Your customers are rebuilding your product to find out what replacing you would cost. They’re measuring the part of your business that runs on a calendar.
TL;DR: Your customers can now rebuild enough of your product to see what replacing you would cost, then throw the result away. What they cannot reproduce is the part of your business that runs on a calendar. A SOC 2 window has to actually pass; a reference account needs a year. The buildable list keeps growing, and the waiting list does not move. And if an exit is in your future, your acquirer prices you on a wait that outlives your own horizon. Count it in months, on a vendor before you count yourself.
You have been watching your competitors. Wrong window. An engineer at a company that already pays you was asked in a budget meeting whether what they buy could simply be built instead; nobody had a confident answer, so they went and found out. They never finished. Deciding they probably could was enough, which is a much cheaper way to have strong leverage at renewal than actually building the thing. This is not one customer having a clever afternoon. It happens often enough now that the big annual surveys carry a line item for it.
Acquirers run the same experiment, with better catering and a data room they will not need (the point of the exercise is that it works without one). The test does not always come back badly. A diligence team at one of the big consultancies ran it on two targets and came back with opposite answers. Against the first, the technology looked reachable. Against the second, the company looked well dug in, and what separated them was how thoroughly the customers had already moved in.
So the useful question is smaller than the panic. Of everything your product is, how much could five motivated engineers with enough tokens reproduce in a week, and how much would they simply have to wait for? That second list is the only one that was ever protecting you. It is also the one that has yet to appear on a single slide in your possession.
Building is work, and work is exactly what got cheap. The rest of your defensibility was never work at all. It is waiting, and waiting has never been on sale. A SOC 2 window has to actually pass, and calendars are indifferent to your roadmap. A reference customer has to spend a year not being let down. Usage data arrives at the rate your customers do their jobs, a rate you have never once been able to influence. And the files, the settings, and the workflows your customers have bent around your product took years, and no amount of money makes them go faster. Nobody can speed that up. Including you.
You will want a figure. There is none. Nobody publishes how long that wait runs for any kind of software, and the research group that set out to measure something far simpler, how much faster AI actually makes a developer, stopped trusting its own results and went back to redesign the experiment. That’s because the developers in the study flatly refused to work the old way just to be measured. The instrument broke on contact with the thing it was measuring, which is a very modern sort of failure.
So you estimate it yourself, or you go without. And yes, a range of months on a board slide is a guess wearing a suit. It will be read as a measurement and quoted back to you in a year and a half. I do not have a good answer to that. I have only noticed that the alternative is an adjective, and no adjective has ever survived a room full of people doing arithmetic.
How long the wait needs to be is not a property of your product. It belongs to whoever is asking. A renewal wants one answer, a payback period another, an exit a third. The exit is the one worth losing sleep over, because whoever buys your company is not really thinking of you in the present. They are already thinking of the person they intend to sell it to years later, and they are pricing accordingly. You are being valued on a wait that has to still be running long after you have stopped caring how it ends.
I sat on the buying side of a carve-out not long ago where the question underneath every other question was how quickly AI would replace the software. Put less politely: how fast does this revenue go to nothing? That is the same instrument, pointed at somebody else’s company, and it turns build, buy, or partner into arithmetic. Run it outward and the logic inverts. A short wait means building it yourself just became a real option. A long one means the price is for the clock, not the code, and an acquisition premium becomes payment for elapsed time.
For the next vendor on your desk, generate two lists. What five engineers could reproduce in a week. What they would have to wait for. Put months against the second one, estimated by whoever owns the audit, keeps the reference accounts, and watches what customers have built inside the product. Write the reasoning down so somebody can argue with it later. Then run those two lists on your own product. If the waiting list comes back short, start the slow things now. Nothing else will still be running when somebody runs this test on you.
Every software company is now both examiner and specimen. The only real choice left is how soon you find out which one you are.
Sources
Becker, Joel, Nate Rush, Tom Cunningham, David Rein, and Khalid Mahamud. “We are Changing our Developer Productivity Experiment Design.” METR, 24 February 2026. https://metr.org/blog/2026-02-24-uplift-update/ Accessed 2026-09-15
Farmer, Benjamin, Gene Rapoport, Richard Lichtenstein, Emmanuel Coque, Amy Wall, and Parker DeRensis. “New Diligence Challenge: Uncovering AI Risks and Opportunities.” Bain & Company, 27 October 2025. https://www.bain.com/insights/new-diligence-challenge-uncovering-ai-risks-and-opportunities/ Accessed 2026-09-15
Tinkoff, Dan, Lieven Van der Veken, Michael Chui, and Tara Balakrishnan. “The state of AI in 2026: On the road to ROI.” McKinsey Global Survey, 25 August 2026. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai Accessed 2026-09-15
Originally published at www.orionplaybook.com.