16T15:27:53Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
    <item>
      <title>Manulife Selects Akka to Operationalize Agentic AI within its Enterprise AI Platform</title>
      <link>https://akka.io/blog/manulife-selects-akka-to-operationalize-agentic-ai</link>
      <description>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/manulife-selects-akka-to-operationalize-agentic-ai" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/Screenshot%202026-03-09%20at%207.46.40%20AM.png" alt="Manulife Selects Akka to Operationalize Agentic AI within its Enterprise AI Platform" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;blockquote&gt; 
 &lt;h3&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;“Akka illustrates how enterprises can build and operate agentic systems with the speed, predictability, and governance required in highly regulated environments.”&lt;br&gt;&lt;/span&gt;&lt;/span&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;&lt;/span&gt; &lt;br&gt;&lt;/span&gt;&lt;span&gt;&lt;a href="https://www.linkedin.com/in/jodiewallis/"&gt;Jodie Wallis &lt;/a&gt;Global CAIO, Manulife&lt;/span&gt;&lt;/h3&gt; 
&lt;/blockquote&gt; 
&lt;p style="font-weight: bold;"&gt;&lt;br&gt;Executive Summary:&lt;/p&gt; 
&lt;ol&gt; 
 &lt;li&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;Manulife selected Akka as the platform on which to build and operate&amp;nbsp;agentic AI&amp;nbsp;&lt;/span&gt;&lt;/span&gt;&lt;/li&gt; 
 &lt;li&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;Akka provides secure and scalable foundation to support high volume of business applications&lt;/span&gt;&lt;/span&gt;&lt;/li&gt; 
 &lt;li&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;&lt;/span&gt;&lt;/span&gt;Manulife expects AI to generate $1 billion+ of enterprise value by 2027, with roughly one-fifth expected to come from improved efficiency.&lt;/li&gt; 
 &lt;li&gt;Following this announcement, Manulife's Global CAIO shares immediate &lt;a href="https://akka.io/blog/go-slow-to-go-fast"&gt;cost-optimization&lt;/a&gt; results by using Akka. &amp;nbsp;&lt;/li&gt; 
&lt;/ol&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;March 10, 2026, &lt;a href="https://www.manulife.com/ca/en/about-us/ai"&gt;Manulife&lt;/a&gt; announced that Akka has been selected as its latest partner supporting the development of its enterprise agentic AI platform. By bringing Akka’s durable and highly available runtime into the ecosystem, Manulife will strengthen the platform’s security, reliability, and performance as it embeds AI into critical workflows and expands adoption across the organization.&lt;/span&gt;&lt;/p&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;In June 2025, Manulife was named the &lt;a href="https://www.manulife.com/ca/en/about-us/news/manulife-named-number-1-life-insurance-company-for-ai-maturity-by-evident"&gt;#1 life insurance company for AI maturity&lt;/a&gt; in the inaugural Evident AI Index for Insurance, reinforcing the company’s commitment to innovation and customer-centric solutions.&amp;nbsp;&lt;/span&gt;&lt;br&gt;&lt;br&gt;&lt;span style="color: #ffce4a;"&gt;“Our insurance and investment businesses are built on trust, and that same principle guides our approach to AI,”&lt;/span&gt; said Jodie Wallis, Global Chief AI Officer for Manulife. “As we explore technologies that can help us scale reliable, compliant and resilient AI systems, solutions like Akka illustrate how enterprises can build and operate agentic systems with the speed, predictability, and governance required in highly regulated environments. Their focus on orchestration, safety, operational SLAs, and system reliability reflects exactly the kind of rigor that supports responsible AI and consistent customer value.”&lt;br&gt;&lt;br&gt;Manulife’s enterprise AI platform, now in beta testing, provides a secure, integrated foundation for building and deploying AI agents – intelligent systems that understand tasks, help make decisions, and take action to support customers and colleagues. The platform streamlines development, reduces operational costs, and enables high volume, business critical AI solutions, all while embedding strong governance, safeguards, and Responsible AI practices.&amp;nbsp;&lt;br&gt;&lt;/span&gt;&lt;/p&gt; 
