--- title: Weekly Dose #12 - Open Weights, Stateless Agents and the... description: A curated weekly roundup of five major AI/ML developments from late July to early August 2026. Moonshot AI released full weights for Kimi K3, a... canonical: https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Weekly Dose #12 - Open Weights, Stateless Agents and the New AI Control Problem | daily.dev og:description: A curated weekly roundup of five major AI/ML developments from late July to early August 2026. Moonshot AI released full weights for Kimi K3, a... og:url: https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e og:image: https://api.daily.dev/og/posts/p1fcBez5E.png og:image:alt: Weekly Dose #12 - Open Weights, Stateless Agents and the New AI Control Problem og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Weekly Dose #12 - Open Weights, Stateless Agents and the New AI Control Problem **[Machine Learning Pills](https://daily.dev/sources/mlpills)** · 9 min read · 2 upvotes · 0 comments ## Summary A curated weekly roundup of five major AI/ML developments from late July to early August 2026. Moonshot AI released full weights for Kimi K3, a 2.8-trillion-parameter MoE model requiring data-center-scale infrastructure. The MCP Go SDK shipped a stateless protocol rewrite (version 2026-07-28), removing session IDs and enabling standard load balancing. Microsoft announced Project Perception, a multi-agent security system using specialist model routing to cut costs by ~50%. The EU AI Act's Article 50 transparency requirements took effect, mandating AI disclosures and machine-readable content markings. GitHub fully retired its Models service while AWS moved several first-generation AI services (Bedrock Agents, Kendra, Q Business) to maintenance mode for new customers. The overarching theme: AI systems are becoming easier to connect but harder to own, with managed services carrying lifecycle risk and open weights not equating to accessible infrastructure. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What changed in the MCP protocol version 2026-07-28 compared to earlier versions? The 2026-07-28 MCP protocol removes the initialize handshake and Mcp-Session-Id, sends client information through per-request metadata instead, and introduces server/discover for capability negotiation. This makes MCP stateless so any request can reach any server instance rather than being pinned to the session-creating instance, allowing remote MCP servers to run behind ordinary round-robin load balancers without sticky routing. The official MCP Go SDK shipped full support for this version on 28 July, though it also deprecates Roots, Sampling and Logging while preserving them during a transition, and can negotiate down to the older 2025-11-25 protocol for compatibility. _Teams migrating MCP servers to the new stateless protocol can track breaking changes like these on daily.dev._ ### How many parameters does Moonshot's Kimi K3 model have and what hardware does it need to run? Kimi K3 is a mixture-of-experts model with 2.8 trillion total parameters, 104 billion activated per token, 896 experts (16 selected per token), and a one-million-token context window. Moonshot recommends supernode configurations with 64 or more accelerators for deployment, using MXFP4 weights and MXFP8 activations, with support through Transformers and vLLM. Its hosted API costs $0.30 per million cache-hit input tokens, $3 for uncached input, and $15 for output. _Teams weighing self-hosting versus hosted APIs for large open-weight models like this can find deployment cost breakdowns on daily.dev._ ### What happened to GitHub Models and which AWS AI services stopped accepting new customers? GitHub Models was fully retired for all customers on 30 July, removing its playground, model catalogue, inference API, and bring-your-own-key endpoints, with GitHub redirecting users to Microsoft Foundry and Copilot. The same day, AWS stopped accepting new customers for Amazon Bedrock Agents (renamed Bedrock Agents Classic), Amazon Kendra, and Amazon Q Business, recommending new agent workloads move to Bedrock AgentCore, though existing customers retain support in what amounts to maintenance mode rather than shutdown. _Engineers depending on managed AI services can plan exit strategies before retirements like these force a rewrite, tracked on daily.dev._ ## Similar posts on daily.dev - [This Week in AI: Open Weights, Frontier Models, Sandbox Escapes, Voice & AI Detection](https://daily.dev/posts/this-week-in-ai-open-weights-frontier-models-sandbox-escapes-voice-ai-detection-tx92vwcwt) · ThursdAI · 1 upvotes · 1 comments - [Weekly Dose \#11 - AI Agents Are Getting Easier to Build, and Harder to