"AI Will Do It in One Evening": Three Estimation Methods Show 300 Hours Is Below the Industry Floor

An author on Habr estimated an extension of a corporate accounting system at 300 hours. The client replied that this was inflated and hinted: "AI will do it for me in one evening." In response, the developer recalculated the same scope with three independent methods and got a result the client did not expect.

Three Methods, One Conclusion

The developer applied three standard approaches to the same scope of work and did not rely on any single one of them alone.

• Bottom-up by user actions — a detailed breakdown of every end-user operation in the system
• Function points with ISBSG benchmarks — the scope is converted into function points and compared against the ISBSG (International Software Benchmarking Standards Group) database of thousands of real projects
• COCOMO II — an algorithmic model from the University of Southern California that accounts for project scale, logic complexity and team characteristics

All three methods produced consistent results. According to Habr, each approach showed that 300 hours is not a ceiling but a lower bound: the figure came out below average industry values.

Why Is 300 Hours Too Little?

The ISBSG database accumulates production data from thousands of real enterprise software projects worldwide. The average productivity of a team building accounting systems is considerably lower than it looks from the outside: function points account not only for writing code but also for requirements analysis, integration with existing modules and regression testing.

"300 is below the floor for the industry," the author states after

running the calculations with three independent methods.

COCOMO II adds coefficients for business-logic complexity and technical debt. For corporate accounting systems, where every change touches adjacent modules, these coefficients raise the final estimate rather than lower it.

Where the "AI in One Evening" Argument Goes Wrong

Modern AI tools do speed up boilerplate code, but they do not eliminate the key stages: auditing the existing architecture, analyzing requirements, and integration and regression testing. These stages make up most of the real workload when extending corporate systems.

The client's mistake is a common one: AI is visible in a quick demo ("a prototype in one evening"), but invisible in everything the prototype lacks — reliability, compatibility and data security. ISBSG and COCOMO II account for these hidden stages explicitly.

What This Means

Three independent methods gave the same answer: 300 hours for extending a corporate accounting system is a realistic lower bound, not an inflated figure. The case shows that a well-grounded estimate built on industry methodology is a workable answer to the "AI will do it faster" argument.

Frequently Asked Questions

What is ISBSG?

ISBSG (International Software Benchmarking Standards Group) is a non-profit organization that collects production data from real software development projects worldwide. Its database lets teams compare a function-point estimate against the industry median and defend it to a client with statistics rather than intuition.

How does COCOMO II differ from bottom-up estimation?

COCOMO II is an algorithmic model that produces a probabilistic estimate based on project scale, team characteristics and technical risks. Bottom-up estimation is a detailed decomposition into specific tasks. When both methods give similar results, it settles disputes about "expert subjectivity" and increases trust in the estimate.