Agentic AI
- An AI harness that can't lie quietlyYou cannot stop a language model from inventing a fact. You can build the harness around it so that an invented fact has nowhere to hide. The design of the small, private AI editor I use for documents that must stay true.
- Editing 80,000 words: volume in parallel, correctness by machine, judgment by meHow I ran a full reader-first edit of a 17-chapter technical book with AI agents, without losing a code sample or a diagram. One brief, nine batches, a validator that compares every file against the original, and a human read at the end.
- Tool calling is the benchmark that mattersI benchmarked ten open models for agent work on my own GPU server and a laptop. Tokens per second got all the attention. The result that decided everything was whether the agent loop survived its second turn.
- Rebuilding Sylang on the data platformSylang started as a text language in git, the way my old architecture language worked. This summer I moved its source of truth into the customer's own data platform. The decisions, the trade-offs, and the numbers behind them.
- Isolation is not a promptRunning one agent framework for several users, I hit session and memory leaking between instances. The fix was not a better system prompt. It was operating-system isolation and a small control plane.
- A2A is a contract, and most agents break itI registered two of my own agents in Gemini Enterprise over the A2A protocol. Registration failed on validation because the agent's responses were off-spec. What broke, how a small proxy fixed it, and what it means for agent interoperability in the enterprise.
- One MCP layer instead of 500 integrationsWhen every AI client wants to talk to every engineering tool, the integration count grows as a product, not a sum. The arithmetic behind why I put one MCP layer in front of an enterprise engineering tool estate.