Skip to content

For teams building their own tools with AI

Your AI-built tools run. Now make them safe to keep using

Plenty of teams without an engineering background have used AI to build small programs, such as batch emailers or scheduled scrapers for government open data. They run, but people worry that no one can fix them when they break, are afraid to change them when needs change, and end up going back to off-the-shelf software. We can help you set up the architecture first, work alongside your team as you build, or advise when you need it.

Tell us about your toolsSee the three ways to work with us

Common situations

The program runs, but nobody feels sure about it

Teams that build their own tools with AI often run into these.

  • No one knows what the code does

    AI wrote it, and the person who built it can't explain each part. Changing one small thing means asking the AI to rewrite the whole section.

  • Failures go unnoticed

    An email run stops halfway, or a site redesign breaks the scraper, and nobody knows until someone notices data is missing.

  • Afraid to change anything

    Adding a field or a new list of recipients feels risky, because something else might break. So it stays as it is.

  • It only runs on one person's computer

    The program and its schedule live on the builder's machine. If that computer is switched off or replaced, or that person leaves, the work stops.

  • Passwords are in the code

    Email accounts and website logins are written into the code, and they go along with it whenever the program is shared or uploaded.

  • No one knows which programs are running

    Everyone builds their own and there is no list. When something breaks, you first have to ask around to find which program it is and who owns it.

So many teams stop after a few programs and go back to off-the-shelf software. Where packaged software fits, keep using it. We want to help with the part it can't cover and that feels too risky to maintain on your own.

Why this happens

AI makes writing code faster. The maintenance work stays

Getting a program to run is only the starting point. Fixing bugs, changing requirements and connecting to other systems afterwards is where most of the time goes.

AI amplifies the approach you already have. Decide early where programs run, where data lives, how passwords are kept and who is notified when something fails, and AI builds on that structure, keeping the code tidy. Without it, the faster AI writes, the more there is to clean up later.

The full write-up is on the founder's blog:Vibe Coding Tips: AI Is an Amplifier — Don't Let It Magnify Your Tech Debt

60–80%

Share of a software system's lifetime cost that comes after launch. The figure comes from IEEE research cited in the article.

What to add

Six things small programs usually need before you can rely on them

Take batch emailers and scheduled scrapers as examples. You don't need all six at once; start with what is most likely to go wrong.

  1. Keep the code in Git

    Every change is recorded, a bad change can be rolled back, and you can see who changed what and why.

  2. Store passwords separately

    Take them out of the code and keep them in a dedicated store, so they don't travel with the program.

  3. Run schedules on a fixed server

    Run them on a cloud server in your own name, without relying on someone's computer staying on.

  4. Get notified on failure

    When an email run fails or data can't be fetched, the person in charge gets a message instead of waiting for someone to notice.

  5. Write a handover note

    Note how to change it, how to redeploy it and where to look first when it fails, so someone new can take over.

  6. Test what matters

    When you ask AI to change something later, one test run tells you whether existing features still work.

Three ways to work with us

Pick one to start, based on where your team is

Choose any one on its own, or set up the architecture first and then have us coach the team for a while.

Architecture setup

Before building, we set up the structure together. Then your team builds on top of it with AI.

  • Review the programs you have and the tools you want to build next
  • Decide where programs run, where data lives and how passwords are kept
  • Set up a project template: Git, automated deployment, failure alerts
  • An architecture guide written for your team

Good for: teams about to start, or with a few programs that need tidying up

Talk about this option

Coaching

Your team builds and we review together on a regular schedule. What you build stays understandable, and you can keep changing it.

  • Regular online check-ins, at a pace that suits your team
  • Review what the AI changed and work through blockers together
  • Add tests, automated deployment and failure alerts step by step
  • Practise describing requirements to AI and checking what it produces

Good for: teams with someone willing to build, who wants an experienced person alongside

Talk about this option

Advisory

Handle most things yourselves and call us when a decision needs a second opinion.

  • Whether to build it yourselves or buy packaged software
  • Which tools and cloud services to use
  • Whether a program should be fixed, rewritten or retired
  • Checking AI-written code for security risks such as leaked passwords

Good for: teams that manage most things on their own and occasionally need someone to weigh in

Talk about this option

Packaged software or your own

If packaged software fits, use it

We often recommend existing software ourselves. Here is what to weigh.

Packaged software fits better when

  • Your process is much like most companies', such as bookkeeping, shift scheduling or leave requests
  • You need a vendor to handle security and regulatory updates
  • No one on the team has time to look after programs

Building your own is worth it when

  • Your process is specific to you and packaged software needs many workarounds
  • You need to connect data across several systems, such as forms, spreadsheets and email
  • You use one or two features but pay for the whole package

FAQ

Yes. We explain the architecture in plain terms and write the handover notes for your level. In the first coaching sessions we set up the tools and walk you through what the AI changed, so you don't need to learn programming syntax first.

Related articles

Write-ups on our founder's blog, covering how it was done and the problems along the way.

Tell us about the programs you have now

Let us know which programs you have, what they do and what worries you most. We'll suggest where to start, and if packaged software would suit you better, we'll say so.

Fill in the contact formWant one-to-one vibe coding lessons? See tutoring