The week’s AI moves, a 5-minute read.
A tight Tuesday brief for CTOs, CFOs, product leads and founders — published as a LinkedIn Newsletter so you read it where you already work.
The week, a 5-minute read
One block of prose. No bullet soup. Written by the Pulse24 desk; edited by the Fact Auditor.
Pulse24's read
A dedicated opinion column — where we say what the evidence supports, nothing more, nothing less.
This week's numbers
Stories tracked, busiest categories, most active subjects, with deltas from last week.
Past issues/ 20 of 50
Full archive on LinkedIn →Repo Leaks, Token Overhead, and Record Fab Revenue.
GitHub agent repo leak, Claude Code's harness token overhead, and TSMC's record $39.62B quarter — Pulse24's AI briefing for 6–13 July 2026.
Prompt Markers, Reasoning Ceilings, and One-GPU Agents.
A researcher decoded hidden Claude Code prompt markers and Anthropic pulled them; developers mapped GPT-5.5 reasoning tokens clustering at 516; Ornith-1.0 targets one GPU.
Custom Silicon, Ford's Reversal, and Computer Use.
OpenAI unveiled its first inference chip with Broadcom, Ford brought back 350 engineers after an AI push fell short, and Google built computer use into Gemini Flash.
Open Weights, Code Review, and Agent Identity.
US curbs pushed buyers to open-weight models, two reports argued AI coding's hard part is no longer writing it, and agents got identities of their own.
Model Halts, Agentic Git, and Open Reasoning.
A US order halted Anthropic's new models, AI agents rewrote Git in Rust, and Hugging Face open-sourced a DeepSeek-R1 reasoning recipe.
Coding Costs, Apple on Gemini, and Open-Weight Gains.
AI coding tools may cost providers 10x what users pay, Apple rebuilt Siri on Google's Gemini, and DeepSeek's V4 reportedly beat GPT-5.5 on precision.
Chips, Code, and the Edge.
Memory drives most of your inference bill, a cost-optimisation layer forms around developer tools, and agentic models reach consumer silicon.
Agents and Zero-Days.
OpenAI and Notion ship agent platforms while Google confirms the first AI-driven zero-day exploit. This week's AI briefing for engineering leads and security architects.
Million Tokens on a MacBook.
Six months ago, running a frontier model locally was a research curiosity. This week, antirez shipped million-token inference on a MacBook — and Big Tech's cloud AI bill hit three quarters of a trillion dollars.
Nine Seconds.
AI code generation hits 80%, PyTorch Lightning compromised by supply chain malware, and a coding agent destroys production in nine seconds.
The Agent Stack Week.
OpenAI ships workspace agents, YourMemory open-sources agent memory, and DeepSeek V4 offers frontier performance at lower cost. New options across the agent stack — while server CPU supply tightens.
The Asymmetry Week.
Anthropic launches Claude Design knocking Figma 7%, valuation hits $800B on $30B revenue, and Microsoft fields in-house AI models on Azure.
Mythos, Anthropic's Enterprise Run, and Regulation Closes In.
Anthropic's Mythos Preview — an AI that finds and chains zero-days — just redrew frontier AI deployment. Project Glasswing, Wall Street trials, and a DoD supply-chain-risk label all landed in the same week.
Capital Pressure, Agent Integrity, and Chip Gaps.
Big Tech squeezes AI labs toward exits. Copilot injects ads into 1.5M PRs. Chinese execs quantify a 5-10 year chip gap. Three enterprise constraints shifted.
Sycophancy, Courts, and Inference Costs.
AI systems affirm harmful requests 49% more than humans. Court blocks Pentagon's Anthropic designation. TurboQuant cuts LLM memory 6x. OpenAI shuts Sora three months after Disney's $1B pledge — four enterprise AI assumptions shifted this week.
Superapps and Safeguards.
OpenAI merges three apps into one superapp. Meta AI agent leaks data in Sev 1. This week's trade-off: bigger platforms, bigger blast radius.
Agents, Approvals, and Reshuffles.
Amazon gates AI code after outage. OpenAI forms $10B PE venture. Meta and Atlassian cut thousands citing AI. Chrome opens DevTools to agents. Edition 33.
Compliance, Capital, and Confidence.
US tightens AI vendor control terms. Oracle cancels Texas expansion. Adoption stalls at 38%. Three constraints now gate AI scale-up more than compute does.
Policy Gaps, Power Pledges, and Stalled Adoption.
Claude was used in Iran military strikes while Anthropic fought the Pentagon in court. Oracle cancelled a Texas expansion. Adoption stalled at 38%.
The Vendor Reckoning.
Trump blacklists Anthropic over military AI safeguards. OpenAI wins Pentagon deal hours later. Block cuts 40% citing AI, stock surges. Australia sets March 9 enforcement deadline.