PromptPartner DE

We build your app with coding agents, under an engineer's rules.

So an app for one of your jobs pays off with 20 or 200 staff, and it keeps working after the first change.

A colleague sketches a process on a whiteboard while two others discuss it at the table
Change 118: inspection appIllustration with sample data
  1. Written by a coding agentto the spec
  2. Tests on past inspectionsall passed
  3. Data model unchangedchecked
  4. Review by our engineerwaiting

Nothing goes live without a review

Why an app for one job is now worth it

For years, software written for one firm needed a team of developers. So most firms rented a standard product and bent their process to fit it. The approval sheet that gets printed and signed, or the inspection list kept in three files, is often what is left over.

In our projects, coding agents now write most of the code. That lowers the cost of building, so an app for one job of yours can pay off where it never did before.

The work around the code stays the same: understanding your process, designing the data, testing, and looking after the app once it runs. That is where our engineers spend their time.

Vibe coding is fine for a prototype

Vibe coding means describing an app to an AI and taking the code without reading or testing it.

Where it helps

It helps to show an idea in a meeting or to try a screen with the people who will use it. We do it too, as a first sketch.

Where it breaks

Once real data and real people are involved. Nobody designed the data model, the access rights are missing and nothing is tested, so the first change breaks something else.

Often nobody can say why the app does what it does, and the next step is starting again.

What production software needs

QuestionVibe codingHow we build
What is the app built from?Vibe coding: A promptHow we build: A written spec of your process, agreed with the people who do the job
Who decides how the data is stored?Vibe coding: The AI, case by caseHow we build: An engineer, before the first line of code
Who may see what?Vibe coding: Often everyoneHow we build: Access rights per role, with sign-in through your Microsoft Entra ID, Google or Okta account
What happens on a change?Vibe coding: You try it and hopeHow we build: Tests run on every change, and an engineer reviews it before it goes live
Can you see what happened?Vibe coding: RarelyHow we build: A log of what the agents decided and what your people did
Who looks after it next year?Vibe coding: Whoever wrote the promptHow we build: We do, every month, against fixed examples

We lay the foundation before the first screen

A screen is easy to change later, so we start with the data underneath it.

  • The data modelWhich records exist, such as an inspection, a defect or an approval, and how they belong together.

  • A decision logWhat each agent decided and why, and which tasks are waiting for one of your people.

  • Access rightsWho may read, change and approve what, set up per role from the first day.

The rules our coding agents work under are public

We publish them as AgentSmith, an open-source tool that sets the rules for AI coding agents. AgentSmith installs a shared operating agreement and a work profile into a project. The agent runs the relevant checks, exercises the real path, records evidence, and leaves a handoff the next session can check.

Your IT team can read every rule on GitHub before we write a line for you.

Your IT tried a coding assistant and stopped?

That happens often, and the reason is usually the setup. A coding assistant without a spec, without tests and without knowing your process writes code that looks right and fails on the third real case.

With rules, tests and a person who reviews, the same tools build software that lasts. That is the setup we bring to every project.

Questions about how we build software

Can our own IT build this with a coding assistant?

Often, yes, for parts of it. The difference lies in the setup around the tool: a written spec, a data model, tests on every change and a person who reviews. The rules we use are public as AgentSmith, so your IT team can read them.

How do you check code that an AI wrote?

The same way as any other code. Tests run on every change, and one of our engineers reviews it before it goes live. Each app also keeps a log of what the agents decided and what your people did.

Do you also build prototypes?

Yes, often as a first sketch to show the idea to the people who will use it. A prototype becomes an app for daily use once the data model, the access rights and the tests are in place.

Who looks after the app once it runs?

We do. We host it in your dedicated environment and look after it every month: small changes, a check against fixed examples and a one-page report.

We build your app with coding agents, under an engineer's rules.

In a 30-minute first call we look at the job with you and tell you whether it needs an app at all.