Trails

Traditional UX has a hidden assumption: humans on both sides. That assumption is breaking.

We obsess over prompt engineering when we should obsess over constraint engineering. The prompt is a request. The constraints are the physics. The code is what ships.

I've spent a month running a personal AI agent on a Mac Mini. The technology works. The promises don't — at least not the way we were told they would.

A technical breakdown of multi-agent orchestration: Claude's Agent Teams, OpenAI's Agents SDK, Google Antigravity, Gas Town, Beads, and the community alternatives.

On the quiet ethos connecting Buttercup, Enpass, pass, and VeraCrypt — and why it matters more now that AI agents need your credentials too.

I built an internal context engine for our 1,765-doc knowledge base. My teammate independently found a better one. The difference was a reranking step I hadn't thought of. Here's what the emerging context engineering stack actually looks like — and why no single tool gets it right yet.

Backend testing is a solved problem. Frontend testing for stateful, streaming, visual UIs is not. That's exactly where AI coding agents are generating the most code.

Design is no longer artifact creation—it's constraint architecture. The three-layer model for the agentic economy.

When anyone can spin up a coding agent and ship something workable, what actually matters? Not the word soup — the real answer.

What interviewing looks like when AI can write the code but not make the call — and why the same break is happening in every field that used output as proof of understanding.

Infrastructure config is collapsing into application code — the same pattern as build tooling a decade ago. Void showed us the destination. The question is whether you want to ride the platform or steal the pattern.

Large engineering orgs already solved coordination for thousands of humans. How much transfers to thousands of agents — and where does entropy win?

Every event must carry everything a repair bot needs to act without asking questions. Don't scatter breadcrumbs — ship holograms.

We don't sell answers. We sell receipts. The receipts are what earn the right to call something true.

How Convex challenges our mental models of databases—not relational, not NoSQL, but something new.

Anthropic wiped $285 billion off software stocks this week. The market is panicking about AI replacing tools. But the real disruption is what it means for the people who build them.

The agent era is splitting infrastructure into two camps: those building new primitives and those retrofitting old ones. The bet you make reveals what you think agents are.