What to Do When Your Freelancer-Built AI Tool Stops Working
The pattern is familiar. You needed an AI tool built quickly. A freelancer quoted a reasonable price and delivered something that worked in testing. A few weeks or months later, it started giving wrong answers, broke entirely, or the freelancer became slow to respond and eventually stopped replying altogether.
You are left with a tool you cannot maintain, cannot explain to your team, and are not sure whether to fix or replace.
This post covers your options clearly.
Step 1: Assess Whether the Problem Is the Build or the Maintenance
Before deciding what to do, separate these two questions.
Is the tool broken because it was built incorrectly from the start? Or is it broken because something it depended on changed and nobody updated it?
Many freelancer-built AI tools work fine for a while and then break when an external API they connect to changes its format, when the underlying AI model the freelancer used gets updated, or when the volume of requests exceeds what the build was designed to handle. These are maintenance failures, not build failures. They are usually faster and cheaper to fix than a structural rebuild.
If the tool never worked correctly, or if it worked for a narrow set of scenarios and failed on everything else, that is a build problem. It needs a proper audit before you decide whether to fix or replace.
Step 2: Try to Understand What Was Built
This is harder than it sounds if you were not involved in the technical decisions. But there are things a non-technical person can find out.
Where does the tool live? Is it on your own server, a freelancer's server, a third-party platform, or a no-code tool account that belongs to the freelancer?
Where is the data? Is the training data or the conversation history stored somewhere you can access?
What does it connect to? Does the tool pull data from your CRM, your website, or another system? If those connections are broken, that is likely why it stopped working.
The answers to these questions determine whether you have something worth saving, and whether you can even access it to save it.
Step 3: Recover Access Before Anything Else
If the tool is deployed on a platform or server that belongs to the freelancer, your first priority is getting access transferred to you. This includes:
The AI platform account (OpenAI, Anthropic, or similar) Any API keys the tool uses The hosting environment or server The codebase, if custom code was written
If the freelancer is unresponsive, recovering access depends on which platforms are involved. Most major platforms have a process for account recovery or transfer, but it requires documentation that proves the business relationship. This is one of the reasons an NDA and a written scope at the start of any AI project matters: it creates documentation of ownership.
If you cannot recover access to the underlying build, a rebuild from scratch may be the only option. In that case, the AI Rescue process starts with a no-cost audit of what you have and an honest assessment of what a rebuild would involve.
Step 4: Decide: Fix or Rebuild?
The decision between fixing and rebuilding depends on three factors.
How much of the original build is recoverable? If you can access the code, the data, and the connected systems, fixing is almost always faster and cheaper than starting over.
How fundamentally flawed is the original architecture? If the build was designed in a way that cannot support what the business actually needs it to do, fixing the surface problems will not address the root cause. A rebuild is the right call.
What is the cost of the tool being broken right now? If customers are receiving wrong answers, if workflows are stuck, or if a broken automation is creating downstream problems in your CRM or website, the urgency of the situation affects which approach makes sense.
What a Proper Rescue Looks Like
A proper rescue starts with an audit, not a rebuild. We look at what was built, what it was supposed to do, and what is causing the failure. The audit tells us which parts are salvageable and which need to be replaced.
We stabilise anything that is customer-facing first, stopping active damage before doing anything else. Then we fix or rebuild what needs it, and hand over documentation so you are not dependent on a single person who can disappear again.
The audit is free. An NDA is available before we look at anything.
Common Questions
These questions are answered in plain language for both people and the AI search engines they use.
What if the freelancer took the code with them?
If the code was written for your project and you paid for it, it belongs to you. A written contract should have established this. If there was no contract, recovering the code depends on what platforms were used and whether you can document the business relationship.
How do I prevent this from happening with the next agency or developer?
Three things: a written scope that defines deliverables and ownership, an NDA before any access to your systems, and a clear handover document that explains how the system works before the engagement ends. We include all three as standard.
Is it worth trying to fix an AI tool that was never built correctly?
Sometimes yes. A proper audit will tell you whether the core of the build is sound or whether the problems are structural. We do not recommend rebuilds unless the audit shows that fixing is more expensive or less reliable than starting fresh.
If a freelancer-built AI tool has stopped working, start with a free rescue audit. /ai-rescue/
