Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Multi-agent LLM coordination in a Pokémon 2v2 environment
Oliver Grabner introduces an open environment where two LLM agents share a side in 4-player Pokémon Showdown battles and can talk before every turn, measuring whether natural-language coordination wins more battles and what it costs.
3 min · 737 words
Will Lachance refines his Implementation Decision Records (IDRs), short documents meant both to get better output from LLM coding agents and to leave a durable record of implementation decisions, arguing they should capture the problem, goals, non-goals and intended result rather than a step-by-step plan.
3 min · 609 words
Kyle Kukshtel writes up his Indie Rodeo talk on making games more legible to AI coding agents: practical ways to structure a game project, tooling and feedback loops so agents like Claude Code can iterate faster and more reliably on game development.
27 min · 6,159 words
Geoffrey Huntley argues that teams who cannot make time to experiment with AI agents and loop engineering are effectively preparing themselves to be replaced by those who do.
10 min · 2,250 words
Harness²: Recursive Agent Harnessing for an Open WorldRecursive Agent Harnessing for an Open World — Google Cloud AI Research & Northwestern
Google Cloud AI Research and Northwestern introduce Harness², an open-source framework that recursively improves an AI agent's execution harness (prompts, tools, memory) via Propose–Probe–Compose without ground-truth verifiers, gaining up to 11.5 points on WorkBuddy-Bench.
13 min · 3,004 words
What Is poppy.json? The /.well-known File That Tells AI Agents How to Work With Your Company
A walkthrough of poppy.json, the Personal Agent Protocol (PAP/Poppy) discovery file at /.well-known/poppy.json: required fields for organization, OAuth, agent/API/web interfaces and extensions, plus a fictional worked example of Draft 0.1.
4 min · 888 words
500+ Billion Tokens Later: Letting AI Agents Decompile A First-Person Shooter
Maurice Heumann describes three months spent orchestrating AI coding agents to produce an accurate, compilable C++ decompilation of a popular first-person shooter, focusing on agent orchestration, infrastructure, harness design and verification rather than the game itself, and estimates 600 to 700 billion tokens were spent.
10 min · 2,206 words
Five months treating bugs like patients and coding agents like a medical team
Cockroach Labs' Adam Storm describes the "hospital", an agent system that triages GitHub issues and assigns specialist coding agents. It added IBM Db2 support to the MOLT migration tool in under two days for a $4,172 token bill, versus nine months for Oracle in 2024, and the post covers what broke, human review and training junior engineers.
19 min · 4,382 words
Voxlocal: a minimal voice agent written in Rust
Sam Khawase walks through building Voxlocal, a minimal local voice agent in Rust, explaining each stage of the pipeline from audio capture and Whisper speech recognition through text normalization, embeddings and retrieval, LLM tool routing and Piper speech synthesis, and measuring where the time goes.
13 min · 3,024 words
Prime Intellect describes how Prime Agent orchestrated a swarm of over 2,000 agents across 10,000+ Prime Sandboxes and 200 billion tokens to rewrite itself from TypeScript to Rust in two weeks, covering why Rust, finite-state-machine orchestration, architecture redesign, and cold start, memory and install-size gains.
9 min · 2,133 words
Why Are Coding Agents So Dumb?
Michael Lynch argues that while LLMs keep getting better, the coding agents that drive them have barely improved since early 2025: he separates the model from the agent, catalogs the workflow gaps and bugs he still hits in Claude Code, Codex, OpenCode, Cline and Pi, and sketches what a better agent would do.
10 min · 2,316 words
a16z's Seema Amble, Ivan Makarov and David Borecky argue AI is reshaping the CFO role from accountant to banker to builder: data stops being the bottleneck, finance engineers build their own tools, teams get smaller, planning becomes continuous, controls become product features, and finance's mandate grows.
9 min · 2,003 words
Benchmarking Bot Detection Systems Against Modern AI Agents in 2026
Proof of Human researchers revisit their bot detection benchmark, testing Proof of Human, Google reCAPTCHA v3, hCaptcha, Cloudflare Turnstile and Cloudflare Bot Management against scripted browsers, stealth browsers and AI browsing agents completing a five-question survey. Detection rates ranged from 9% to 99%; they discuss where each system failed, limitations, and why behavioural signals matter more.
6 min · 1,293 words
A green exit code is not evidence that the work happened
Why agent-run jobs that exit 0 can still have done nothing—and how to assert on real work products and alert on silent zero-effect “success.”
10 min · 2,257 words
Production-grade LLMs and agents: a field guide
Stack Overflow’s maturity map for taking LLM agents from demos to production: determinism, evaluation, confidence, safety, operability, and disciplined coding-agent builds.
3 min · 704 words
The tests your agent writes for itself
Mashhood Rastgar responds to Kun Chen's DeepSWE v1.1 experiment, where banning Claude Sonnet 5.5 from writing its own tests left success flat while cutting time and cost, arguing the real lesson is about unrequested agent-written tests versus human-specified tests that encode intent.
3 min · 768 words
Microsoft Execution Containers: Policy-driven containment for AI agents
Microsoft announces general availability of Microsoft Execution Containers (MXC), a Windows containment layer that lets developers and IT define which files and network destinations an AI agent may use and enforces those policies at runtime, alongside upcoming agent identity via Microsoft Entra and manageability tooling.
8 min · 1,940 words
Introducing Personal Agent Protocol
Sierra and Meta, with partners including Genesys, Instinct, Rocket, Shopify, Stripe and Walmart, announce Personal Agent Protocol — an open standard for how personal AI agents authenticate, get customer consent and interact with businesses via websites, APIs or company agents, with a v0.1 spec planned later this month.
3 min · 739 words
Agentic Systems Are the Future of Brands
A Little Plains research note arguing that growing teams and AI agents cause brand drift, and that studios should hand over agentic brand systems: living kits of positioning, voice, visual rules, assets, tokens and working components that people can browse and agents can read, so every new touchpoint strengthens the brand.
9 min · 2,158 words
Introducing the Personal Agent Consent & Trust Protocol (PACT)
Decagon open-sources PACT, co-developed with Instinct, so businesses can verify which personal AI agent is acting for a customer and what that customer authorized — built on Agent2Agent and OAuth 2.0 — and joins the Personal Agent Protocol working group around Meta’s Muse.
4 min · 910 words