Plan what you can, evolve what you can't
Plan the ordinary work agents can execute, and reserve evolutionary design for the genuinely unknown parts.
Zoom all the way out and the work has a shape:
discovery → plan & hand over → delivery → production → back to discovery
Discovery is where you figure out what problem to solve and prototype a solution. Delivery is where you build it for real. Production is where it meets users and tells you the truth. It’s a loop, not a line — what production teaches feeds the next round of discovery.
[ILLUSTRATION: the loop as a cycle, not a pipeline. Highlight the join between discovery and delivery — the seam — as the riskiest handoff.]
Which revives an old argument. The agile community spent years debating how much design to do up front versus how much to let emerge as you build. Pre-AI, the honest answer was both. Post-AI it’s still both — but the balance shifts toward planning, because the plan is the specified work an agent can actually be handed. The less you plan, the less you can delegate.
The good news is that planning, like coding, just got cheaper — agents are strong at it, so you can plan more, and faster. But cheaper isn’t delegated: agents draft and expand the plan; you own it. The plan is the one thing you don’t hand off blind.
So the recipe is to plan as much as the problem allows. For ordinary, well-understood cases — the bulk of most products — plan them fully against a standard and hand them over. For the genuinely new or hard parts, don’t force a plan you can’t yet trust; that’s where evolutionary design wins, and where discovery earns its keep — you learn the design by building it.
This course covers plan & hand over and delivery. (The discovery course is coming soon.)