Will AI make hourly billing obsolete? The debate affecting freelancers

AI can speed up a task without increasing the income of the person doing it. For a freelancer, three questions remain separate: how much time is saved, what quality is delivered, and who benefits from the saving? A productivity study does not automatically answer the third.

The NBER working paper Generative AI at Work, issued in April 2023 and revised in November 2023, reports an average 14% increase in issues resolved per hour across 5,179 customer support agents. Effects differ with experience. This is customer support research, not research on freelance quotes: applying its percentage to your own business would be unjustified.

Conversely, METR’s randomized experiment published on July 10, 2025 found that tasks took 19% longer with the permitted AI tools, among 16 experienced developers completing 246 tasks in familiar open-source projects. That result concerns early-2025 tools and conditions; it does not show that every AI tool slows every developer today.

More recently, METR’s May 11, 2026 survey distinguishes work speed from work value among 349 technical workers. Responses are self-reported, from a convenience sample, with explicitly discussed biases. None of these sources establishes that hourly billing is disappearing or that freelance earnings are generally rising. The useful lesson is to measure your own delivery cycle rather than turn a feeling of speed into a pricing assumption.

If your agreement charges for actual time worked, do not invoice eight hours because the task used to take eight hours before AI. Instead, track scoping, preparation, tool use, checking, corrections and communication under the agreed rules. An automated process running in the background does not, by itself, amount to billable human attendance.

Quality control is not an embarrassing extra: it is part of the service. Drafting landing-page copy may be quick, but checking claims, adapting the tone, testing the form and checking links still matter. Equally, personal experiments or learning a new tool do not automatically become client time.

Before starting, clarify:

• which activities are billable and which remain your responsibility;
• the budget cap and when additional authorization will be needed;
• which data may enter the tool and any client restrictions;
• the activity record that will make the invoice understandable.

If you want to change your rate or move to a fixed fee, propose a new agreement for future work. Do not quietly change the rules halfway through an assignment. This is commercial discipline, not universal legal advice: applicable obligations depend on the agreement and circumstances and may require professional advice.

Keep hourly pricing when the scope is still uncertain. Diagnosing a poorly documented system, unpredictable maintenance or exploratory consulting means the client is buying intervention capacity. A spending cap and progress reviews limit surprises. AI may reduce execution time without removing uncertainty.

Choose a fixed fee when the deliverable and acceptance criteria are clear. You sell a defined output rather than a presumed number of hours. Efficiency may improve your margin; substantial rework may reduce it. The client gains a known price while you take on some execution risk. Keep an internal time estimate even when time is not the unit you sell.

Consider value-based pricing when the benefit is understandable and defensible. This does not mean taking an arbitrary share of hoped-for revenue. Identify the starting situation, a useful improvement and the factors outside your control. For an agency, a better-designed page does not guarantee sales: traffic, the offer and sales follow-up also matter.

A hybrid can be more cautious: paid hourly discovery, fixed-fee implementation, then maintenance with defined scope and capacity. Before packaging a service, the guide to freelance services for small businesses helps you start with the client problem rather than the tool.

A poorly scoped fixed fee can turn AI savings into unlimited feedback rounds. Specify what will be handed over, in which format, with which acceptance criteria and which client dependencies. “A website” is not a workable scope; named pages, a tested form and approved content provide a much better starting point.

Separate correcting an error, which is necessary to satisfy the agreed scope, from changing the request. Set the number of included feedback rounds, how feedback should be consolidated and how additional requests are handled. A revision limit does not replace the requirement to deliver correctly what you promised.

Use a checklist suited to your work: sources and accuracy for content; functionality, security and accessibility checks for a website; duplicates and consistency for a data file. Do not submit an output simply because the model sounds confident. Name the person responsible for final approval and retain important decisions.

Also define access handover, expected source files and any maintenance. Confidentiality conditions and AI-use restrictions deserve discussion before data is shared—not after the client objects.

Here is a fictional calculation, not a client result. A project previously estimated at eight hours is sold at €75 an hour: €600. With AI assistance, it requires three hours of production and two hours of checking and communication, or five actual hours. At the same hourly rate, the invoice is €375. Time savings therefore do not automatically increase revenue.

If a €600 fixed fee had been agreed for the same deliverable, those five hours would yield €120 of revenue per project hour. With €30 of tool costs allocated to the assignment, the calculation becomes (€600 − €30) / 5 = €114 per hour, before other expenses, contributions and taxes. That is not take-home income. If three additional hours of rework occur, the ratio falls to €71.25. A fixed fee offers less protection than it appears to when revisions expand.

To test your offer, log production, checking, coordination and rework separately. Allocate tool costs without counting the same subscription twice. Then compare sufficiently similar projects with comparable quality and scope. One easy delivery is not a sound reason to cut every future estimate.

Include sales and administrative time in your wider assessment of the business. A freed hour that you do not sell adds no revenue; it may still enable rest, training or better quality. The article on why AI does not automatically make you money develops this distinction between producing quickly and building a viable offer.

Do not defend your price by pretending AI changes nothing. Explain what it accelerates, what you check and what the client is buying. With a fixed fee, the discussion concerns deliverables, quality, deadlines and limits. With hourly billing, it concerns actual work and the agreed tracking rules.

One possible wording to adapt: “AI assistance speeds up certain steps. This proposal covers scoping, execution and validation of the listed deliverables, with the specified feedback rounds included. Out-of-scope requests will require a separate agreement.” This is not a ready-to-sign legal clause.

If the client asks for a reduction, explore a simpler scope, a less demanding deadline or fewer variants instead of hiding hours. If AI use is prohibited, check that your method and estimate still work. Do not introduce the tool secretly against agreed conditions.

Run a pilot on a repeatable service rather than changing every contract:

• fix the scope, price and acceptance criteria;
• track all hours and rework;
• compare the contribution after costs with your internal threshold;
• collect objections raised at quotation and delivery;
• keep, adjust or drop the model according to observed results.

For an agency restructuring its offer, SprintLead’s agency page can be a product entry point for prospecting, not proof that the new fee will sell. Qualify the need and budget first: a better billing model cannot rescue an offer without a buyer.