Skip to content
Agent Engineering Lab
Menu

Build agents that survive contact with reality.

Most agents are workflows in costume. Reliability lives in the harness, the loop, and the gate, not the model. This lab publishes the evidence.

Start here Read the book The evidence
An engraved tightrope walker heading toward a brick-red sun

In the Lab

  1. The Book Agentic AI for Serious Engineers, from first tool call to production 19 chapters, 7 free here
  2. Architecture Blueprints for what a system can do, may do, and can prove it did 8 blueprints
  3. Governance Standing up AI governance from nothing, one step at a time 7 of 22 steps
  4. Patterns A pattern language for shipping agents, anti-patterns included 26 patterns
  5. Labs Experiments with the repo, the data and the seed published 7 reports
  6. Field Notes Notes from inside the work of building agents 4 notes
  7. Signal What to keep and what to skip in the talk around agents 4 pieces

New here? Pick a path. The Lab's position is in the manifesto, its terms in the glossary.

Latest work

LAB-003 · Lab

The baseline got the right answer and emitted one dispatchable action in thirty runs

Lab-002's sixty committed run records, scored twice. Once on the answer and once on the path. The two metrics rank the two systems in opposite directions, and the single assertion that explains it is whether the proposed remediation could actually be executed.

Sep 21, 7 min read
  1. STEP 1 · GuideUpdated Nov 11, 9 min read

    Agree what governance is for

    Before a policy, a committee or a single form, agree what governance is for: deciding what ships, naming who decides, and keeping the evidence behind each decision.
  2. ARCH-002 · BlueprintUpdated Sep 21, 13 min read

    Detector Architecture

    Four detection techniques, a sourcing decision people mistake for a fifth, one interface contract, and the calibration problem that breaks every ensemble built by averaging scores that do not mean the same thing.
  3. LAB-004 · LabSep 21, 10 min read

    Prompt hardening moved three models three different ways. The gate moved all of them to zero.

    Fifty support tickets, thirty carrying injected instructions, run through three guard configurations on three local models. Compliance and reachability are measured separately, because they are not the same number and only one of them is under your control.
  4. LAB-005 · LabSep 21, 9 min read

    The poison was stored, retrieved and cited every time. What changed was what it was allowed to cause.

    An agent reads an incident ticket carrying a false causal claim, remembers it, and meets a matching incident later with a clean ticket. Four configurations, measured on three separate things: whether the claim is stored, whether it is cited, and whether it reaches a forbidden action.
  5. LAB-006 · LabSep 21, 9 min read

    A second attempt was worth six points. Deciding when to take one was worth one.

    Lab-001 changed two things at once and reported a seventeen point gap. This varies them independently across five arms, reproduces both original numbers exactly, and finds that the metric Lab-001 called routing quality was measuring agreement with a keyword map.