SECTION 01 / OPENING
Notes on harness engineering
Where reliable AI agent capability actually comes from — and how to build it.
In engineering, AI capability does not live in the model alone. It is distributed across tools, verifiers, control flow, and the design of the operating environment. The Harness is a working notebook on that problem — writing from the intersection of agentic AI and architecture, engineering, and construction.
SECTION 02 / TOPICS
Harness Engineering
The tools, verifiers, orchestration, and process design that make agentic work reliable. Not the model — the system around it.
10 articles 02Agent Evaluation
Benchmarking agents on real engineering tasks. What to measure, how to measure it, and why most evals miss what matters.
5 articles 03AI in AEC
Applying agentic AI to architecture, engineering, and construction — where the work is instance-bound, constraint-heavy, and intolerant of generic answers.
9 articlesSECTION 04 / RECENT
- data-engineering Mediation, Not Intermediation Why the 'fix your foundations before AI' message has it backwards: agentic workflows are the way out of legacy data, and governance worth having is co-designed from practice, not committees.
- ai-evaluation Task Worlds and Meta-Harnesses How task worlds, Badiou, Plasticity, and the AEC-Bench meta-harness turn task prose, evidence, review, governance, and repair into runnable machinery.
- workflow-reliability Plausible Answers, Failed Workflows An AEC-Bench release evaluation read as workflow reliability, not prose quality. Chapter by chapter: why a model can produce a plausible answer and still fail the durable record a project has to audit.
- prime-lab Making aec-bench Trainable with Prime Lab How aec-bench and Prime Intellect's Lab turn engineering benchmarks into verifier-backed RL environments, adapter training runs, and inspectable traces.
- autoformalisation Executable Standards Better tools and verifiers are not enough. The next harness boundary is the clause itself — turning standards, briefs, and codes into versioned predicates and replayable certificates.