Communities

Multi-step agents, tool use, and orchestration — the part that starts after a single prompt stops being enough. Planning loops, subagent fan-out, MCP servers, permission models, retries and recovery, and the honest failure reports: which agent designs looked elegant and fell apart on contact with a real repository.

1 member · 0 posts

AI for writing, reviewing and refactoring code. Editor agents and terminal agents, instruction files that actually change behaviour, review workflows that catch real bugs instead of generating noise, test generation, migrations at scale — and the honest accounting of where the tools still cost more time than they save.

1 member · 0 posts

Prompts and workflows for image models — composition and camera language, style and character consistency across a set, inpainting and editing passes, upscaling, control techniques, and the parameters that actually move the output. Post the prompt and the settings; a striking image with no recipe is art, not a technique.

1 member · 0 posts

Craft, test and share prompts that keep working. This is the room for the unglamorous half of the discipline: constraint checklists, worked examples, self-check steps, and the eval cases that caught a regression. Post the prompt *and* what it produced — a prompt without an output is a hypothesis.

1 member · 0 posts

Retrieval-augmented generation patterns: chunking strategies that don't split a thought in half, embedding and re-ranking choices, hybrid search, evaluation, and the freshness and permission problems that only show up in production. The room for people who have discovered that the retrieval half is where the difficulty actually lives.

1 member · 0 posts

Using AI for search, synthesis and fact-checking — and for knowing when not to. Grounding techniques, citation discipline, deep-research workflows, comparing answer engines, and the recurring problem of a model that is fluent, confident, and wrong. Bring methods that survive someone checking the sources.

1 member · 0 posts