
Customers buy a result that works in their business, not an AI tool.
Customers pay millions of won for a single website or chatbot.
If I look only at the work I do, that price seems strange. I meet the customer, clarify the requirements, and hand the implementation to Codex or Claude. Work that used to take days can sometimes be finished in hours.
Yet it is still cheap for the customer. Something that would have cost ten million won at a conventional agency can now be delivered for a few million. Making it has become easier. Buying it is still difficult.
AX, or AI transformation, means putting AI to work inside a company's actual operations. The market is growing, but the money can seem almost too easy. Will this gap disappear once everyone becomes fluent with AI? Or will profitable teams of specialists remain?
Companies Buy the Execution of an Expert Team
Explaining why customers pay by saying that they simply do not understand AI misses something important. Knowing a tool and putting it into a workflow are different things. Someone still has to turn an ambiguous request into a workable scope, connect the data, and decide how the result will be checked.
Customers often ask, “Can this be done?” Separating what is possible now from what needs a different approach is more than coding. It is translating technology into the customer's problem.
Consulting works in a similar way. McKinsey and BCG assign people to each project and solve the problem again, so they do not scale in the same way a product does. Even so, they earn substantial fees by building expert teams and a concentration of talent that companies struggle to maintain internally. When an important problem arises, the customer borrows that team's judgment and ability to execute. If each engagement is valuable enough and the firm can keep attracting strong people, solving problems one at a time can still make a large business.
AX could take a similar form. The difference is that the work can end in a functioning system rather than a report. Customers are buying a result that fits their operations and the team capable of producing it. Bringing in a small group of specialists when needed can be faster and cheaper than maintaining that capability in-house.
Accenture's new generative AI bookings nearly doubled from $3 billion in fiscal 2024 to $5.9 billion in fiscal 2025. Companies still find it difficult to produce results internally, so a market for professional services is opening before many of the products are mature.
New Technology First Creates a Market for Outside Experts
When the Internet became a priority for businesses in the late 1990s, e-consulting firms appeared in large numbers. What they sold was not simply a homepage. They built businesses on the Internet, including online stores and customer portals.
Companies without the experience bought entire teams of strategists, designers, and developers from outside. The market expanded rapidly, but it did not last. According to Harvard Business School, more than 100 firms were considered e-consultants in January 2000, and more than half had disappeared by December 2001. When the dot-com bubble burst, customers started asking about ROI rather than speed.
Viant's revenue more than doubled from $61.3 million in 1999 to $127.2 million in 2000, then fell 73% to $34.6 million in 2001. Professional services expenses declined by only 38% that year. Demand could disappear immediately. Payroll and office costs could not.
A new technology creating a large services market and a company in that market being a good business are two different things.

Demand can change quickly. A project-based firm's fixed costs rarely adjust at the same speed.
The Survivors Sold Operational Capability
Not every firm disappeared. Sapient had to cut its revenue base and workforce substantially, but it survived as a company that handled everything from defining a business problem to operating the resulting systems. After demand contracted, it expanded delivery across regions and increased recurring operational work, including application management and quality assurance.
Even a project-based firm can lower the cost and risk of its next engagement by accumulating industry knowledge, development methods, and operating processes. Productizing the business does not have to mean abandoning all service work to build a single SaaS product.
Mobile agencies were also acquired by large consultancies such as Deloitte and Accenture, and their capabilities became part of those firms' core offerings.

Code, industry knowledge, and operating practices can make the next project less costly to deliver.
This Time, Model Capability Can Raise Productivity Without More People
AX differs from what came before. A professional services firm's output has traditionally been proportional to its headcount and working hours. An AX firm's output could look more like headcount × model capability.
The same person can deliver a broader result when the next version of Codex or Claude can handle more of the work. A team's production capacity can grow simply because an external model improves, without the team hiring anyone else.
If the number and difficulty of projects one person can manage keep increasing, a firm may be able to grow while remaining small. That challenges the assumption that a services company must become a SaaS business before it can scale.
Of course, better models do not automatically create an advantage. Sales, trust, data access, an understanding of the work on the ground, and verification of the result still depend on people. A model can write the code in an hour and produce no value if it solves the wrong problem.
An AX firm's core capability may be less about prompts than about dividing a problem well: turning an ambiguous request into executable units of work and placing the model's output inside a system people can trust.

A small team's capacity can grow with better models, while judgment and verification remain essential.
A Large Market Does Not Have to Produce Only Large Companies
I see three possible forms for AX firms.
The first is an expert boutique selling execution and trust in a particular industry. The second is a delivery business that reuses code and operational knowledge to make its work more repeatable. The third is absorption into a large consultancy or a company's internal organization. In that case, the label “AX company” itself may become less distinct.
Prices for straightforward outsourced work will fall, but I do not think the gap will disappear entirely. The range of work models can perform may expand faster than companies can change how they operate. As today's difficult work becomes easy, outsourcing can move from websites to internal automation, and then to redesigning the work itself.
The core of the AX market is not an early AI user selling instructions to a late adopter. It is connecting the rapidly moving edge of technology with organizations that move more slowly.
A large market does not necessarily produce only a few enormous companies. AX could support many small, profitable teams of specialists. The firms that last may not be those that know the most about models. They may be the ones that increase the value they give customers every time the models improve.