Code was never the bottleneck.
AI is great at building software right. It can't tell you whether you're building the right thing. Why verification and validation are different questions, and why the second matters more now.
The short answer
The hard part of software has always been building the right thing, the right way. AI is great at the second half, but it doesn't replace human judgment or the nuances of how real users think, behave and decide. AI makes building cheaper, so validating what to build matters more, not less. Only your customers can answer it.
Code was never the bottleneck. The hard part of software has always been building the right thing, the right way. AI is great at the second half of that sentence. But it doesn’t replace human judgment, or the subtle nuances of how real users think, behave and decide.
Why isn’t shipping more the answer?
AI is now everywhere in development. Google’s 2025 DORA report, based on survey responses from nearly 5,000 technology professionals, found that AI adoption among software development professionals has reached 90%, and it’s now linked to higher delivery throughput, meaning teams are shipping more software.
But shipping more was never the problem. Pendo’s usage data shows that 80% of features in the average software product are rarely or never used. Faster delivery doesn’t fix that. It just gets you to the wrong place sooner. DORA puts it well: AI acts as an amplifier, magnifying the strengths of high-performing organizations and the dysfunctions of struggling ones.
What is the difference between verification and validation?
Verification asks: are we building it right? AI is genuinely useful here. It speeds up delivery and helps confirm the software works as specified.
Validation asks: are we building the right thing? No tool can answer this for you. Only your customers can.
Every new project starts as a hypothesis: a solution that might address a real need. AI can help you build and test that hypothesis faster, but success depends on proving it actually delivers the value you expected to the people you built it for.
What is the uncomfortable truth?
You can build the right thing with bugs and still create value, then improve it based on feedback. You can also build the wrong thing perfectly, with zero bugs, and still fail.
You can also build the wrong thing perfectly, with zero bugs, and still fail.
So treat it as a cycle of continuous learning. The evidence you gather from validation is the real decision point: keep going, pivot to a better hypothesis, or drop it and go solve a problem that matters more.
Why does validation matter more now?
AI makes building cheaper. That makes validating what to build more important, not less. Reading weak signals, noticing what users do versus what they say, deciding when the evidence is strong enough to commit or walk away: that’s still human judgment, and it’s where projects are won or lost.
What I’d do this quarter
- Treat every new project as a hypothesis, and set out to prove it delivers the value you expected to the people you built it for.
- Use AI to build and test that hypothesis faster.
- Watch what users do versus what they say, and read the weak signals.
- Use the evidence from validation to decide: keep going, pivot to a better hypothesis, or drop it and solve a problem that matters more.
Sources: Google Cloud · Google · DORA · Pendo
Written by Luciano Ayres. Edited with AI assistance.