AI is changing how businesses make money. Who gets the wealth?

PhotoCraft is an open-source, Photoshop-style editor with a familiar interface and public source code: https://github.com/storytold/photocraft. It is part of ArtCraft’s creative tools, whose AI-assisted development was covered in October. The project’s own site still labels it early alpha, not a finished replacement for Adobe’s suite.[10][11][12]

Adobe, meanwhile, reported record revenue of $6.76 billion in its third fiscal quarter of 2026, up 13% year over year.[1] Those results cover the quarter ended August 28, before ArtCraft’s reported October 3 release. This contrast is not evidence that the clone has already failed to affect Adobe; the reporting periods do not allow that conclusion.[1][12]

A business can be growing today while its future pricing model comes under pressure. AI can affect both how cheaply competitors build software and how many people customers need to operate it.

Those possibilities are why I find the business-model question more useful than another debate about whether AI is good or bad.

AI could create wealth, erode existing sources of profit and make some people’s working lives less secure at the same time. There is no contradiction. What matters is where the value comes from, who can charge for it and what happens to the people whose work changes.

The problem with selling access to human users

Adobe’s Creative Cloud team offerings include individual licences. Its documentation describes a Photoshop team licence as giving one person the right to use the application.[6]

Now imagine a customer delivering the same work, at comparable quality, with two or three employees assisted by AI instead of ten. If each person needs one licence, the customer could go from ten paid seats to two or three. At unchanged prices and without additional charges, the vendor would earn less from that account even though the tool helps the customer maintain output.

This is an illustrative scenario, not a measured productivity gain or a claim that every team can shrink to three people. Demand could grow, teams could retain staff, and quality or review requirements could prevent such a reduction. But even alongside the reported 13% revenue growth, the scenario exposes a potential weakness in charging per human user.

For the customer, the saving could increase profit, fund new projects or support lower prices. For employees no longer needed in that workflow, the issue is whether they can move into useful, adequately paid work. Pricing changes alone do not solve that transition.

Adobe already combines subscriptions with generative credits and offers additional credit plans.[2][5] The challenge is how to keep earning as the relationship between human headcount and production changes.

The same thought experiment applies to any business whose revenue depends on the number of people using a tool or the hours required to deliver a service. If AI reduces the labour needed, selling more labour becomes a less comfortable growth strategy.

New ways to charge, and new things to sell

Changing a price unit is not the same as changing a business model. Billing by the token will not rescue a product customers no longer need.

I see several possible directions, each with its own trade-off.

Charge for consumption

A supplier can charge for the resources a customer uses rather than the number of people involved. Snowflake uses consumption-based pricing across compute, storage and data transfer.[3] Datadog’s pricing includes units such as monitored hosts and ingested log volumes, depending on the product.[4]

These are useful reference points, not proof that consumption pricing always wins. They also predate the current generative-AI wave. AI makes this approach relevant to more workflows; it did not invent it.

For an AI product, consumption pricing can connect revenue to activity and help cover variable operating costs. But it can also charge customers for failed attempts. More generated content does not necessarily mean more useful work.

Charge for a defined outcome

Intercom’s Fin illustrates a different approach: its published pricing includes charges for outcomes such as resolutions and configured procedure handoffs.[9]

The definition matters. A resolution can be confirmed by the customer or assumed when the customer leaves without requesting further assistance. Intercom also describes conditions under which a charge is reversed if the customer returns for more help.[9]

That is more specific than saying “we charge for success.” The contract has to define success, handle disputes and identify who carries the cost of mistakes.

Outcome pricing is easier to imagine for a verifiable completed task than for a vague promise to “grow your business.”

Sell an operated service, not another tool

An AI-enabled business could take responsibility for a workflow: preparing a reporting pack, maintaining a product catalogue or triaging incoming requests, with human review where necessary.

The buyer would be purchasing delivered work rather than hiring staff to operate another application. This resembles a managed service, but AI could change its cost structure and the amount of work a small team can deliver.

That opportunity comes with responsibility. Integration, quality control, security and exception handling are part of the service. A demo that works on a clean example is not yet a dependable business.

