All founders building AI tools share the same question. When the next version of a foundation model (FM) comes out, will my product be reduced to just one of its features? It is a reasonable concern.

On that point, Julien Bek (줄리엔 벡), a partner at venture capital firm Sequoia Capital, recently stressed in a post shared on social media platform X (Twitter) that AI service companies should sell the work, not tools.

His message can be summed up as autopilot rather than copilot. “Copilot” means selling a tool to experts, while “autopilot” means selling the work directly to companies. He believes AI tool companies should pursue an autopilot model to survive as AI models evolve rapidly.

He said, "Until recently, AI models were still under development in both intelligence and judgement, so building a copilot first was the right direction. We put AI in the hands of experts and let them decide what to do. Harvey sold to law firms, and Rogo sold to investment banks. Experts were the customers, and the tool made them more productive. Now models have become smart enough. In some areas, it is better to start with autopilot from the beginning."

The intelligence and judgement he mentioned can be explained as follows.

He said writing code is mostly about intelligence, while deciding what to build is a judgement issue. Turning specs into code, testing and debugging are complex but rules-based tasks. Judgement is different. Deciding which feature to build next, whether to take on technical debt, or whether to launch before completion is intuition built from long experience.

He said, "Drafting documents, translating code and classifying data are intelligence tasks. Deciding what to build and how to set strategy are judgement tasks. AI has already reached a level where it can autonomously handle most intelligence work. Software engineering crossed that line first, and other jobs are following one by one."

In a situation where AI increasingly covers the intelligence domain, Bek says selling AI work, not AI tools, is advantageous.

He said, "In any job, work budgets overwhelmingly exceed tool budgets. Autopilot targets a bigger market from the start. A company can now spend $10,000 a year on QuickBooks and $120,000 on an accountant, but the next generation of leading companies will have AI handle everything up to closing the books without a separate accountant," adding, "If you sell tools, you have to compete on model and speed, but it is different if you sell the work. The better the model gets, the faster and cheaper the service becomes."

He pointed to outsourced work as the area where autopilot will take hold the fastest.

He said, "Companies have accepted outsourcing, there is a budget line item, and buyers are already purchasing outcomes. Replacing an outsourcing contract with an AI service is a supplier change. Replacing internal staff is a reorganisation. The former is much easier."

Autopilot's potential is also appealing in terms of market size.

He said, "Insurance brokerage is as large as $200 billion, and standardised paperwork is the core. Accounting and auditing is an $80 billion market in the United States alone, and there is no successor to fill the gap left by 340,000 accountants who have left over five years. Healthcare billing coding, insurance claims adjustment, tax advisory and legal contract work have the same structure." He added, "If you add IT management, supply-chain procurement and hiring, a multi-trillion-dollar market is waiting for autopilot." He continued, "The fastest-growing AI companies in 2025 were copilots. In 2026, many companies will try to shift to autopilot. Selling the work means making it so your customer cannot do that work themselves. While existing copilot companies hesitate, new companies that started with autopilot from the beginning are exploiting the gap."