Field notes on building meaningful AI.

Insights from the front lines of the AI maturity curve.

The Craft of AI

July 21, 2026 · 8 min read
Deployable agent teams
Throughput is up, and so is the exhaustion: hand-juggling 10+ Claude sessions makes you an overstressed middle manager, not a superhero. The way out is the orchestration ceiling’s opposite — deployable agent teams that work a queue, follow explicit workflows, and check in only when necessary. What your harness needs, and the measured impact of running three teams.
By Luke Lin
Read article

A new way to work: autonomous agent teams

July 8, 2026 · 7 min read
Friday of my first week running the GTM team, the weekly brief had merged context and learnings from every GTM activity that week — without me managing a single file. How I graduated from manually steering one Claude agent to running a coordinated, autonomous team of 11 from Slack.
Read more

Harness engineering when reasoning is exponential

June 17, 2026 · 6 min read
Same model, same task, two very different Mondays. A harness is the scaffolding around a model that turns raw reasoning into work that reliably gets done. Six principles for building one that delivers today and grows with the models of tomorrow — when reasoning 10x’s again.
Read more

The Return of the Eval

June 10, 2026 · 6 min read
A year ago the loudest voices called evals dead. At Arize:Observe, 700+ builders proved otherwise. The saga from “evals are everything” to “evals are a scam” to the position that settled: evals are the telemetry of agent performance — and three reasons you need them now.
Read more

Design MCPs for the Agent, Not the API

June 5, 2026 · 6 min read
An agent isn't a developer reading your docs — it's a new kind of user that pays for every token and makes worse choices the more options you give it. The MCP design principles we earned shipping 50+ servers: fewer, more expressive tools, agent-readable outputs, and auth that doesn't trade safety for convenience.
Read more

Multi-Player Agents

June 3, 2026 · 8 min read
Building an agent for yourself is easy. Pointing it at a team turns it into a negotiation across identity, memory, guardrails, orchestration, and cost. A four-tier framework for the compounding complexity of multi-player agents.
Read more

The AI Coding Maturity Curve

May 20, 2026 · 7 min read
Hand your engineers Claude Code and Lovable without rigor and you industrialize AI Slop Debt. This is the maturity curve from raw agents to coding harnesses, and the move that holds up past 2026.
Read more

Let’s build something meaningful.