Stack Overflow's 2025 survey found that 51% of professional developers use AI tools daily. On many teams, the number is even higher.
AI tools are no longer experimental. They are part of the standard development workflow.
The question is no longer whether AI changes software development.
The question is how much it changes the economics of software projects.
One framework that helps explain this is the 40-20-40 rule.
Roughly 40% of project effort goes into planning and design. Another 20% goes into coding. The final 40% goes into QA, testing, integration, deployment, and finishing work.
AI has clearly compressed the coding portion.
For many projects, that 20% may now be closer to 8% to 12%.
That is a meaningful improvement.
But the headline often misses an important detail.
If you reduce a 20% activity to 10%, you save roughly 10% of total project effort.
Not 60%.
The remaining 80% of the project still exists.
Planning still requires stakeholder discussions, architecture decisions, requirements gathering, risk analysis, and product tradeoffs.
Testing still requires validating real-world workflows, checking edge cases, reviewing integrations, ensuring security, and confirming production readiness.
Every AI-generated function still needs review.
Every AI-generated test still needs validation.
Every AI-generated architectural suggestion still needs experienced engineers to evaluate whether it will survive real production environments.
The verification burden is becoming one of the most important factors in AI-assisted software development.
Recent surveys suggest that 45% of developers find debugging AI-generated code more time-consuming than debugging code they wrote themselves.
Another survey found that 96% of developers do not fully trust AI-generated output without human review.
In other words, some of the time saved during generation gets spent during verification.
The net productivity gain is still positive.
But it is rarely as large as the most optimistic headlines suggest.
This is why many agencies and engineering teams continue to price projects based on the full software delivery lifecycle rather than just the coding portion.
The coding got faster.
The engineering did not disappear.
FoundersBar published a full breakdown of how the 40-20-40 rule applies in the age of AI, including what agencies, clients, and software teams should realistically expect from AI-assisted development.
👉 Read the full article: foundersbar
What's your experience so far? Is the review overhead offsetting some of the gains from AI-assisted coding, or are you seeing meaningful net improvements across the entire development lifecycle?
Top comments (0)