What If the Price of Your AI Workflow Tripled?

Why are people saying AI tools are underpriced?

Because the unit economics have not closed. The argument going around in 2026 frames it as an inference subsidy, with reporting suggesting the tools businesses run on are priced well below what they cost to serve. The specific percentages circulating vary by source and deserve to be read as directional rather than precise.

The argument itself is not new. Sequoia published a widely read version of it in September 2023. What changed is that more businesses now have production workflows sitting on top of the assumption.

Is this a reason to stop using AI?

No. Treating it as a prediction about the market is the least useful response available, because nobody running a mid-sized company can act on it. Treating it as a question about your own operation is different, and it takes an afternoon to answer.

What is the actual risk to an operator?

Every place an AI step sits inside a live workflow, there is a price assumption underneath it that never got stated. Nobody wrote it down because at current prices it never had to be defended. That is fine until the renewal notice arrives with a different number and the decision has to be made in a week.

How do you find your exposure?

Walk the workflow and sort every AI step into three groups.

• Still worth it at three times the price, because the value is high and the volume is low.
• Done by hand instead, because the task was always cheap labor and the tool was a convenience.
• The step stops, and takes the surrounding process with it, because nothing else was built to do that job.

That third group is the one worth knowing about. Most companies have one or two entries in it, and they are rarely the ones people expect.

Why is the third group so easy to miss?

Because the steps that quietly become load-bearing are the ones that work. A summarization step that never fails stops being reviewed. A classification step that routes correctly stops being questioned. Six months later a process depends on it and no manual fallback exists, because nobody thought to keep one for something that costs so little.

Does this apply to automation generally?

Yes, and the AI framing is a distraction from a more durable point. Any step you did not build, cannot replace, and do not control the price of is a dependency. AI made those dependencies cheap enough to accumulate quickly, which is the only genuinely new part.

What is the practical next step?

Take the workflow closest to revenue and mark each AI step with which of the three groups it belongs to. If the answer for any step is that the process simply stops, decide now what the manual version looks like and write it down while there is no pressure.

That document costs an afternoon. Producing it during a renewal negotiation costs considerably more.