Who Captures the Productivity Gains from AI?

💡 Summary: Key Takeaways

• AI can reduce your share of the value without eliminating the work itself. A salaried employee may produce more for the same pay, while an hourly provider may bill fewer hours for the same deliverable.
• The person using AI does not automatically capture the productivity gain. With salaried work, the first gain may go to the employer; with actual-hour billing, it may go to the client.
• There is no easy pricing escape. You cannot bill for hours you did not work, competition can compress fixed fees, and productisation works only when it creates recurring value and a defensible difference.

“Will AI take your job?” is too narrow a question for understanding AI's effect on work. AI does not always eliminate the work. The client may still want the same deliverable, and the employee or provider may continue producing it. What changes is how the productivity gain created by AI is divided between them.

Suppose an agency performs a specific client task that used to take three hours. Let the hourly rate be R. Under the old arrangement, the agency completed the work, billed three hours, and earned 3R.

AI reduces the task to one hour. If the agency bills the client for the actual time worked, the invoice becomes R. Now suppose the agency uses the remaining two hours to complete two similar jobs for two more clients. At the end of the same three-hour period, the agency has served three clients but still billed only three hours in total.

The amount of work delivered has tripled. Revenue has not changed.

AI has clearly increased productivity. The agency can serve more clients. Clients receive the result faster and pay less for the same work. But the agency's revenue per working hour has not increased. Finding three clients, explaining the process three times, holding more meetings, and collecting three payments may also require more effort than managing one client.

The important question is therefore not whether AI saves time. It is who captures the economic value of the time saved.

More work does not mean more revenue

When actual working time is what gets billed, the equation is simple:

Revenue = hourly rate × actual hours billed

If AI reduces the time required for one job from three hours to one while the hourly rate remains unchanged, revenue from that job falls from 3R to R.

If the agency has enough clients and enough work, it can fill the remaining two hours with work for two other clients. Revenue after three hours returns to 3R. But the agency has delivered three jobs instead of one, managed three client relationships, and taken on more volume merely to preserve the old revenue. The additional jobs did not create additional income. They replaced the billable hours lost on each job.

If the agency does not have enough clients or work, the outcome is even clearer. It finishes the job in one hour, bills R, and cannot bill the remaining two hours. The same client job now produces less revenue.

AI does not solve the income problem for someone who already had empty hours in their calendar either. Capacity was not the constraint; demand was. Finishing the available work sooner reduces billable time and revenue while increasing unused time.

For a provider working at a fixed hourly rate, the direct income effect therefore has two possible forms. With enough demand, they perform more work to preserve their old revenue. Without enough demand, their revenue falls. More clients and more deliverables are not, by themselves, a financial gain for the provider.

A salaried worker may keep their income while losing their share of the gain

Now imagine that the same work is performed by an employee. The employee earns a monthly salary of M. Before AI, they completed one unit of work in three hours. With AI, they complete three units of work in the same time. Their salary remains M.

The employee's monthly income has not suddenly fallen to one third. But the company now receives three times as much work capacity for the same wage. If that capacity becomes more sales, shorter queues, or a lower unit cost, the first economic gain goes to the employer.

The employee can share in that gain. A pay rise, performance bonus, profit sharing, shorter working hours, lower workload, or stronger job security can all distribute part of the productivity benefit back to the employee. None of these follows automatically from the use of AI. If the company simply raises output targets while leaving pay unchanged, the employee produces more without receiving additional economic compensation for the increase.

Same work, different contract

A developer, designer, or SEO specialist can perform the same work as a salaried employee, an hourly provider, or a business charging a fixed price for a deliverable. The AI tool and the speed gain may be identical, while the first recipient of the economic benefit changes with the contract.

This mechanism is not limited to consulting. A developer may prepare a first implementation, test drafts, or technical research faster. A designer may produce more options and reduce repetitive editing. An SEO specialist may complete a site audit, keyword clusters, or content briefs in less time.

In none of these examples does the first output mean the work is finished. Software still needs review, a design still needs to work in context, and SEO recommendations still need to fit the brand and search intent. My article on why AI made code cheap but did not make verification cheap explains that boundary through software.

Even so, some preparation, production, and repetitive work genuinely does become faster. The same speed gain expands company capacity when the worker is salaried, but can reduce the invoice, and therefore income, for a service provider selling their time. Producing productivity and turning that productivity into income for the person using AI are two different things.