Control](https://daily.dev/posts/weekly-dose-11---ai-agents-are-getting-easier-to-build-and-harder-to-control-ynvfl6cn9) · Machine Learning Pills · 1 upvotes · 0 comments - [The State of Open Source AI — V1.0 · July 2026](https://daily.dev/posts/the-state-of-open-source-ai-v1-0-july-2026-gjzwgazbw) · Hacker News · 1 upvotes · 0 comments - [Weekly Dose \#9 - AI Is Becoming an Access-Control Problem](https://daily.dev/posts/weekly-dose-9---ai-is-becoming-an-access-control-problem-odg9xd3us) · Machine Learning Pills · 2 upvotes · 0 comments - [The Current State of Agentic AI](https://daily.dev/posts/the-current-state-of-agentic-ai-yp4wym5dv) · Machine Learning Mastery · 3 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#ai-governance](https://daily.dev/tags/ai-governance), [#ai-gateway](https://daily.dev/tags/ai-gateway) [View this post on daily.dev](https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Weekly Dose #12 - Open Weights, Stateless Agents and the New AI Control Problem","url":"https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e"},"datePublished":"2026-08-02T15:04:51.654Z","dateModified":"2026-09-13T19:55:00.671Z","description":"A curated weekly roundup of five major AI/ML developments from late July to early August 2026. Moonshot AI released full weights for Kimi K3, a...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/a19b2f9de45de5e4f3115bb744593333?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/a19b2f9de45de5e4f3115bb744593333?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Machine Learning Pills","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Machine Learning Pills","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/4a65b589ef234a99a2fbc0c4cf043d7f","url":"https://daily.dev/sources/mlpills"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,mcp,ai-governance,ai-gateway","timeRequired":"PT9M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Machine Learning Pills","item":"https://daily.dev/sources/mlpills"},{"@type":"ListItem","position":3,"name":"Weekly Dose #12 - Open Weights, Stateless Agents and the New AI Control Problem"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/weekly-dose-12---open-weights-stateless-agents-and-the-new-ai-control-problem-p1fcbez5e#faq","mainEntity":[{"@type":"Question","name":"What changed in the MCP protocol version 2026-07-28 compared to earlier versions?","acceptedAnswer":{"@type":"Answer","text":"The 2026-07-28 MCP protocol removes the initialize handshake and Mcp-Session-Id, sends client information through per-request metadata instead, and introduces server/discover for capability negotiation. This makes MCP stateless so any request can reach any server instance rather than being pinned to the session-creating instance, allowing remote MCP servers to run behind ordinary round-robin load balancers without sticky routing. The official MCP Go SDK shipped full support for this version on 28 July, though it also deprecates Roots, Sampling and Logging while preserving them during a transition, and can negotiate down to the older 2025-11-25 protocol for compatibility. Teams migrating MCP servers to the new stateless protocol can track breaking changes like these on daily.dev."}},{"@type":"Question","name":"How many parameters does Moonshot's Kimi K3 model have and what hardware does it need to run?","acceptedAnswer":{"@type":"Answer","text":"Kimi K3 is a mixture-of-experts model with 2.8 trillion total parameters, 104 billion activated per token, 896 experts (16 selected per token), and a one-million-token context window. Moonshot recommends supernode configurations with 64 or more accelerators for deployment, using MXFP4 weights and MXFP8 activations, with support through Transformers and vLLM. Its hosted API costs $0.30 per million cache-hit input tokens, $3 for uncached input, and $15 for output. Teams weighing self-hosting versus hosted APIs for large open-weight models like this can find deployment cost breakdowns on daily.dev."}},{"@type":"Question","name":"What happened to GitHub Models and which AWS AI services stopped accepting new customers?","acceptedAnswer":{"@type":"Answer","text":"GitHub Models was fully retired for all customers on 30 July, removing its playground, model catalogue, inference API, and bring-your-own-key endpoints, with GitHub redirecting users to Microsoft Foundry and Copilot. The same day, AWS stopped accepting new customers for Amazon Bedrock Agents (renamed Bedrock Agents Classic), Amazon Kendra, and Amazon Q Business, recommending new agent workloads move to Bedrock AgentCore, though existing customers retain support in what amounts to maintenance mode rather than shutdown. Engineers depending on managed AI services can plan exit strategies before retirements like these force a rewrite, tracked on daily.dev."}}]} ```