Combine a subscription with variable charges

My expectation is that many businesses will use a hybrid: a base fee for access and administration, an included allowance, then additional charges for usage or defined outcomes.

This is a hypothesis about commercial incentives, not a forecast that subscriptions will disappear. Predictable budgets still matter to buyers. Predictable revenue still matters to suppliers.

AI can create value while destroying someone’s margin

Suppose AI makes a service cheaper to deliver. Several things could happen.

The supplier could keep the saving as profit. Competitors could lower prices and force it to share the saving with customers. A new entrant could offer a simpler service at a price that was previously uneconomic.

In the last case, the incumbent might lose revenue while customers gain access to something useful. A fall in one company’s sales is not, by itself, evidence that society has become poorer.

The reverse also holds: higher AI-company valuations or profits do not prove that most households are better off. Revenue, profit, productivity and household wealth measure different things.

I would expect the ability to capture value to depend on more than access to a model. Customer relationships, distribution, useful data, reliable delivery and responsibility for the final result could all matter. Which of these becomes defensible will vary by market.

Lower production costs could help small firms enter markets. They could also leave those firms dependent on a few infrastructure suppliers. Neither outcome is automatic.

More wealth does not guarantee more secure employment

The International Labour Organization’s 2025 index estimated that one in four workers globally was in an occupation with some exposure to generative AI. It identified job transformation as the most likely overall impact because most occupations still contain tasks requiring human input.[7]

Exposure is not a job-loss count. It measures potential technological reach, not how many people have been dismissed.

Nor does “transformation” guarantee a painless transition. A role can survive while its pay, bargaining power or entry requirements change.

It helps to separate three questions:

• Can a task be automated?
• Will an employer reorganise or reduce jobs because of it?
• Can an affected worker find another suitable job at a reasonable wage?

The third is about employability. It cannot be answered by showing that the economy still has jobs somewhere.

A risk worth watching is the loss of entry-level learning opportunities. If firms automate the routine work through which beginners gain experience, they will need another way to develop future specialists. That is a concern to investigate, not a claim that all junior roles are disappearing.

Distribution also matters. A 2025 IMF working paper, using household data and a calibrated model, describes how AI adoption could increase wealth inequality through capital returns and firms’ adoption choices. Its findings are model-based, not a measured forecast of what every economy will experience.[8]

An economy can therefore become more productive while particular workers lose income or face a difficult transition. “AI creates wealth” does not settle who receives it.

What businesses should measure next

For a business adopting AI, I would start with one real workflow and track more than hours saved.

Measure the cost of an accepted result, including review and rework. Check whether quality holds, whether demand expands and whether the supplier’s pricing makes the benefit worth paying for. Then decide explicitly what happens to the capacity released: more output, staff development, shorter working time or lower headcount.

For a founder, the test is whether a customer will pay for the delivered benefit after operating costs and failures are included. Selling access to AI is different from delivering something a customer can rely on.

For workers, gaining practical AI experience can help, but it is not an insurance policy. Employers also need to provide training, preserve routes into skilled work and support people whose roles change. Public policy affects how costly a transition becomes for those who cannot move easily.

My view is that AI can build wealth while weakening some existing business models and displacing some workers. Whether the gains become broadly shared prosperity depends on business decisions, competition, ownership and the support available during the transition.

Before calling an AI rollout a success, I would ask: what useful work became cheaper, who received the saving, and what happened to the people doing that work?

Sources

• [1] Adobe: Q3 FY2026 results
• [2] Adobe: generative credits FAQ
• [3] Snowflake: pricing model
• [4] Datadog: pricing
• [5] Adobe: additional generative credits
• [6] Adobe: Creative Cloud for teams
• [7] ILO: generative AI and occupational exposure
• [8] IMF working paper: AI adoption and inequality
• [9] Intercom: Fin AI Agent outcomes
• [10] PhotoCraft: source code
• [11] PhotoCraft: project and early-alpha status
• [12] Techstrong: ArtCraft and AI-assisted development