Latest writing

A hand-drawn map showing the Harness Engineer between developer experience, platform engineering, quality engineering, and AI systems
Featured

harness-engineeringresearch

Harness Engineer: Role, Skills, and Roadmap

A practical definition of the Harness Engineer role, its core skills, a five-stage maturity model, and a focused 90-day adoption roadmap.

A hand-drawn verification pyramid rising from deterministic checks to human judgment

Verification-First: Measure Evidence, Not Confidence

A hand-drawn stack of six harness layers from context and tools to feedback and verification

Harness Engineering: Six Layers Around an Agent

A hand-drawn system map placing a model inside context, tools, runtime, task state, verification, and human control

Coding Agent Is More Than a Model

A hand-drawn ReAct loop connecting a working thought, tool action, external environment, observation, and updated plan

Reason + Act: When Models Started Using Tools

More notes

07
  1. A hand-drawn staircase showing coding assistance expanding from token completion to a shared AI pair programming workflow

    From Autocomplete to AI Pair Programming

    How coding assistance expanded from local symbol completion into a conversational, review-driven AI pair programming workflow.

  2. A hand-drawn continuous integration loop connecting a small commit, build, tests, evidence, and diagnosis

    How Continuous Integration Turned Feedback into Infrastructure

    How continuous integration turned private test runs into fast, shared infrastructure for learning from every small software change.

  3. A hand-drawn test harness loop arranging state, acting on a system, observing results, and deciding with an oracle

    What Test Harnesses Taught Us Before AI

    Five design lessons from test harnesses that still define reliable automation—and now shape how we build coding-agent environments.

  4. A hand-drawn timeline tracing harness engineering from test harnesses to agent-first systems

    Harness Engineer: From Test Harnesses to Agent-First Engineering

    How test harnesses, CI, tool-using agents, and feedback systems evolved into the emerging discipline of harness engineering.

  5. Abstract paths converging around a luminous Astra core

    GPT-6 Astra Costs More. Use It Where It Compounds.

    GPT-6 Astra costs more per token. Here is where larger, more capable workflows may still return more value—and how to measure that.

  6. Glowing geometric rules transforming into structured JSON objects

    Changing Rules Without Rebuilding the App

    How to ship remotely managed business rules safely, especially when client releases pass through stores or platform review.

  7. Neovim editing the dotfiles Lua plugin configuration with the project explorer open

    My Current Neovim Workflow: Snacks, Blink, and Profiles

    A tour of the Neovim workflow I actually use today: focused profiles, Snacks navigation, Blink completion, and explicit tooling.

Series

01
A hand-drawn map placing reliable agent work between context, tools, feedback, and guardrails

Complete9 / 9 published

Harness Engineering

How test harnesses, CI, tool-using models, and coding agents became the discipline of designing environments where agents do reliable work.

  1. 01Harness Engineer: From Test Harnesses to Agent-First EngineeringPublished
  2. 02What Test Harnesses Taught Us Before AIPublished
  3. 03How Continuous Integration Turned Feedback into InfrastructurePublished
  4. 04From Autocomplete to AI Pair ProgrammingPublished
  5. 05Reason + Act: When Models Started Using ToolsPublished
  6. 06Coding Agent Is More Than a ModelPublished
  7. 07Harness Engineering: Six Layers Around an AgentPublished
  8. 08Verification-First: Measure Evidence, Not ConfidencePublished
  9. 09Harness Engineer: Role, Skills, and RoadmapPublished
1 / 1