&lt;blockquote&gt; 
 &lt;h3&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;“Manulife demonstrates the rare understanding of what it really takes to deliver agentic AI on a global scale.”&lt;br&gt;&lt;/span&gt; &lt;br&gt;&lt;/span&gt;&lt;span&gt;&lt;a href="https://www.linkedin.com/in/tylerjewell/"&gt;Tyler Jewell&lt;/a&gt; CEO, Akka&lt;/span&gt;&lt;/h3&gt; 
&lt;/blockquote&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;“Partnering with a leader like Manulife is an honor for us, and it’s a testament to our 15 years of enterprise scale and regulated industry experience,” said Tyler Jewell, CEO of Akka. “The Manulife team demonstrates the rare understanding of what it really takes to deliver agentic AI on a global scale. Without consistent engineering practices that address a complex and continually changing set of environment factors, AI systems, which are inherently random, will not be trusted to deliver business outcomes.”&lt;br&gt;&lt;br&gt;Manulife's partnership with Akka reinforces its commitment to Responsible AI and sustainability as outlined in its publicly available Responsible AI Principles. &lt;/span&gt;&lt;/p&gt; 
&lt;h3&gt;&lt;span style="font-weight: normal;"&gt;&lt;span style="color: #ffce4a;"&gt;By using&amp;nbsp;Akka, Manulife will be able to:&lt;/span&gt;&lt;/span&gt;&lt;/h3&gt; 
&lt;ol&gt; 
 &lt;li&gt; &lt;p&gt;&lt;span style="font-weight: normal;"&gt;Prioritize customer, colleague, and organizational safety through sound delivery and governance processes.&lt;/span&gt;&lt;/p&gt; &lt;/li&gt; 
 &lt;li&gt; &lt;p&gt;&lt;span style="background-color: transparent;"&gt;Design energy-efficient AI solutions that require less infrastructure to run and maintain&lt;/span&gt;&lt;/p&gt; &lt;/li&gt; 
 &lt;li&gt; &lt;p&gt;&lt;span style="background-color: transparent;"&gt;Build AI solutions that are explainable and reliable, while maintaining the appropriate human accountability for decision making.&lt;/span&gt;&lt;br&gt;&lt;/p&gt; &lt;/li&gt; 
&lt;/ol&gt; 
&lt;span style="background-color: transparent;"&gt;&lt;/span&gt; 
&lt;br&gt; 
&lt;blockquote&gt; 
 &lt;h3&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;“Akka provides a secure, scalable software foundation to support high volumes of business applications to accelerate meaningful value creation across our franchise.”&lt;br&gt;&lt;/span&gt; &lt;br&gt;&lt;/span&gt;&lt;span&gt;&lt;a href="https://www.linkedin.com/in/shamus-weiland-50648613/?originalSubdomain=ca"&gt;Shamus Weiland&lt;/a&gt; Global CIO, Manulife&lt;/span&gt;&lt;/h3&gt; 
&lt;/blockquote&gt; 
&lt;p&gt;&lt;span style="background-color: transparent;"&gt;&lt;br&gt;&lt;/span&gt;&lt;span style="background-color: transparent;"&gt;“Manulife is embedding AI across nearly every part of our business while equipping our colleagues with the capabilities to design, operate, and scale solutions,” said Shamus Weiland, Global Chief Information Officer at Manulife. "Akka provides a secure, scalable software foundation to support high volumes of business applications to accelerate meaningful value creation across our franchise. This partnership is rooted in our Responsible AI Principles, with a strong focus on governance, human oversight, and safety, as we advance toward becoming an AI-powered organization.”