Between 10 and 15%. That is the share of the productivity gain that, according to its chief executive, TCS passes on to the client in most cases. Measuring development in hours made sense while a person wrote every line of code. Today some vendors code with coding agents, and their invoice still arrives in hours.
I propose to call it the empty hour: a billed hour that no longer tells you how much work it holds or what was delivered in it. It may be an hour of a person writing code, an hour reviewing what an agent wrote or an hour of waiting while the agent tries again. On the invoice they weigh the same.
The thesis
I am not saying that paying by the hour is always a mistake; I am saying that, when your vendor writes the code with agents, the hour stops measuring what you buy. What protects the buyer is contracting for what is specified: a written scope, with acceptance criteria that can be checked and quality criteria so you do not pay for technical debt, the code that works today and is expensive to change tomorrow. The hour is still useful for work that cannot be specified, such as exploring or researching. What stops being useful is paying by the hour for what can be specified.
The empty hour has two outcomes, and neither informs whoever pays. Either the vendor does not say it uses agents and bills as before, or it bills fewer hours for the same work and your team cannot tell whether it went faster or was reviewed less. The specification tells you what you get. The hard part moves to writing it well, and that work can be hired out too. Your own team already feels it with its tools: coding agents are billed by usage, and a poorly defined request costs more.
It has limits. The sources talk mostly about large vendors, and little about a mid-sized company buying a project. And if the specification is badly written, a fixed scope pays for exactly what was badly written.
What the data says
The price of the hour is already being debated in the market, above all among large vendors.
- Kotak Institutional Equities estimates a 3 to 3.5% hit to the revenue of India’s IT services sector over the next two fiscal years, because clients ask for AI productivity to show up in the price at renewal. Analytics India Magazine reported it in June 2026; Kotak’s report is not publicly available.
- The 10 to 15% figure comes from TCS’s chief executive, quoted by Ted Schadler, an analyst at Forrester, in October 2026. It is a vendor’s statement, with no independent measurement.
- Gartner expects that by 2028 more than half of companies will stop paying for assistive AI and will choose platforms that commit to the result of the workflow (press release, April 2026). The prediction is about AI tools and leaves development services out.
What cuts the other way counts too. Sandeep Poojary, a partner at Deloitte, admits in the same Analytics India Magazine article that nobody has cracked outcome-based pricing yet. In CIO magazine, in June 2026, Jasper Geurts, CTO of Software Improvement Group, warns that paying only for functional outcomes means paying vendors to create technical debt. That is why the specification carries quality criteria as well as functional ones.
What it looks like on the invoice
- This month’s invoice shows the same hours as before the vendor mentioned it works with agents.
- Or it shows fewer hours for the same functionality, and nobody on your team knows whether that means more efficiency or less review.
- The contract talks about profiles and days, and nowhere says what has to happen for a delivery to be accepted.
- The renewal is negotiated on the hourly rate, and the functionality delivered stays out of the conversation.
Why it is understandable
The hour has real advantages for the buyer. It is easy to compare across vendors, procurement understands it and it fits an annual budget. When the scope is unclear, paying hours looks safer than fixing a price for something nobody has written down. It suits the vendor too, because the hour protects its margin when the project gets complicated.
Leadership has its share. For years development has been bought by profiles and days, and few companies have invested in writing specifications a third party can meet. If nobody in-house knows how to write an acceptance criterion, the hour is the only unit you can ask for. Public bodies face the same question when they draft a software development tender with AI.
What you pay for without seeing it
With agents, the link between hours and results breaks. A task that used to take two days can be done in an afternoon, or it can take three days of reviewing what the agent got wrong. The invoice cannot tell the two cases apart, and your company pays for both the same.
Incentives do not help either. A vendor that bills hours earns less if it works faster, so it has reasons not to mention it. And quality that nobody wrote down does not get checked either: technical debt shows up months later, when the code is already paid for and the vendor may be gone.
What to do
Whoever does the work (the technical team that commissions and receives the code):
- Write checkable acceptance criteria for every story in the next request before asking for a quote.
Manel Marcos, cofounder and technical lead at onext, would ask the contract to set criteria that check themselves on every delivery: that it compiles, that all tests pass and that there is one test per requirement and per edge case; that no secrets or credentials enter the code; size limits on what is new (files under 300 lines, functions under 50); that versions and dependencies are not changed without a written decision; and that if the code changes, the documentation changes with it. And a single human rule: no specification goes into development with open questions.
Whoever leads:
- At the next renewal, move at least one request from hours to a fixed scope, with acceptance and quality criteria in writing.
- Ask for the contract to state whether the vendor uses agents and which automated checks each delivery passes.
- Budget the time to write the specification and decide who writes it: your team or someone you hire for that work.
What can be checked
At onext we stopped selling the Quick-start by the hour, for three reasons. When we started working with agents, hours stopped reflecting the work: a task that used to take a week came out in two days, and charging for it by the hour meant charging less for doing it better. The conversations went into negotiating the rate and the number of hours, and what the team would have at the end was left out. With a fixed price you talk about the result: the harness set up and the team working with the flow. And the client did not know what they were buying. With a fixed price and a defined deliverable, the decision is easier and faster.
The empty hour will stay on the invoice as long as nobody writes down what is expected of each delivery. When an agent writes the code, what your company can buy and check is what it put in writing beforehand.
Frequently asked questions
What is the empty hour?
It is an hour billed by a development vendor that no longer tells you how much work it holds or what your company got in return. It appears when the vendor codes with coding agents and keeps charging by the hour.
Why does hourly billing fit worse with a vendor that uses agents?
Because the link between hours and results breaks: a task can be done in an afternoon or take days of reviewing what the agent did, and the invoice charges both cases the same. On top of that, a vendor that bills hours earns less if it works faster.
Should you pay for outcomes?
The sources are cautious. A Deloitte partner admits that nobody has cracked outcome-based pricing yet, and the CTO of Software Improvement Group warns that paying only for functional outcomes means paying vendors to create technical debt. That is why quality criteria should be added to the functional ones.
What does contracting for what is specified mean?
Contracting a written scope with acceptance criteria that can be checked and quality criteria, such as automated tests or how easy the code is to maintain, so that each delivery is accepted on an objective check.
Which quality criteria should go into the contract?
The ones that can be checked automatically on every delivery, such as the tests that pass in your continuous integration environment or static analysis of the code. That way quality is reviewed on every delivery without relying on one person looking at everything.
Who writes the specification?
The buying company, with its technical team, or someone it hires for that work. What matters is budgeting that time, because writing the specification well becomes the hard part of the job.
Sources
- Ajay Rag, “Indian IT is Tearing Up Its Billing Manual, But Unsure What to Charge Next”, Analytics India Magazine, 18 June 2026 (Kotak Institutional Equities estimate and remarks by Sandeep Poojary of Deloitte).
- Ted Schadler, “For Service Providers, AI Giveth And AI Taketh Away”, Forrester, 16 October 2026.
- Gartner, press release on assistive AI and outcome-focused workflows (2 April 2026), in its official distribution on PR Newswire of 9 April 2026, with remarks by Alastair Woolcock.
- Grant Gross, “IT hurtles toward the ‘Great Enterprise Pricing Reset’”, CIO, 16 June 2026.

Bernat López is founder and CEO of onext, an AI boutique. He helps development and product teams work with AI with method —specification before building, automated checks and Spec-Driven Development— and applies to his own company what he proposes: onext runs on its own agentic system.
LinkedIn →