Grounded writing on AI you can verify.
The state of the field, with sources attached: models and agents, knowledge and logic, and the evidence behind the headlines. No hype in either direction.
The Harness: Why the System Around the Model Decides
The same model scores 93% in one tool and 77% in another. The difference is the harness: everything in an agent except the model. This article maps what a harness is made of, why its load-bearing parts must be structures rather than prompts, and why a temporal knowledge graph fits that job.
Read the article →An LLM Is Not Enough
A language model gives every workflow the same thing: a plausible answer. Workflows differ in what an answer must pass before it counts. Where that gap matters most (time), and how working systems close it with explicit, dated facts and a verifier.
Read the article →AI in 2026: What It Can Do, What It Can't, and What Most People Miss
A complete, sourced map of the field in mid-2026: frontier models and agents, the machine learning that quietly runs the economy, the symbolic half most coverage skips, the money, the evidence on work, and the open problems that don't go away by waiting for the next model.
Read the article →