Code Is a Commodity Now. Stop Buying It Like It Isn’t.

Writing code used to be the bottleneck and the bill. With AI-assisted development, the marginal cost of producing working software has collapsed. Most enterprise contracts are still priced as if it hadn’t.

The shift, plainly

For thirty years, software cost what it cost because the people who could write it were scarce. That assumption is no longer holding. A competent generalist with the current generation of AI tooling now ships in a day what used to take a senior engineer a week. The work has not disappeared. The marginal cost of producing it has collapsed.

This is not a forecast. It is observable in any team that has adopted AI-assisted development with any kind of discipline.

What is still expensive

The interesting question is what remains scarce. After a year of watching teams adopt AI tooling at speed, the honest answer is short:

• Knowing what to build: AI does not tell you which problem is worth solving. It cheerfully helps you build the wrong thing faster.
• Integration with a real organisation: the time-consuming parts of enterprise software are not "writing the function". They are the contract with the data team, the legal review, the procurement signoff, the change management. None of that has been compressed.
• Being right about correctness: a model can produce code that compiles, passes tests, and looks plausible while being subtly wrong in production. Reviewing for actual correctness is harder, not easier, because the floor of what looks right has risen.
• Domain and regulatory judgement: a model trained on the open internet does not know your contracts, your audit obligations, your incident history.

These are the parts that used to be hidden inside "engineering hours". They are now the entire bill.

The buyer’s mistake

Most enterprise software contracts are still priced in engineer-days. The vendor sells hours. The buyer pays for hours. Neither side has caught up to the fact that the cost of one of those hours has dropped.

The result is predictable. Vendors keep their rate cards, pocket the productivity gain, and call the AI tooling an "internal efficiency". Buyers keep paying 2022 prices for 2026 productivity. The gain ends up in someone’s margin, not in the buyer’s value-for-money.

What done-right looks like

A serious AI-assisted engagement does not look like the same hours, billed the same way, with a model on the side. It looks like:

• Fewer people, doing more, in less time.
• Pricing tied to the outcome: FTE-hours returned, decision cycles compressed, defect rate down.
• A clear contract about what the team must still own (judgement, integration, governance).
• Ownership transfer at the end, not vendor dependency.

If your AI consultant cannot commit to those, they are still selling hours.

Where we sit

This is the reasoning behind our AI-Assisted Development practice. We do not sell engineer-days dressed up with AI. We help teams lift their ceiling, and we price the engagement against the gain. When code stopped being scarce, the work stopped being "more code". It became knowing what to build, integrating it cleanly, and being right about it.

The bill follows from that, not from a timesheet.