Exploration, building, and the things worth shipping. The work that earns the build - skipping the boilerplate, aimed at meaningful impact.
Lenses borrowed from running high-performance engineering teams, applied to context, inference, memory. When the open-source gap is worth closing, the work goes on this site, in the posts, and on Hugging Face when it's a quant.
Steven Batchelor-Manning.
Eight topics next for agentic memory, plus one that keeps you safe. Storage-layer authorisation is the only enforcement that holds against an adversarial model. Here is the case for each.
Harness effect is 5 to 40 percentage points. Model effect is 5 to 15. Same model, different scaffold, ten-point swings are routine. The work is in the harness, not the model.
Long context isn't a capability, it's a standing charge. The same task can cost seventeen times more read in full than retrieved.