&amp;nbsp;&lt;/span&gt;&lt;/p&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;Manulife has been actively investing in and scaling AI capabilities since 2016. In the past three years, the firm has significantly increased the value enabled through advanced analytics and AI by expanding its AI solutions portfolio and increasing efficiency through data and AI platform investments, all driven by the company’s responsible AI framework. &lt;span&gt;In December 2025, Manulife announced that Adaptive ML would provide the reinforcement-learning-powered engine for finetuning and optimizing models within its enterprise AI platform. &lt;/span&gt;Manulife expects AI to generate $1 billion+ of enterprise value by 2027, with roughly one-fifth expected to come from improved efficiency.&lt;/span&gt;&lt;/p&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;&lt;a href="https://www.newswire.ca/news-releases/manulife-selects-akka-to-operationalize-agentic-ai-within-its-enterprise-ai-platform-804452481.html"&gt;GlobeNewswire release&lt;/a&gt;&lt;br&gt;&lt;/span&gt;&lt;/p&gt; 
&lt;h3&gt;&lt;span style="font-weight: normal; color: #ffce4a;"&gt;Related Information:&lt;/span&gt;&lt;/h3&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;Jodie Wallis speaks to &lt;a href="https://akka.io/blog/go-slow-to-go-fast"&gt;early results with Akka&lt;/a&gt;, setting a firm foundation of governance-first agentic AI, and the idea of "Go Slow to Go Fast".&lt;/span&gt;&lt;/p&gt;</description>
      <content:encoded>&lt;div class="hs-featured-image-wrapper"&gt; 
 &lt;a href="https://akka.io/blog/manulife-selects-akka-to-operationalize-agentic-ai" title="" class="hs-featured-image-link"&gt; &lt;img src="https://akka.io/hubfs/Screenshot%202026-03-09%20at%207.46.40%20AM.png" alt="Manulife Selects Akka to Operationalize Agentic AI within its Enterprise AI Platform" class="hs-featured-image" style="width:auto !important; max-width:50%; float:left; margin:0 15px 15px 0;"&gt; &lt;/a&gt; 
&lt;/div&gt; 
&lt;blockquote&gt; 
 &lt;h3&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;“Akka illustrates how enterprises can build and operate agentic systems with the speed, predictability, and governance required in highly regulated environments.”&lt;br&gt;&lt;/span&gt;&lt;/span&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;&lt;/span&gt; &lt;br&gt;&lt;/span&gt;&lt;span&gt;&lt;a href="https://www.linkedin.com/in/jodiewallis/"&gt;Jodie Wallis &lt;/a&gt;Global CAIO, Manulife&lt;/span&gt;&lt;/h3&gt; 
&lt;/blockquote&gt; 
&lt;p style="font-weight: bold;"&gt;&lt;br&gt;Executive Summary:&lt;/p&gt; 
&lt;ol&gt; 
 &lt;li&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;Manulife selected Akka as the platform on which to build and operate&amp;nbsp;agentic AI&amp;nbsp;&lt;/span&gt;&lt;/span&gt;&lt;/li&gt; 
 &lt;li&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;Akka provides secure and scalable foundation to support high volume of business applications&lt;/span&gt;&lt;/span&gt;&lt;/li&gt; 
 &lt;li&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;&lt;/span&gt;&lt;/span&gt;Manulife expects AI to generate $1 billion+ of enterprise value by 2027, with roughly one-fifth expected to come from improved efficiency.&lt;/li&gt; 
 &lt;li&gt;Following this announcement, Manulife's Global CAIO shares immediate &lt;a href="https://akka.io/blog/go-slow-to-go-fast"&gt;cost-optimization&lt;/a&gt; results by using Akka. &amp;nbsp;&lt;/li&gt; 
