Six months ago, a feature that would take a team two weeks to build can now be prototyped in an afternoon. AI coding assistants, code generation tools, and increasingly capable LLMs are making the act of writing software dramatically cheaper.
This changes everything about how we should think about planning.
The Old Calculus
When building was expensive, planning was how you avoided waste. You planned exhaustively because every wrong line of code was costly. Waterfall made sense when a single engineer cost $150K/year and a rewrite meant months of lost work.
Even in agile teams, sprint planning existed to minimize the cost of building the wrong thing. The implicit formula was: planning time saves building time.
The New Calculus
If building is nearly free, the math flips. Why spend two days planning a feature when you could build three versions and test them with real users in the same timeframe?
Some founders are already operating this way. "Just build it" is their mantra. And for simple features with clear success metrics, this works surprisingly well.
But here's where it gets dangerous.
What Cheap Building Doesn't Fix
AI can build what you ask for. It cannot tell you what to ask for. The harder problems in product development were never about code:
- Which problem should we solve? Building three solutions to the wrong problem is still waste — just faster waste.
- What are the second-order effects? A feature that's easy to build can still be hard to maintain, hard to explain, or hard to undo.
- How does this fit the system? AI can build a module. It struggles to understand how that module changes the dynamics of an entire product.
- What should we NOT build? This is the most valuable question in product development, and it requires judgment, not generation.
The Right Balance
Planning should shift from "how will we build this?" to "should we build this and why?"
The new planning rhythm looks like:
- Problem framing — 80% of planning time. What exactly are we solving? For whom? How will we know it worked?
- Architecture sketching — Light touch. What are the major pieces and how do they connect? Not detailed specs.
- Build + test — Fast. Multiple small bets. Throw away what doesn't work.
- Reflection — What did we learn? What surprised us? This is the new "retro."
The Meta-Shift
The real change isn't that planning matters less. It's that the type of planning that matters has shifted upstream. When building is cheap, thinking is the bottleneck. Strategy is the bottleneck. Understanding your user is the bottleneck.
AI makes it cheaper to build the thing. It doesn't make it cheaper to decide what thing to build.
Spend your time there.