Article
LSP for AI Agents: Real Promise, Messy Reality
The Language Server Protocol has been around since 2016, and AI agents could genuinely benefit from it. Here's why most of them don't — and what would need to change.
#AI Agents
20 articles
Article
Codebase graph servers: MCP for AI agents, genuine breakthrough or just middleware?
Code graph MCP servers index your codebase as a symbol graph and expose it to LLM agents. Powerful on large projects — but useful for 90% of codebases?
Deep Dive
An Agent Hallucinated an Attack, and Proving It Took Three Hours
A sub-agent invents a malicious instruction out of thin air. A forensic walkthrough of telling hallucination apart from a real compromise.
Article
You can't follow everything. Stop trying.
In agentic AI, trying to follow everything has become the best way to master nothing. Here's why exhaustive monitoring makes you less competent — and what to do instead.
Brève
DESIGN.md — making your design system readable by agents
Google Labs releases DESIGN.md, a plain-text format that gives coding agents persistent design context: YAML tokens + Markdown rationale, with a CLI for lint, diff, and export.
Brève
Open Knowledge Format: Google standardizes AI agent memory
Google just shipped OKF, an open spec for representing knowledge as plain Markdown files — the format AI agents needed to share context without vendor lock-in.
Article
Agent Skills: giving your agents procedural memory
Agent Skills is an open format for injecting contextual instructions into AI agents. A look at the standard, its progressive disclosure architecture, and skill-creator for industrializing skill production.
Article
Lost in the Middle: Why Your LLM Ignores What You Give It
LLMs don't read long contexts well, not because of capacity limits, but by design. Two studies measure the gap. What it means for how you architect your systems.
Article
Token Economy: Three Tools Attacking the Problem Differently
RTK compresses tool outputs. Caveman forces LLM brevity. DCP prunes context history. Three tools, three layers — and a counterintuitive argument: fewer tokens, better reasoning.
Article
Agentic AI: what it actually is
Between devs convinced AI will automate everything and those who think it will never truly understand code, there's a more nuanced reality, and a more useful one.
Brève
pure.md — a markdown proxy for AI agents, with built-in MCP
pure.md markdown proxy for AI agents — converts any URL to clean markdown in one GET request. Handles JS SPAs, PDFs, bot detection. 28K tokens vs 143K with Jina.
Article
Devoxx France 2026: AI everywhere, but with nuance
A recap of 3 days at Devoxx France 2026 — from generative AI to CI/CD security, Docker Sandboxes to LLM guardrails, here's what the conference says about the state of the art (and its limits).
Article
The Agentic AI Foundation — standardizing agentic AI without shooting yourself in the foot
The Linux Foundation launches AAIF with MCP, goose, and AGENTS.md. Open governance to prevent ecosystem fragmentation.
Brève
WebMCP — MCP in the browser, finally
A JavaScript API that turns your site into an MCP server. Agents talk to your app through structured tools instead of scraping the DOM.
Deep Dive
AI Agent Guardrails: Every Source of the Problem
AI agents have a fundamentally different attack surface than chatbots. A complete map of the threats — from indirect prompt injection to tool supply chain attacks.
Deep Dive
Building an agent-first harness with OpenCode custom tools
How to give an AI agent all the tools it needs to run, analyze, and see your application — and become truly autonomous in detecting and fixing errors.
Article
The review-manager: anatomy of a code review orchestrator
A deep dive into the review-manager powering opencode-team-lead — how it selects reviewers, isolates their contexts, arbitrates disagreements, and produces a structured verdict without reading a single line of code.
Article
opencode-team-lead: an agent that plans, delegates, and never writes code
On long tasks, an all-in-one agent loses track and reviews its own code with the same blind spots that produced it. I built opencode-team-lead to fix both problems with a strict delegation pattern.
Article
Your agents don't need memory
In a codebase that serves as the source of truth, everything an agent needs to 'know' should be mechanically verifiable. Giving your agents memory is patching in prose what should be encoded in the harness.
Brève
Harness engineering: what Anthropic and OpenAI are actually doing with their agents
Anthropic and OpenAI published their harness engineering post-mortems weeks apart. Different problems, same conclusions — here's what to take away.