Editorial coverage of artificial intelligence — the models, the infrastructure, the businesses being built on top of them.
Our AI coverage is grounded reporting on the systems, businesses, and second-order effects that actually matter. We audit predictions against outcomes, map production patterns as they emerge, and separate the frontier story from the small-model story. Every piece is reviewed by a human editor with a working understanding of the underlying research.
Editor's Picks
Eighteen months after Anthropic open-sourced the Model Context Protocol, MCP has quietly become the default plug for AI agents. We surveyed the public registry and production deployments to map what works, what breaks, and where the next gap is.
The narrative that big always wins is a partial narrative. We map the class of small, task-focused models that quietly took a growing share of production AI work — where they win, where they still lose, and what changed.
The AI infrastructure conversation is bad because the layers blur. Most arguments about 'the AI stack' confuse five separate markets with five separate competitive dynamics. Here is the framework we use to keep them apart.
Latest
The MCP registry is a directory of hundreds now. Most are hobby projects. We surveyed which servers are actually in production use in 2026, grouped them by domain, and flagged the ones we would not deploy — with reasons.
SWE-bench Verified is the benchmark every AI coding tool cites. But the score you see quoted often obscures what the test actually measures — and what it does not. We audited the methodology and the leaderboard against the claims teams make from them.
The narrative that big always wins is a partial narrative. We map the class of small, task-focused models that quietly took a growing share of production AI work — where they win, where they still lose, and what changed.
The AI infrastructure conversation is bad because the layers blur. Most arguments about 'the AI stack' confuse five separate markets with five separate competitive dynamics. Here is the framework we use to keep them apart.
Eighteen months ago the AI conversation was wall-to-wall predictions. We pulled the most-cited public claims about 2026 from late 2024 — AGI timelines, job displacement, autonomous agents, capital returns — and graded each against what actually happened.
Eighteen months after Anthropic open-sourced the Model Context Protocol, MCP has quietly become the default plug for AI agents. We surveyed the public registry and production deployments to map what works, what breaks, and where the next gap is.
An opinionated, no-nonsense guide to assembling your first production AI stack in 2026 — what to pick, what to skip, and what to defer until Series A.
Most teams treat Claude's prompt caching like a checkbox. Here's the production tuning playbook from three companies that dropped their bills 70-85% in a month.
The Anthropic Agent SDK and Claude Skills ecosystem went from new toy to default in roughly nine months. Here is what they are, why they won, and what to build on them.
The European Commission issued its first AI Act infringement notices in May 2026. A practical compliance plan for small teams: what you owe, what you can ignore, and what costs €38K versus €4K to get right.
Six months and 17 production deployments later, here's when a Claude Skill wins, when an MCP server wins, and the hybrid pattern most teams miss.
Scaling laws stopped buying us reasoning. The next phase of AI is neuro-symbolic, world-model-driven, and considerably stranger than another transformer.