&lt;/ol&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;March 10, 2026, &lt;a href="https://www.manulife.com/ca/en/about-us/ai"&gt;Manulife&lt;/a&gt; announced that Akka has been selected as its latest partner supporting the development of its enterprise agentic AI platform. By bringing Akka’s durable and highly available runtime into the ecosystem, Manulife will strengthen the platform’s security, reliability, and performance as it embeds AI into critical workflows and expands adoption across the organization.&lt;/span&gt;&lt;/p&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;&lt;span&gt;In June 2025, Manulife was named the &lt;a href="https://www.manulife.com/ca/en/about-us/news/manulife-named-number-1-life-insurance-company-for-ai-maturity-by-evident"&gt;#1 life insurance company for AI maturity&lt;/a&gt; in the inaugural Evident AI Index for Insurance, reinforcing the company’s commitment to innovation and customer-centric solutions.&amp;nbsp;&lt;/span&gt;&lt;br&gt;&lt;br&gt;&lt;span style="color: #ffce4a;"&gt;“Our insurance and investment businesses are built on trust, and that same principle guides our approach to AI,”&lt;/span&gt; said Jodie Wallis, Global Chief AI Officer for Manulife. “As we explore technologies that can help us scale reliable, compliant and resilient AI systems, solutions like Akka illustrate how enterprises can build and operate agentic systems with the speed, predictability, and governance required in highly regulated environments. Their focus on orchestration, safety, operational SLAs, and system reliability reflects exactly the kind of rigor that supports responsible AI and consistent customer value.”&lt;br&gt;&lt;br&gt;Manulife’s enterprise AI platform, now in beta testing, provides a secure, integrated foundation for building and deploying AI agents – intelligent systems that understand tasks, help make decisions, and take action to support customers and colleagues. The platform streamlines development, reduces operational costs, and enables high volume, business critical AI solutions, all while embedding strong governance, safeguards, and Responsible AI practices.&amp;nbsp;&lt;br&gt;&lt;/span&gt;&lt;/p&gt; 
&lt;blockquote&gt; 
 &lt;h3&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;“Manulife demonstrates the rare understanding of what it really takes to deliver agentic AI on a global scale.”&lt;br&gt;&lt;/span&gt; &lt;br&gt;&lt;/span&gt;&lt;span&gt;&lt;a href="https://www.linkedin.com/in/tylerjewell/"&gt;Tyler Jewell&lt;/a&gt; CEO, Akka&lt;/span&gt;&lt;/h3&gt; 
&lt;/blockquote&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;“Partnering with a leader like Manulife is an honor for us, and it’s a testament to our 15 years of enterprise scale and regulated industry experience,” said Tyler Jewell, CEO of Akka. “The Manulife team demonstrates the rare understanding of what it really takes to deliver agentic AI on a global scale. Without consistent engineering practices that address a complex and continually changing set of environment factors, AI systems, which are inherently random, will not be trusted to deliver business outcomes.”&lt;br&gt;&lt;br&gt;Manulife's partnership with Akka reinforces its commitment to Responsible AI and sustainability as outlined in its publicly available Responsible AI Principles. &lt;/span&gt;&lt;/p&gt; 
&lt;h3&gt;&lt;span style="font-weight: normal;"&gt;&lt;span style="color: #ffce4a;"&gt;By using&amp;nbsp;Akka, Manulife will be able to:&lt;/span&gt;&lt;/span&gt;&lt;/h3&gt; 
&lt;ol&gt; 
 &lt;li&gt; &lt;p&gt;&lt;span style="font-weight: normal;"&gt;Prioritize customer, colleague, and organizational safety through sound delivery and governance processes.&lt;/span&gt;&lt;/p&gt; &lt;/li&gt; 
 &lt;li&gt; &lt;p&gt;&lt;span style="background-color: transparent;"&gt;Design energy-efficient AI solutions that require less infrastructure to run and maintain&lt;/span&gt;&lt;/p&gt; &lt;/li&gt; 
 &lt;li&gt; &lt;p&gt;&lt;span style="background-color: transparent;"&gt;Build AI solutions that are explainable and reliable, while maintaining the appropriate human accountability for decision making.&lt;/span&gt;&lt;br&gt;&lt;/p&gt; &lt;/li&gt; 
&lt;/ol&gt; 
