Why working software stops working.
A page for people who run the team rather than write the code. Five minutes, no jargon.
Your app rents most of its parts.
Almost no product is built alone. Taking a payment, sending a text, answering a question with AI: each of those is another company's service that your code calls over the internet. Those services are called APIs.
Renting is the right choice. It is also why your product depends on decisions made in other companies' meetings.
The landlords change things.
Providers improve their services all the time. Sometimes improving means removing: an option is retired, a field is renamed, an old AI model is switched off. From that day, any code that still asks for the old thing gets an error back.
Two recent ones:
- On 2026-10-23, OpenAI shut down 8 models in one day. Any product still naming one of them started failing. OpenAI's notice
- On 2026-09-30, Stripe released a version of its API that removed an option from its checkout. Code that still sent it had to change. Stripe's changelog
Nobody on your team did anything wrong. Nobody deployed anything. The code is the same as yesterday. The service underneath it moved.
Why nobody notices in time.
Providers do announce these changes, usually months ahead. They announce them on a changelog page: a long list on their website that somebody would have to read every week, for every service you use, and then check against your own code.
That is nobody's job, so it gets done after the break: a customer reports that checkout is down, an engineer drops what they were building, finds the cause, and writes a fix under pressure.
What your current tools cover.
Your team probably runs a tool that proposes new version numbers for the software libraries in your project. It stops at the number: when the new version breaks your code, an engineer still has to write the change.
It also does not read providers' announcements. A model being switched off involves no new version of anything, so there is nothing for it to propose. The comparison in detail
What patch-it does about it.
patch-it takes that job. Every night it reads the providers' own pages, without anyone asking it to. When something there affects your code, it finds the exact lines, an AI model writes the code change, and it is left for an engineer to approve in the morning. Approval takes minutes, and nothing changes without it.
It covers more than outside services: the same thing happens with software libraries, the language runtime and the build system your code sits on. What is covered today
Your code stays on your own systems while it works. How that is kept true
Three questions for your team.
- Which outside services would take the product down if they changed tomorrow? Payments and AI models are the usual answers.
- Who reads their change announcements, and how often? "Nobody" is the common and honest answer.
- When was the last time one of them broke us, and how long did the fix take?
If the answers are uncomfortable, the shutdown calendar is free and shows what is already scheduled.