The Next Great AI Companies May Sell Work, Not Software

For the last two decades, the default ambition in software has been easy to describe: build a product, sell subscriptions, and grow by adding more users, more teams, and more seats. AI is beginning to challenge that model.

Sequoia's argument is that some of the most important AI companies may not look like software vendors in the traditional sense. They may look more like service businesses, but with software and AI doing so much of the underlying work that the customer is no longer really buying a tool. The customer is buying the result.

That distinction matters because it changes where value sits. If a company sells a tool, it is always exposed to the risk that the next foundation model turns its product into a feature. But if a company sells the completed task, then every improvement in the underlying models can make the business stronger. Better models reduce cost, improve speed, and increase reliability, while the company still owns the customer relationship and the delivered outcome.

A useful way to understand the thesis is through the distinction between intelligence and judgement. Intelligence-heavy work follows complex but learnable rules. Judgement-heavy work depends more on taste, prioritization, context, and experience. Writing code from a spec, testing, debugging, or translating instructions into output is often intelligence-heavy. Deciding what to build, when to ship, or which tradeoffs matter most is more judgement-heavy.

This leads to another distinction: copilots versus autopilots. A copilot helps a professional do the work. An autopilot delivers the work itself. Early AI companies often had to start as copilots because models were not yet strong enough to take responsibility for the full workflow. But in some categories, that is changing. The better the models become, the more viable it is to sell the outcome directly instead of selling productivity software to the person who used to do the task manually.

This matters commercially because the software budget is often much smaller than the labor budget attached to the same workflow. In many categories, companies spend far more on getting work done than on the tools used to support that work. That means the larger opportunity may not be to sell better software into existing tool budgets. It may be to capture part of the much larger spend that currently goes to human-delivered work.

That does not mean the right strategy is to replace whole professions overnight. The more practical pattern is to start where work is already outsourced. If a task is already outsourced, the customer has already accepted that the work can be done externally, a budget line already exists, and the buyer is already paying for an outcome rather than an internal process. Replacing an outsourced vendor is much easier than replacing internal headcount.

This also explains why some categories are better suited than others. The best early markets are usually repetitive enough to structure, valuable enough to justify adoption, and clear enough that buyers can trust the output. In those cases, AI does not need to create a new behavior. It only needs to deliver the same outcome faster, cheaper, or more reliably.

Another important part of the thesis is defensibility. If a company owns the workflow rather than only the interface, it can learn from repeated execution. It gathers edge cases, operational data, exception patterns, quality signals, and domain-specific examples of what good performance looks like. Over time, that can become more defensible than access to a model alone. In this framing, the moat is not just the AI model. It is the workflow, the system design, the controls, and the compounding data generated by doing the work repeatedly in the real world.

Seen this way, the next generation of breakout AI companies may not win because they built the most impressive dashboard or assistant. They may win because they turned labor into software-like outcomes. From the customer's point of view, the experience becomes faster, more reliable, more scalable, and more outcome-based. The customer does not buy help with the task. The customer buys the task being done.

What this means for Aune

This framework is highly relevant to Aune.

Aune's opportunity is not just helping users or AI agents find service providers. The bigger opportunity is helping them get the service job done. In real-world services, that means handling fragmented supply, variable availability, timing constraints, negotiation, and commitment. The value is not in showing a list of providers. The value is in turning demand into an executable outcome.

That is why this thesis fits Aune so well. Real-world services are full of workflows that have historically required coordination by phone, email, quoting, scheduling, and manual follow-up. If AI is going to matter in this category, it cannot stop at discovery. It has to move into execution.

For Aune, this suggests a sharper way to describe the company. Aune should not be seen as a better directory or a smarter lead-generation layer. It should be seen as infrastructure that helps AI agents and users convert fragmented provider capacity into real bookings under real constraints. That is much closer to outcome delivery than to traditional software.

It also reinforces why services are such an important category. Products can often be purchased from static catalogs. Services usually cannot. They involve timing, provider fit, availability, urgency, scope, and negotiation. That complexity is exactly what creates room for an AI-native execution layer.

The deeper implication is about moat. Aune's long-term defensibility is unlikely to come from generic access to models alone. It will come from owning more of the workflow: provider behavior, quote dynamics, response times, fallback patterns, booking outcomes, completions, cancellations, and real-world exceptions. If that loop becomes strong enough, Aune does not just help agents browse the service economy. It helps them operate inside it.

That is where this thesis becomes most powerful for Aune. The AI winners in services may not be the companies that help users search better. They may be the companies that help agents and users get the work done. Aune has the chance to become part of that execution layer.