Thoughts on AI, engineering, and building
Practical writing on AI agents, automation, and the engineering behind them, for the people who run the business.
Harnesses & Toolingdeep-dive
AI coding tools make it cheaper to understand old systems, but proving a migration preserves behavior is still the hard part. What the evidence shows.
· 4 min read
Harnesses & Toolingexplainer
Test the retrieval step of a RAG system on its own: recall@k, precision@k, MRR and nDCG, a small worked example, and how to build a golden question set.
· 4 min read
Harnesses & Toolingdeep-dive
How the Model Context Protocol's 2025-11-25 revision handles authorization and long-running work, and what that means for CRM and ERP integrations.
· 4 min read
Harnesses & Toolingexplainer
A production prompt is a versioned specification with real examples, a schema and an evaluation set. What the vendor docs say to build before you tune wording.
· 3 min read
Harnesses & Toolingexplainer
An AI agent that fails rarely says why. Tracing records each model call, tool call and handoff so you can find out. Here is what to capture and what to watch.
· 3 min read
Harnesses & Toolingnews-analysis
Microsoft Agent Framework 1.0 and OpenAI's Agents SDK update landed within two weeks. What they share, where they differ, and what to ask before you commit.
· 3 min read
Harnesses & Toolingdeep-dive
Models now plan spreadsheet work and read business documents. Why that raises, not lowers, the value of clean pipelines and anomaly detection.
· 4 min read
Harnesses & Toolingbrief
Embeddings turn text into numbers so software can search by meaning. What they are, where they are used, and what to ask a vendor about them.
· 2 min read
Harnesses & Toolingnews-analysis
On December 9 MCP, AGENTS.md and goose moved under a Linux Foundation fund. Why the harness around a model now matters as much as the model.
· 3 min read
Harnesses & Toolingexplainer
Three ways to make a language model work with your business knowledge, what each one changes, and a simple order in which to try them.
· 3 min read
Harnesses & Toolingexplainer
Retrieval-augmented generation lets a model answer from your files instead of memory. What it is, where it fails, and what to ask before you build it.
· 4 min read
Harnesses & Toolingnews-analysis
The MCP Registry preview and Google's AP2 payments protocol show agent interoperability maturing in layers. What each does and where the gaps remain.
· 3 min read