Saturday, September 26, 2026

TL;DR

Microsoft folded personal and enterprise Copilot into one Home/Code/Autopilot app, retreating from the consumer-chatbot race. Akamai signed an $11.6 billion cloud deal with Anthropic. A federal appeals court upheld the Pentagon's Anthropic blacklist. Sanders and Casar introduced a bill to ban AI superintelligence.

Act on this

  • Claude Code 2.1.283 added /doctor prompt-audit; run it to catch CLAUDE.md files, skills and commands still written with prompting patterns tuned for older models. Claude Code changelog #

Signals

The biggest lever on developer productivity has shifted from frameworks to how legible code is to an agent, not a human  #

Thorsten Ball argued Sept 26 that agents matter roughly a hundred times more than framework choice now, citing a roughly 100x drop in token cost over the past year - his own argument, not a measurement. Simon Willison dissented Sept 24, writing that the more he works with coding agents, the more convinced he is they make software engineering harder, not easier. Ball · Willison

A single success story with formal-verification tools doesn't hold up under controlled testing  #

Dan Luu's Sept 26 controlled test of 26 testing approaches on AI agents building a Zstd-like compressor found the default, no-special-instructions condition beat most guided techniques; formal methods including Lean and TLA+ clustered near or below average. That complicates Boris Cherny's Sept 22 account of 16 bug-fixing pull requests from a few Lean/TLA+ prompts. Demonstrated study versus one anecdote. Dan Luu · Cherny, via X

Current AI progress looks like 'lossy self-improvement,' not the recursive self-improvement labs and critics both assume  #

Nathan Lambert argued Sept 19 that inference-time scaling and mass agent deployment produce real but bounded efficiency gains, not compounding intelligence gains: automatable research is too narrow, parallel agents hit diminishing returns, and peak intelligence remains the hardest exponential to accelerate. He calls this 'lossy self-improvement,' not true RSI. His own argument, direct dissent from RSI framing. Nathan Lambert

News

SHIPMicrosoft relaunches Copilot as one Home, Code and Autopilot app #

Microsoft folded personal and enterprise Copilot into one app with Home, Code and Autopilot tabs, embedding Office apps inside it. Nadella called it a new OS for work; Bloomberg called it an exit from the personal-chatbot race. Microsoft · Bloomberg

WATCHAkamai signs an $11.6 billion, seven-year cloud deal with Anthropic  #

The deal, expandable toward $20 billion, covers CPU capacity rather than GPUs. Akamai issued Anthropic a warrant for up to 5% of its stock, vesting as spend crosses $3 billion increments, and expects $5.5 billion in related capex. Akamai · TechCrunch

WATCHFederal appeals court upholds the Pentagon's Anthropic blacklist, 2-1 #

The D.C. Circuit found the Defense Department had 'ample support' for barring Anthropic's Claude after it refused to allow use for 'all lawful purposes,' citing its own limits on autonomous weapons and domestic surveillance. Anthropic may seek en banc review. CNN · CNBC

WATCHSanders and Casar introduce a bill to ban AI superintelligence  #

The bill would create a Cabinet-level Department of Artificial Intelligence, pause advanced AI development pending new federal rules, and ban models exceeding human performance across most domains, with penalties up to corporate dissolution and 20 years in prison. Sanders Senate release · NBC News

SHIPClaude Code 2.1.283 adds a prompt-audit doctor check and gateway hints #

The release adds /doctor prompt-audit to flag CLAUDE.md files and skills written for older models, a gateway hint header for grouping requests, availableModelsMatch and deniedModels admin controls, and a Bedrock Mantle provider option. Claude Code changelog

The long view

A year ago Microsoft was pushing Copilot as a personal companion to rival ChatGPT, hiring Mustafa Suleyman to give it its own voice, while its weekly users stayed near 20 million against ChatGPT's roughly 400 million. This week Microsoft folded personal and enterprise Copilot into one Home, Code and Autopilot product built around work, where its 400 million-plus Microsoft 365 users already sit - Bloomberg called it an exit from the personal-chatbot race. If this holds, the assistant market settles along the line Microsoft just stopped contesting: work tools from Microsoft, personal chat left to OpenAI, Google and Meta.

Also noted

  • llama.cpp build b11195 (Sept 25) adds tiled matrix-multiply kernels for k-quants, 3 to 6x faster on large operations. llama.cpp release #
  • OpenRouter: GLM 5.3 Flash and DeepSeek V4.1 Flash each routed about 18 trillion tokens the week of Sept 17-22. OpenRouter  #
  • SemiAnalysis's ClusterMAX 3.0 (Sept 23) rated 323 GPU cloud providers; Nebius joined CoreWeave in the top Platinum tier. SemiAnalysis  #
  • Hamel Husain and Shreya Shankar published an updated AI Evals FAQ, Sept 18-21. Hamel Husain #
  • Jesse Vincent's Superpowers 6.4.1 (Sept 21) added OpenCode 2.0, Meta Muse and Qwen Code harness support. Jesse Vincent #