The Definition of Done

Published 2026-06-30 · Updated 2026-06-24 · By Alan Wizemann

Topics: AI & Machine Learning, Engineering & Development, Product Strategy & Development, Startup

I have been building my own things lately, partly because I have always wanted to, and partly because anything I build might one day be worth something to someone else. So I have had a front-row seat to where software actually gets shipped right now, and the numbers coming off it are hard to believe. Product Hunt used to see twenty to forty new products on a launch day; now it is routinely in the hundreds. The app stores are buckling under the same wave, and someone close to the review side told me they are seeing something like a hundred submissions a minute (whether or not that figure is exactly right, you can feel it). I had an app sit in review for more than five days, when that process usually takes about thirty minutes. The reviewers are not slow; they are drowning, and so is everyone trying to push a real product through the same pipe as the flood.

Because that is what it is, a flood, and what makes it strange is that most of what pours out actually looks fantastic. The screens are clean, the onboarding is smooth, the marketing site is gorgeous. Then you get under the hood, and a great deal of it is a pile of shit – brittle code, security holes you could drive a truck through, no real architecture, nothing that survives contact with actual users at actual scale. It presents to the public as a sophisticated, finished tool, and it is, in roughly the way a movie set presents as a town. The front of it is beautiful, and there is nothing standing behind it.

This is the oldest lesson in building anything, and it is the one I keep coming back to. You can have it good, cheap, and fast, but you only ever get to pick two of the three. What AI did, more or less overnight, was make fast and cheap nearly free at the same time, and so almost everyone reached for those two and quietly let go of good. They convinced themselves nobody would notice, because the part you can actually see looks great, and for a while nobody does notice, which is precisely the trap. The polish is free now, but the substance never got any cheaper, and most of these things simply skipped it.

I watched a version of this up close years ago, when I was evaluating whether to acquire a company. The demo was clever and the front end was lovely, while the real story sat in the codebase you would inherit and the structural problems nobody mentions in a pitch. The difference today is only one of degree, but it is an enormous degree. The gap between how good something looks and how good it actually is has never been wider, and it has never been cheaper to manufacture the looking-good half. You used to need a real team and the better part of a year to fake this convincingly. Now four people in a garage can do it over a weekend, and if the thing demos well, they can raise money at a valuation with a startling number of zeros on it before anyone has stress-tested a single line of it.

I will admit, plainly, that I do not entirely understand the money part of this. There are companies firing experienced engineers, the very people who told them, directly, that the AI approach is not working yet, at the same moment that four-person teams with no track record and a polished prototype are handed billion-dollar valuations out of the gate. One group is being punished for telling the truth about where the technology actually sits, and the other is rewarded almost entirely for how finished its prototype happens to look. That can hold for a while, but I do not believe it holds forever, and so I keep circling the same question you ask about anything that looks too good to be true, which is simply this: what craters first? Maybe it is the indie ecosystem, all those polished, hollow apps that fall over the instant they meet real usage. Maybe it is the enterprise side, where companies bought the demo and are slowly discovering what was, or was not, behind it. Maybe a whole new tools market emerges to audit and harden and clean up the mess, and that quieter corner is where the next real businesses actually get built. Or maybe it is something none of us can see yet. I do not know which, and I am confident only of one thing, which is that the gap between presentation and substance is energy stored up somewhere, and stored energy, sooner or later, gets released.

None of this is an argument against the tools themselves. I am using them, they are remarkable, and fast and cheap are real advantages rather than illusions. It is an argument against confusing the output of those tools with a finished product, because looking done and being done have now come completely unbundled, and a great many people are about to learn the difference the expensive way, in production, with their customers watching. In product teams we already have a name for the thing that closes that gap, and we call it the definition of done. It is not a mood, and it is certainly not a demo. It is a written standard you attach to every story and every epic before the work even starts, agreed by the people who will build the thing and the people who will own it. Done means tested, and it means the security review actually happened, and it means the thing is documented and instrumented and holds up when ten thousand people hit it at once rather than when you click through it once for a screenshot. You write it down and bolt it to the work precisely so that done stops being a matter of opinion. It cannot be faked, because the standard is shared and visible, and it cannot be skipped, because the story does not close until the definition has been met. A demo passes the eye test, and the definition of done is what passes the eye test and everything sitting behind it, which is exactly the discipline this whole frenzy is racing straight past.

So if you are building something right now, my advice is simply to borrow it. Use the tools to move as fast as they will let you, then hold the work to a definition they cannot satisfy on their own, and refuse to call the thing shipped until it has genuinely cleared that bar. The market, for this strange moment, is paying enormous sums for a beautiful front and an empty back, but that is a quirk of a frenzy and not a law of physics, and frenzies tend to end with the people who skipped the hard part holding the bill. The screens have never been cheaper to make beautiful, and the work behind them costs very nearly what it always did. You do not get paid for fancy. You get paid for done, the real and written-down definition of it, the one you set on purpose so that it could not be quietly faked and could not be quietly skipped.