By Ryan Richardson · 14 November 2026
Ask your builder for a written list of everything left, with a time against each item, and have an independent person read the code. Stop paying for more work until you have both, then decide whether to fix or rebuild.
I was brought in to rescue a mobile vet booking product that had sat at eighty percent for two years, through two builders and around $30,000.
Almost done is a feeling, and it is no use as a project status. Done and almost done can be a week apart or a year apart, and the person saying it often cannot tell which.
So ask for the list, with every item that is left and an estimate of time next to each one. If your builder cannot write it, the build is not done. When almost done stretches on for months, the scope was nearly always underestimated and nobody wants to be the one to say so.
I have changed the founder's personal details, and her scenes are a sketch. The build, the money and the timeline are as they happened.
She knew exactly what she needed. Booking, routing a day between houses, reminders, and notes attached to an animal and its owner. The designer who had helped her think it through offered to build it. It worked in a demo and then did not hold.
A developer was hired to finish it and went in circles, fixing one thing and breaking another. Every time she asked what was wrong she got a real bug, the bug got fixed, and the product was still at eighty. The trouble was a tangle of small bugs with no single place anyone could point to.
Meanwhile she chased reminders by hand, in the car, between jobs.
She asked whether I was telling her to bin it because binning it suited me. The honest answer was that finishing the tangle would be the long, open-ended job.
We kept her thinking and threw the code away. In two days I rebuilt everything they had built, using common, reliable parts that real products already run on. The product is live today with real users.
Common parts and planning are where that kind of speed comes from. On any build I keep the choices simple and standard. Two of my usual habits, given here as examples and not as a record of that rebuild, are ClickSend rather than Twilio for messages and the same navigation bar on every page, so one pattern repeats and each change stays small. Where every page was made its own way, a developer can take a day to make one wording change, by the time they work it out, edit, review and push it. Once a build is systemised, the same change takes minutes, even without AI. Most of this predates AI. By my estimate, AI has added another 50 to 100 times improvement on top. In your own build, ask your developer how long one small wording change takes from request to live, and put the answer beside the list of what is left.
The full story is chapter three of my $5 book, Turn Your Expertise Into a Product.
Paying for more of the current work before you have the facts keeps you on the path that stalled. Starting again on a hunch feels like progress, and it can be the most expensive move, because it throws away work that functions.
The independent read should answer four questions.
Rebuild when the read finds the problems tangled together and what your product needs already works in other products. That was the vet build. Fix in place when the problems can be named one by one, each with a time against it, and most of what exists works.
If a new developer says it needs a full rebuild, and that includes me, ask them to name each problem, how long each would take to fix in place, and what a rebuild would cost and take by comparison. Then ask what they would do with their own money.
A builder chasing the work tends to give you vague talk about deep problems in how it was built, and a firm push toward the rebuild. If they cannot name the problems one by one, get a second opinion before anything is torn up.
The code, the wiring and the screens had eaten the two years. The founder also held her knowledge of which five things matter in a mobile veterinary business and which fifty don't.
Thousands of products already do booking, reminders and notes reliably. What was hers was the judgement about which of them matter in a house with a frightened cat under the bed, and in what order.
Agree it in writing before anyone starts. My own definition has six parts, and you can check each one without reading code.
The questions to put to the next builder before you sign their quote are in how to read a software quote.
Send your builder one message today asking for that written list of everything left, and hold the next payment until it arrives.
If you cannot read code, you judge a build by time and money, and both lag behind the work. They tend to show a problem around eighteen months after it started going wrong, which is when acting on it costs the most.
It is close. People come to me saying they moved fast at the start and have now been stuck at ninety percent for months. When an app is built by typing one request at a time into an AI tool, what comes out is a set of single features that do not fit together. Start the same way, with the written list of what is left.
Rarely. Ask whether the current language is causing a real problem or is just not their favourite tool. Real reasons look like a speed problem that cannot be fixed another way, or being unable to hire anyone who knows the current language. "It would be cleaner in the other one" does not count. Get a second opinion from someone other than your developer before you throw out code that runs.