&lt;span style="background-color: transparent;"&gt;&lt;/span&gt; 
&lt;br&gt; 
&lt;blockquote&gt; 
 &lt;h3&gt;&lt;span&gt;&lt;span style="color: #ffce4a;"&gt;“Akka provides a secure, scalable software foundation to support high volumes of business applications to accelerate meaningful value creation across our franchise.”&lt;br&gt;&lt;/span&gt; &lt;br&gt;&lt;/span&gt;&lt;span&gt;&lt;a href="https://www.linkedin.com/in/shamus-weiland-50648613/?originalSubdomain=ca"&gt;Shamus Weiland&lt;/a&gt; Global CIO, Manulife&lt;/span&gt;&lt;/h3&gt; 
&lt;/blockquote&gt; 
&lt;p&gt;&lt;span style="background-color: transparent;"&gt;&lt;br&gt;&lt;/span&gt;&lt;span style="background-color: transparent;"&gt;“Manulife is embedding AI across nearly every part of our business while equipping our colleagues with the capabilities to design, operate, and scale solutions,” said Shamus Weiland, Global Chief Information Officer at Manulife. "Akka provides a secure, scalable software foundation to support high volumes of business applications to accelerate meaningful value creation across our franchise. This partnership is rooted in our Responsible AI Principles, with a strong focus on governance, human oversight, and safety, as we advance toward becoming an AI-powered organization.”&amp;nbsp;&lt;/span&gt;&lt;/p&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;Manulife has been actively investing in and scaling AI capabilities since 2016. In the past three years, the firm has significantly increased the value enabled through advanced analytics and AI by expanding its AI solutions portfolio and increasing efficiency through data and AI platform investments, all driven by the company’s responsible AI framework. &lt;span&gt;In December 2025, Manulife announced that Adaptive ML would provide the reinforcement-learning-powered engine for finetuning and optimizing models within its enterprise AI platform. &lt;/span&gt;Manulife expects AI to generate $1 billion+ of enterprise value by 2027, with roughly one-fifth expected to come from improved efficiency.&lt;/span&gt;&lt;/p&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;&lt;a href="https://www.newswire.ca/news-releases/manulife-selects-akka-to-operationalize-agentic-ai-within-its-enterprise-ai-platform-804452481.html"&gt;GlobeNewswire release&lt;/a&gt;&lt;br&gt;&lt;/span&gt;&lt;/p&gt; 
&lt;h3&gt;&lt;span style="font-weight: normal; color: #ffce4a;"&gt;Related Information:&lt;/span&gt;&lt;/h3&gt; 
&lt;p&gt;&lt;span style="font-weight: normal;"&gt;Jodie Wallis speaks to &lt;a href="https://akka.io/blog/go-slow-to-go-fast"&gt;early results with Akka&lt;/a&gt;, setting a firm foundation of governance-first agentic AI, and the idea of "Go Slow to Go Fast".&lt;/span&gt;&lt;/p&gt;  
&lt;img src="https://track.hubspot.com/__ptq.gif?a=45500578&amp;amp;k=14&amp;amp;r=https%3A%2F%2Fakka.io%2Fblog%2Fmanulife-selects-akka-to-operationalize-agentic-ai&amp;amp;bu=https%253A%252F%252Fakka.io%252Fblog&amp;amp;bvt=rss" alt="" width="1" height="1" style="min-height:1px!important;width:1px!important;border-width:0!important;margin-top:0!important;margin-bottom:0!important;margin-right:0!important;margin-left:0!important;padding-top:0!important;padding-bottom:0!important;padding-right:0!important;padding-left:0!important; "&gt;</content:encoded>
      <category>AI</category>
      <category>Operations</category>
      <category>Resources</category>
      <category>News</category>
      <category>Memory</category>
      <category>Governance</category>
      <category>CAIO / CIO</category>
      <pubDate>Tue, 10 Mar 2026 00:55:53 GMT</pubDate>
      <guid>https://akka.io/blog/manulife-selects-akka-to-operationalize-agentic-ai</guid>
      <dc:date>2026-03-10T00:55:53Z</dc:date>
      <dc:creator>Team Akka</dc:creator>
    </item>
  </channel>
</rss>
