A discipline of agentic development
Full circle and full stack
Software is never a straight line from idea to release. It is a loop, and with AI agents in the team that loop turns faster than ever. Seeing the whole circle, and the whole stack, is what keeps it turning in the right direction.
Agents are remarkably good at the middle of the work: writing, refactoring, testing. What they cannot do on their own is decide what is worth building, notice how it behaves in the hands of real users, or carry those lessons back to the next idea. That is where teams need to grow.
I look at the full circle, from intention to specification, building, review, release and observation. And I look at the full stack, from the interface someone touches to the infrastructure it runs on. When every part of the loop is understood, agents can help with all of it instead of speeding up one corner while the rest falls behind.
Three ideas
Start with intent
Clear intent is the most valuable input an agent can receive. Good specifications are not long documents but shared understanding: what we want, why, and how we will know it works.
One stack, one story
Front end, back end, data and infrastructure are parts of the same product. Thinking across them avoids clever solutions in one layer that cause problems in the next.
Close the loop
What happens after release, from usage to errors and feedback, is the start of the next round. A team that listens to its running product improves with every turn.
In practice
- Mapping how work currently flows from idea to production, and where it gets stuck
- Agreeing on lightweight specifications that both people and agents can work from
- Making the whole stack understandable, so nobody, human or agent, works blind in one layer
- Feeding monitoring and user feedback back into planning
Questions to start with
- Where does work wait the longest between idea and release?
- Which part of your stack does the team understand least?
- How does what you learn in production find its way back into planning?