Sell the work, not the tool — model improvements compound for services, against software

The strategic fork for AI companies is stark: “If you sell the tool, you’re in a race against the model. But if you sell the work, every improvement in the model makes your service faster, cheaper, and harder to compete with.” A company might spend $10K a year on QuickBooks and $120K on an accountant to close the books — the next legendary company will just close the books. This reframes LLMs selectively destroy vertical software moats — 5 fall, 5 hold from the buyer’s perspective: the moats that fall (learned interfaces, custom workflows) are all tool moats, while the real opportunity is capturing the work those tools enabled.

This inverts the typical startup anxiety about Frontier companies absorb every useful agentic pattern into their products. If your product is the outcome rather than the tool, platform absorption actually helps you — better models mean better margins on the work you deliver. The same dynamic explains why Domain-specific skill libraries are the real agent moat, not core infrastructure: accumulated domain workflows represent the ability to do the work, not just sell the tool for doing it. The margin gap matters structurally: Platform economics beat labor arbitrage — margins fund flywheels that body shops cannot shows how 50%+ margins fund R&D flywheels while competitors trapped at 10-15% margins cannot invest at all. Intercom’s trajectory demonstrates the full arc: Self-disruption follows the value chain downward — software companies must eat their own agent layer before someone else does — they went from selling software (tool) to selling agent resolution (work) to selling the models themselves (the engine), deliberately cannibalizing each layer before competitors could.