A software development lifecycle in the era of agentic collaboration. Intelligent delegation for efficient collaboration.
Two markdown files to drive shared context for both humans and agents:
vision.md: Humans need to contribute to the shared, concise, overall vision - as new requirements and R&D require change throughout. Less detail, more level-setting and grounding, what is the problem we’re trying to solve?plan.md: Humans need to contribute to a shared, concise, windowed plan of action - this document should be the new standup board. Short enough to avoid agent context window pollution, detailed enough to execute. “Windowed” means staying relevant to the immediate priorities of the week/month (vertically-scoped incremental delivery). How are we going to implement change this week?
Then the team follows SDLC using agents, following a “code of agentic conduct”, while maintaining the above context docs:
- Leadership can onboard humans and share the vision, stay on target.
- UX can use agents to rapidly prototype and demo visual solutions, with no expectation for production code
- Product can analyse the market, with clear scope, imagine and collaborate, use agents to rapidly prototype with no expectation for production code or UX
- FE can take the above, and breakdown for maintainability, performance (iterate)
- BE can take the above, and do the same
All steps can happen in parallel and influence any other step at any time, as long as key decisions are captured within the context documents for all to see.