Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
David Heinemeier Hansson follows up his Rails World talk telling programmers to put down the pencils: coding agents are now good enough that writing most code by hand is no longer economically viable, AI is more like gaining capable new coworkers than another tool, and programmers who refuse to collaborate with agents are the ones who should worry.
2 min · 382 words
Prompt injection in HTTP headers: what 217 million web responses show
Signals Corps scanned Common Crawl HTTP responses and found 7,959 sites sending headers aimed at AI agents, including 30 prompt-injection attempts. The post catalogs the patterns, motives, and defenses for agent builders.
28 min · 6,512 words
Jev-Driven SRE Diagnosis: What Worked and What Failed
The SREGym team builds a Kubernetes incident-diagnosis pipeline with no LLM agent: a programmatic collector gathers cluster evidence and TypeSafe AI's Jev decision model picks root-cause candidates and supporting evidence. It passed 80 of 105 diagnoses (76.2%) across 21 SREGym-Lite faults with a 14.6s median, close to an LLM agent's score at far lower cost, and the post dissects the faults it failed.
7 min · 1,572 words
A Terminal Protocol for Program Status (OSC 7501)
Mitchell Hashimoto introduces OSC 7501, a generic terminal escape sequence that lets programs report whether they are idle, working, blocked on the user, finished or failed, and why. He argues heuristics and non-terminal APIs fall short, especially for agentic inboxes tracking many coding agents, and describes implementations in libghostty and Rex.
5 min · 1,236 words
Armin Ronacher explains why Pi 1.0 added MCP support through Codemode despite his long preference for bash and scripts: bash can only compose programs that run in the execution environment, while Codemode lets the model orchestrate harness-side tools like image reads and sub agents in a sandboxed QuickJS/WASM runtime, plus the open problems that remain.
12 min · 2,672 words
Meta's Muse Is An Adorable Privacy And Security Dumpster Fire
Karl Bode argues that Meta's agentic AI assistant Muse, launched with a zero-day flaw that let attackers spy on Mac users, shows how tech giants are rushing out highly privileged AI agents with little regard for privacy or security while lobbying against regulation and open alternatives.
4 min · 825 words
Fork-and-Flush: Escaping Idea Basins in Autoresearch Agents
Microsoft Research paper on idea basins in long-horizon autoresearch agents, and a periodic fork-and-flush intervention that keeps the workspace but resets chat context—reporting +66% vs single-run under equal compute across 13 tasks.
2 min · 400 words
Property Testing with Agent Swarms
Inanna Malick shows how to have coding agents build property-testing machinery (reference models, generators and assertions) for a repo, then turn the failures into small, reviewable fixes. Pointed at OpenAI's Codex, the public playbook found thirteen distinct correctness bugs, and runs on jj, uv, Prometheus and Babel produced merged upstream fixes.
6 min · 1,325 words
hutch: local code reviews in emacs for the mildly disenfranchised
The author introduces Hutch, a Magit-based code review interface for Emacs that runs an LLM review agent locally on staged, unpushed or branch changes, favours verified SEARCH/REPLACE patches over prose comments, traces agent runs with Perfetto, and reports eval results across 40 PRs and several models.
8 min · 1,836 words
Against the Personal Agents Theory of Everything
In a guest essay for The Leverage, Nathan Baschez argues that personal agents are not the end game: the next paradigm is AI factories, shared, specific, and scrutinized systems that teams build together, which he says are economically inevitable because they enable specialization and gains from trade. He also offers a pattern language for building one.
10 min · 2,223 words
Strands Decider: Why Not an Encoder?
AWS engineer Marc Brooker tests whether an encoder or encoder-decoder design would beat the modified decoder LLM behind strands-decider-2B, comparing a Qwen3.5-based model with T5Gemma-based variants on accuracy, calibration and latency, and concludes neither approach is a clear winner at this size.
3 min · 749 words
The Missing Model: How AI Agents Behave Over Months
Muhammad Hanzla argues labs lack a study design for how AI agents behave across months of real work, and outlines why that longitudinal model should be built before longer autonomy.
15 min · 3,495 words
Stefano Saitta pushes back on the claim that coding agents kill unit tests, arguing for a pragmatic mix of E2E, integration, and still-useful unit tests in an AI-assisted workflow.
4 min · 839 words
We're going to need default hard budget caps on pretty much everythingHard spend limits should be the default for usage-based APIs and agent-driven systems.
Simon Willison argues that pay-by-usage services need default hard budget caps—not soft email warnings—so coding and personal agents cannot rack up surprise bills. He points to AWS spend limits and Google Cloud Spend Caps as early moves in that direction.
2 min · 529 words
The Agent Said It Was Done. The Database Disagreed.
Microsoft’s ThinkingBox grades AI agents on backend records they leave behind—not chat claims—running isolated MCP tool sessions and scoring terminal state consistency across repeated trials; now available on Hugging Face.
13 min · 3,044 words
Agents Don't Need Memory. They Need Documentation.
Kevin Liao argues agent systems fail less from missing long-term memory than from missing written project docs—README, architecture, and runbooks that agents can read and update like humans.
4 min · 1,011 words
Lessons from building an automated research scaffold
Arcadia Impact / Equistamp / UKAISI report: an automated alignment-research scaffold failed to uplift researchers on conceptual work, but surfaced auditing, monitoring, and scaffold–model interaction failure modes worth studying next.
4 min · 989 words
Swarms v16 'Overclock': Token Accounting, Decision Models, MCP Deployment, and Runs That Start Clean
Swarms v16 Overclock changelog: per-agent token accounting, DecisionModel, MCPDeployer, TreeOfThoughts, cleaner run isolation, and faster tool handling — Sept–Oct 2026 release notes.
9 min · 2,045 words
OpenClaw integrates Tencent's AI-Infra-Guard into ClawScan so every skill and plugin uploaded to ClawHub runs through AIG as part of security review, with benchmarks and a feedback loop.
2 min · 392 words
Three AI Agents, Two Countries, and One Very Uneven World Wide Web
Roya Pakzad compares GPT, Claude, and Muse on multilingual research tasks across languages and jurisdictions, documenting gaps in source access, human-in-the-loop oversight, and observability.
5 min · 1,206 words