The bottleneck moved, and most teams missed it
AI did not kill the engineer. It killed the vague ticket. With companies like Snap reporting that a large share of their new code is now AI-generated and shipping with smaller teams, the constraint on software delivery has quietly shifted. The bottleneck is no longer typing speed. It is specification quality.
This is easy to miss because it feels like the opposite should be true. If an agent can produce a working module in minutes, surely we are faster across the board. Sometimes. But speed in the wrong direction is just faster waste. When generation is cheap, the cost of a fuzzy requirement does not go down. It compounds. You ship in an afternoon, then spend a week debugging behavior nobody specified because nobody defined the edge cases.
Why fuzzy specs cost more now, not less
In the old world, a vague ticket was throttled by how long it took to write the code. An engineer would hit the ambiguity, stop, and ask a question before sinking days into the wrong thing. That friction was an accidental quality gate.
Agents remove the friction. They will happily interpret an underspecified request and produce something plausible and confidently wrong. The ambiguity that used to surface during implementation now surfaces in production. The slack that protected you is gone.
So the discipline has to move upstream. The person who can describe a system precisely is now more valuable than the person who can type it fastest.
What good teams are doing differently
The teams getting real leverage out of AI coding share a few habits:
- Specs include explicit non-goals, not just goals. Telling an agent what not to build is often more valuable than telling it what to build, because it bounds the space of plausible-but-wrong outputs.
- Acceptance criteria are written before the prompt, not after the PR. If you cannot describe how you will know the work is correct, you are not ready to generate it.
- Engineers review the spec like they used to review the code. The highest-leverage review now happens before a line is generated, not after.
If your AI output feels inconsistent, the fix usually is not a better model. It is a sharper brief.
This reframes what an engineering role even is. It is shifting from author of code to author of intent. The craft of decomposing a problem, naming the edge cases, and stating the contract precisely was always part of good engineering. AI just made it the part that matters most.
A practical move for this quarter
Pick one habit and make it non-negotiable. The easiest high-impact one: require a short non-goals section on every spec. It forces the hard thinking that a goals list lets you skip, and it gives any agent or engineer a fence to work inside.
Clarity was always a competitive advantage in software. Now it is the competitive advantage, because everything downstream of it has been automated.
At Devdot, we're here to help founders and teams design and build digital products that are built to scale with you, not slow you down. If you're building something and want it built right the first time, get in contact with us today.
The model is rarely your problem. The brief usually is.