If AI helps write code faster, should clients still pay the same number of hours? The honest answer is: sometimes yes, sometimes no — but the work needs to be verifiable. The arrival of AI did not make billing simpler. It made the number on the invoice harder to read.
Hourly billing has always rested on a quiet assumption: that an hour of logged time roughly maps to an hour of effort. AI loosens that link. When part of the work is generated in seconds and part still takes a senior engineer a full day, "40 hours" can describe wildly different things — and the client has no way to see which.
Why the answer is not simple
The instinct is tempting: AI writes code faster, so development should cost less. But software development was never pure typing. Writing the code is one slice of the job, sitting between understanding the problem, making design decisions, and proving the result works in production.
AI accelerates the typing slice. It barely touches the slices around it. So whether a task should cost less depends entirely on how much of it was typing versus thinking — and that ratio is invisible from an invoice.
What AI can reduce
On the right kind of work, the time savings are real and worth passing on:
- Boilerplate — scaffolding and wiring that follow a known shape
- Repetitive UI work — forms and components modelled on existing ones
- Simple code generation — straightforward, well-trodden logic
- Test scaffolding — the skeleton of a test suite
- Documentation — first drafts of comments and docs
For tasks dominated by this kind of work, it is fair to expect AI to shorten the hours.
What AI does not remove
The expensive parts of software remain stubbornly human:
- Architecture — how the pieces fit and scale
- Business logic — the rules that make the product yours
- Debugging — finding why something does not work
- Testing — proving it works, and keeps working
- Security — getting auth, data, and payments exactly right
- Product decisions — choosing what to build and what to cut
- Responsibility for production code — owning what ships, AI-written or not
AI can draft the code. It cannot take responsibility for it. That responsibility is most of what you are paying a real engineering team for.
Why invoices become harder to evaluate
An invoice reports hours. It does not report the mix inside them. The same 30-hour line could be mostly AI generation with light review, or mostly human debugging of code AI got subtly wrong. Those are very different amounts of skilled effort — and they look identical on the bill.
This is the core problem: AI changed the composition of an hour without changing how the hour is reported. The hidden review and rework that AI creates — what we call the verification tax — lands inside the same number, unlabelled.
Why "AI was used" is not enough
Some clients hear "we used AI" and expect an automatic discount. Some teams use it to justify the same rate for faster output. Both are missing the point. "AI was used" is not a price — it is a prompt to explain effort.
Using AI does not automatically mean a task should be cheaper. It does mean the effort behind it should be explainable: what the tool did, what a person did, and why the result took the time it took. A team that can answer that is billing transparently, whatever the number.
What clients should ask from AI-assisted teams
You do not need to be technical to ask the questions that make effort legible:
- Was AI used on this work at all?
- What part was AI-assisted versus written by hand?
- What required human engineering — design, debugging, hard decisions?
- What was reviewed and tested before it shipped?
- Which pull request shows the work behind these hours?
These are not accusations. They are the normal questions of someone who wants billing to be explainable — and a good team will welcome them.
Why pull requests become key billing evidence
That last question is the important one. A pull request is the artifact that survives the AI ambiguity: it shows what actually changed, how broad it was, whether it touched risky areas, and whether it came with tests and review — regardless of how the code was produced.
Tie the hours to the pull request they delivered and the abstract becomes concrete. You stop asking "is 30 hours fair?" in the air and start asking "is 30 hours fair for this change?" — a question that has an answer. (More on this in Pull requests are the new invoice evidence.)
How InsightPM compares logged time with manual and AI-assisted estimates
InsightPM is built for exactly this ambiguity. It connects Jira and GitHub, and for every pull request it produces two independent estimates of expected effort — one assuming manual work, one assuming active AI assistance — then compares both against the hours actually logged.
That gives a non-technical buyer a fair frame for each invoice line: if the logged time sits near the AI-assisted estimate, the team is capturing the speed; if it sits near or above the manual estimate, that is worth a conversation. It does not declare anyone wrong — it makes the effort explainable, in plain English. See why InsightPM exists, or the related read on the verification tax of AI coding.
FAQ
If a team uses AI, should the invoice be cheaper?
Not automatically. AI can reduce time on boilerplate and routine work, but it does not remove architecture, debugging, testing, or responsibility for production code. The right question is whether the logged effort is explainable for the work actually delivered.
Why does AI make hourly billing harder to interpret?
Because the client sees a number of hours, not the mix behind it. The same hour can be AI generation, human review, debugging, or rewriting. Without delivery evidence, you cannot tell which, so the hour stops being a reliable proxy for effort.
Is hourly billing still fair in an AI-assisted workflow?
It can be, as long as the work is verifiable. Time and material billing is fine when the client can see what was delivered and confirm the hours are plausible for the scope. The problem is not the model — it is the lack of evidence.
What should clients ask an AI-assisted team?
Was AI used, and where; what part was AI-assisted versus human-engineered; what was reviewed and tested; and which pull request shows the work. The aim is not to interrogate, but to make effort explainable.
Does "we used AI" justify a higher or lower price?
Neither on its own. "AI was used" is not a price signal — it is a prompt to explain effort. Some AI-assisted work is genuinely cheaper; some shifts cost into review and debugging. Delivery evidence settles which case applies.
How do pull requests help with billing?
A pull request shows what actually changed: scope, files touched, tests, and review. Linked to the hours logged against it, a PR turns a vague invoice line into something you can judge for plausibility without reading code.
Does InsightPM tell me if I am being overbilled?
It gives a signal, not a verdict. InsightPM compares logged hours against independent manual and AI-assisted effort estimates for each pull request and flags where they diverge, so you can ask an informed question rather than guess.