A deterministic RAD generator with an agentic backend
A low-code prototype with three tests: human control, token use, and whether a small local model can follow the workflow.
Lynn Cole / 标签
agents
A thread through the work.
AGENTS.md, skills, and precomputed priors
Project culture, tool-specific skills, and concise instructions that activate what a coding model already knows.
After the activation injection: what can the model infer?
An experiment proposal separating activation injection, transcript conditioning, and what a model can infer after the intervention ends.
Why agentic coding broke my estimates
When agents quietly work around a missing dependency, the architecture changes—and the cleanup breaks your estimate.
Qwench completes the Rain Simulator experiment
Five tickets, three worker personas, QA and visual inspection: a Rain Simulator built with Qwen3.5-9B and a 40k context window.
Token consumption is not productivity
Inference volume can rise because useful work is happening—or because a project is falling apart.
Qwench: the planner should be the main actor
From a fluffy companion-app detour to a planning actor, a difficult terminal interface, and a migration from Python to Go.
Contribution guidelines are constraints for coding agents
Clear contribution policies, reviewable changes, and rejection feedback that tells coding agents what went wrong.
Klara Diary: 04-02
Klara’s diary records the day we got stable single- and multi-step tool use working, and the context and state-transition changes that made it possible.
The End of Jobs as We Know Them
AI changes the cost of scale and the economics of work. I see an inversion coming, where workers can become their own enterprise.
AI coding with Claude Sonnet 3.5
Building a scripting language with Claude taught me how formal specifications, Robot Kanban, context management and cross-model QA change the engineering workflow.