AI‑assisted development: what it changes for your project

What has changed

A coding agent reads a whole project, writes functions, runs the tests and fixes its own mistakes. Tasks that took a day take an hour. The developer’s job shifts: less typing, more deciding and verifying.

What goes faster

• Repetitive codeForms, admin screens, API calls.
• TestsWhich were written too rarely for lack of time.
• MigrationsMoving to a new framework version, renaming everywhere.
• Understanding existing codeYou are taking over.
• DocumentationAnd interface translations.
• PrototypesTo compare two ideas before choosing.

What does not go faster

• Understanding the need and deciding the scope.
• Choosing the architecture and the data schema.
• Waiting for feedback from the client and users.
• Store review, payment providers’ lead times.
• Finding a bug that only occurs in production.

On a complete project, code was never the largest share of the time.

The risks of generated code

Risk	What happens	The answer
Security	Forgotten access rights, secrets in the code	Targeted review, automated analysis
Inconsistency	Three ways of doing the same thing	Written conventions, fixed architecture
Dependencies	Useless or invented libraries	Checking every addition
Fake tests	Tests that pass without checking anything	Reviewing the tests themselves
Debt	A lot of code nobody has read	Small deliveries, reviewed one by one

The method that holds

1. 1A human decides the architectureBefore the first line.
2. 2Written project instructionsConventions, commands, what must not be touched.
3. 3Small tasksOne at a time, each verifiable.
4. 4Tests firstSo the agent has a target.
5. 5A review of every changeAs for a colleague’s work.
6. 6Continuous integrationThat rejects what breaks.

That is how we work.

What the client gains

• More frequent deliveriesHence earlier feedback.
• More finishFor the same budget: tests, accessibility, translations.
• PrototypesTo settle a question instead of debating it.

Pocabase serves 46 languages from a single codebase. Votons covers a website, an API and two mobile apps. Scopes of that size remain within reach of a small team.

The guarantees to ask for

• Who reviews the code produced?
• Are there automated tests, and what do they cover?
• Is the code in my repository, readable by another developer?
• Are my data and my code used to train a model?
• What happens if the tool used changes or disappears?

For a founder

Generation tools let you build a convincing prototype on your own. It is an excellent way to sharpen an idea. Before putting real users and real data into it, have it reviewed. See Building an MVP with AI.