Time measures input, not value

With hourly billing, the unit on the invoice is not the defined work or the business outcome. It is the time the provider spent on the work. When AI reduces that time, the lower invoice does not mean the provider has been treated unfairly. Under the written or verbal agreement, the price is calculated from the time spent.

If a client requested a landing page with defined requirements and the software agency delivered it against the acceptance criteria, the client bought a deliverable before it bought a business outcome. The agency can be assessed on whether the page took three weeks or three days, met the technical requirements, and worked correctly.

The conversion rate does not automatically become the agency's measure of success or failure. It belongs to the agency only if the agency explicitly took responsibility for conversion optimisation and had sufficient authority over the offer, traffic, copy, pricing, and experiment design that influence the result.

This distinction does not mean “stop pricing time and charge for conversion.” Before leaving hourly billing, the provider must establish whether the client is buying defined work, access, a reusable system, or a genuine business outcome. Outcome pricing makes sense only when the outcome is part of the contract, within the provider's control, and credibly attributable to their contribution.

Hourly pricing is not always wrong

If you do not yet know the cause of a failure, you may not know how long it will take to resolve. A fixed fee can create serious risk when investigating a legacy software system, discovering the scope of a legal matter, or working on a project that keeps changing because of client decisions.

In these cases, time can be a reasonable way to share uncertainty. The client pays for the effort that actually occurred. The provider does not carry every scope risk that could not be seen at the beginning.

The decision is not as simple as “hourly pricing is bad; outcome pricing is good.” The pricing unit must fit the uncertainty of the work, what the client is buying, and the risks each party can control.

Charging the client more is not the answer

This diagnosis does not mean, “I earn less because of AI, so I should charge the client more.” If the client bought actual working time and the task took one hour, billing three hours is a false statement. The provider's AI subscription does not automatically create a right to add more to the invoice either.

The fall from 3R to R is therefore not an injustice done to you by the client. It is the direct consequence of hourly billing. You cannot correct it by adding two hours you did not work.

Renaming the price is not enough

Moving from hourly billing to a fixed deliverable price can initially leave the productivity gain with the provider. The agency now completes in one hour the work it previously performed and priced at 3R, while the client still pays the agreed 3R for the deliverable.

That does not mean the agency can preserve 3R indefinitely. Another agency that can complete the same work in one hour with AI may offer comparable quality and trust at a lower price. To the extent that clients can compare offers and the services are similar, competition can pass part of the productivity gain back to clients through lower prices.

The price does not have to fall all the way to R. Trust, expertise, quality assurance, speed, institutional memory, integration knowledge, and responsibility for delivery risk can differentiate one agency from another. But if the old 3R price depended only on the work historically taking three hours, it becomes harder to defend.

A retainer is not a solution by itself either. If the client is really pre-purchasing a number of hours, the same equation continues under another name. Outcome-based pricing makes sense only when the provider genuinely owns and controls the outcome.

AI can therefore compress an agency's income without taking away its work. The same client still wants the same deliverable, and the agency still produces it. But the invoice per job may fall while the agency also pays for AI subscriptions, verification, and the cost of learning new tools.

Change the unit in which you sell labour

If the provider does not want to accept the pressure from 3R towards R, preserving the old hour count is not the answer. The provider has to reconsider what it sells. The aim is not to eliminate labour, but to reduce the linear link between revenue and the hours spent on each client.

Among these options, a productised service or a genuine subscription is the closest structural alternative. The provider can reuse the same method, software, workflow, or operating system across multiple clients. The client buys access to a defined service, a consistent standard, or a continuously operating capability rather than individual hours.

A subscription works only when the need recurs and the value continues. Splitting a one-off deliverable across monthly payments does not create a new business model. A productised service also faces the same pricing pressure if it is easy to copy.

Productisation is therefore not a guaranteed escape. It is an attempt to loosen the linear connection between time and revenue. A durable difference requires more than a reusable system. It may also require trust, distribution, expertise, data, integration knowledge, or responsibility for an ongoing operation.

This is not a prediction that prices will fall at the same speed or to the same level in every sector. It is a mechanism that can emerge when competition is strong, services are comparable, and the productivity advantage becomes widespread.

AI may not take your job. But when it makes the same work possible with fewer human hours, the productivity gain does not automatically go to you. Employees need an explicit pay or working arrangement that shares it. Service providers need a defensible business model that does not tie revenue directly to the hours spent on each client.