The Productivity Mirage: Introducing Micro SDD
Measuring productivity in lines of code was always a mistake. With AI, that mistake accelerates. Micro SDD is the answer: minimal specs, continuous learning, real value.
Originally published at alexandersoto.net
Speed without direction
The AI productivity spike is real. It’s also an illusion.
We’ve always been wrong to measure software development productivity in lines of code. But we kept doing it, and for a long time the damage was contained — writing code was slow enough that the ratio of “thinking” to “typing” stayed reasonable.
AI changed that ratio. Now you can generate thousands of lines in seconds. And with that, the old mistake accelerated. Vibe coding — prompting the AI and seeing what happens — reduced software engineering to a generation act. You write a prompt, the AI writes the code, and if it looks like it works, you ship it.
The real problem was never code generation speed. It was always about understanding the problem deeply and maximizing the value delivered to people. Speed without direction is just drift. Fast drift.
SDD: a step in the right direction, but not enough
Spec-Driven Development (SDD) emerged as a response to vibe coding. The idea: before letting AI generate code, you write a formal specification that describes what needs to be built. The AI then implements against the spec, not against a vague prompt.
This is better. The spec introduces discipline, creates a shared understanding, and gives AI real context instead of assumptions.
But there’s a problem: specs can be too vague and too large. A 200-line spec that describes an entire feature doesn’t eliminate ambiguity — it just pushes it a level up. You still end up building the wrong thing, just faster and with more scaffolding around it.
The core issue is spec size. The difference between rigidity and adaptation. Between presuming the solution and listening to what the problem actually demands. Micro SDD is a revision that takes that tension seriously.
Micro SDD
Purpose Statement
Micro SDD exists to improve people’s lives, evolving toward the right solution step by step, without wasting effort or taking unnecessary risks.
Principles
Principle 0 — Feedback as Fuel: Learning is the most valuable asset. Collecting, measuring and assimilating constant and frequent feedback is the fuel of value generation, always in the direction of the outcome. Its source and its end are the people around it.
Principle 1 — Spec Definition: A spec is small enough to be reviewed easily, and large enough to produce verifiable value in production.
Principle 2 — Spec Validation: Every spec is validated with automated tests, and its design emerges through a disciplined application of TDD.
Principle 3 — Spec Size: The spec is small because the best solution is discovered, not presumed — and it doesn’t need to be estimated either. Each spec reveals information that improves the next one.
Principle 4 — Mandatory Human Understanding: Every spec must be reviewed and understood by humans. A black box only leads to darkness — the agility and longevity of the solution depend on understanding it at the same level as having built it.
Micro SDD is a response born from a pragmatic lean/agile mindset. I’m convinced that, properly understood, it’s of vital importance for the industry. And that’s why it’s “micro” — because a small difference in the initial angle can completely divert a ship, whereas continuous learning can guide you in a world of uncertainty. Speed doesn’t guarantee the destination.