Artificial intelligence in the leading role

Read how we approach AI and what it means for you.

What we use AI for

We have artificial intelligence plugged in across the whole of our work, from design through code and tests to monitoring. It speeds us up but does not replace judgement.

We run entirely on Claude AI. We are trained specialists in it, and what we use ourselves we can deploy at your place too.

Draw a screen out into ten variants
Make custom icons and illustrations
Try out colour and typography sets
Turn a sketch into a clean layout
Send the client previews within the hour
Build variants of the front page
Write most of our code
Rewrite an old component into React
Make a single-use migration script
Explain the code left by the previous supplier
Refactor the modules we keep putting off
Run routine edits through the whole repo
Derive types and schemas from the data
Write a regular expression that works
Put together an SQL query with an explanation
Connect a third-party API from its documentation
Write tests for every new feature
Come up with scenarios nobody would think of
Run the tests on every change in the code
Reproduce a reported bug
Check the accessibility of an interface
Test forms against nonsense input
Load-test the site before launch
Write the first version of the website copy
Rewrite error messages into human language
Shorten a paragraph that overflows
Check whether the site says the same thing everywhere
Fill in the descriptions for search engines
Write the alternative texts for images
Come up with claims to choose from
Put together a newsletter people finish reading
Sum up what changed on the project today
Turn a video call into minutes with tasks
Send out the meeting minutes the same day
Write documentation that would never have existed
Prepare the groundwork for an email reply
Pull a changelog out of the commits
Bring a new colleague into the project
Translate the texts into English
Keep an eye on errors in live operation
Add the context of what happened before the crash
Propose a fix before the customer gets in touch
Sum up the week's swings in operation
Find the slow database queries
Keep the monthly cloud costs in check
Go through the logs after a night-time crash
Set up alerts that do not shout for nothing

How we make design

Designing used to mean betting on one direction and sticking to it. Every further version cost days of work, so in practice they were never made: the client got one, plus an argument for why it was the right one.

Thanks to AI we can explore design directions and generate prototypes fast. What used to take several days now takes one afternoon.

One motif, different graphic directions.

Yes

Trying out new ideas and directions fast, the ones there was never time for.

No

Generating a pile of variants and selling them as finished work.

How we write code

Most of the lines are written by a model today, and it shows mainly in the pace. What has not changed is who vouches for them: somebody has to read the proposal, decide whether it solves the right problem and prove with a test that it really works.

We generate 90 % of our code today. But only what somebody has read and checked goes out.

Yes

Letting the model write code, routine, tests and one-off scripts, and pushing every change through them.

No

Deploying generated code nobody understands.

Security

The model works with code, not with your data. Credentials, keys and production databases stay aside: they never make it into the model's brief, and in the repository a check guards them and will not let them in.

What is sensitive does not go to the model. And what does go, we know exactly.

Yes

We give the model code, documentation and test data. We always know what we are sending it.

No

Sending credentials, keys or clients' live data into the model.

No

Letting the model onto production without a human to approve the change.

No

Adding tools and connections nobody has vetted.

No

Taking instructions from content that reached the model from anywhere other than us.

What does it mean for you?

The rules above are not there for their own sake. This is how they show up in the job.

We save your time

Research, the skeleton of the code, tests, data migration: routine that used to take days is done in a moment. You notice it in the dates: we deliver in weeks, not quarters.
We save your money

We charge for work delivered, not for hours we did not work. What AI saves us, you see in the budget, not in our margin. That way even a small job is worth doing, the kind nobody would have taken on before.
We deliver the highest quality

The time saved does not go in a drawer: it goes back into tests, reviews and the details there was never room for. Under every line that leaves us a human is still signed: the model writes, we decide.