AI Agents · A way to organize delivery
Work with AI as a teammate¶
Clear outcomes. Accountable decisions. Verified results. Keep useful knowledge where the customer can own and reuse it.
One assignment, one current plan¶
AI Agents provides a shared operating model, reusable profiles and a local toolkit. People choose goals and accept consequential decisions. AI delivers bounded increments using the models, tools and permissions available in its host.
-
Frame the outcome
Define value, acceptance and who can resolve open questions.
-
Deliver an increment
Keep the current plan in the assigned issue and evidence in a linked PR.
-
Review the result
Distinguish a decision, checked code, accepted work and deployed behavior.
-
Preserve what matters
Record significant decisions, update knowledge and link iteration outcomes.
Choose your starting point¶
For planned work that needs approval, review the plan locally with Plannotator: comment, revise, approve the current version, then Build with its own model and supported thinking variant. Keep the receipt in the assignment's existing planning home.
| Your situation | Start here | Keep the plan in |
|---|---|---|
| Customer or team delivery on GitHub | Customer workspace starter | The assigned issue |
| Larger local work without a board assignment | Local work records | One .ai/work/ file |
| A small, clear fix | Quick start | The current conversation |
| Personal knowledge work | Second Brain | Your existing private PARA vault |
A small toolkit behind the workflow¶
23 agents · 48 skills · Python 3.11+ · Standard-library runtime
Jörg selects relevant expertise; runtime profiles bring the catalog to OpenCode or Copilot. The local dashboard manages Markdown plans and feedback. Your host supplies execution and any delegation.
This website uses MkDocs. The Docusaurus guide publishes customer-owned workspace sources in an existing Docusaurus site. Editable Markdown and compiled output stay separate.