People decide what gets built. AI does the heavy implementation work, under our direction. Nothing ships until it has been driven the way a real user would drive it.
Design.
We work with you to define what the product needs to do, who it’s for and what done looks like, before anything is built. People make those calls.
Build.
We build with AI, under our direction. We hold the architecture, review what comes back, and own what ships.
Test.
We aim to cover every journey a real user could take. AI agents drive each one, and a person reviews the results.
Support.
Launch is the beginning of the partnership, not the end of the project. We support the product after go-live and keep improving it with you.
How we work
You can see how the work is done, and you deal with the same people throughout. We hold the architecture, review what comes back and own what ships.
Stage 1 · People
We work with you to define what the product needs to do, who it’s for and what done looks like, before anything is built. People make those calls.
Stage 2 · AI, directed by us
We build with AI, under our direction. We hold the architecture, review what comes back, and own what ships.
Stage 3 · Driven like a user
We aim to cover every journey a real user could take. AI agents drive each one, and a person reviews the results. Not just unit tests passing in isolation.
Stage 4 · After launch
Launch is the beginning of the partnership, not the end of the project. We support the product after go-live and keep improving it with you.
Working with us
A long collaboration, not a one-off. Diligence and care on the product, and a bit of craic along the way. The aim is to be your software department, so you do not have to hire one.
01
A named person answers for the work, and you’ll know who they are. We pay attention to the detail, and we treat you and your users with respect.
02
We would rather build a long relationship than hand over a folder and move on. You keep working with the same people as the product grows.
03
We can act as your software team, or work alongside the one you have, so you get the product made without a hiring round.
Example case study
We built Pitched, the AI-assisted marketplace. We designed it, AI implemented it, and before release we put it through an AI-run QA harness.
~180
user paths in the harness
3
screen sizes, each driven the same way
216
screens checked for visual regressions
Pitched was tested as a web app at those three layout widths by an AI-run QA harness. It was not tested on physical handsets.
pitched.ieContact
Find a way through
The bold letters are the walls. Split one and you go back to the start.
Find the opening.
A team of agents tested this maze by playing it, before it went on the page. They moved the square on their own. The letters run in reading order across a grid, and that is the whole game. Where the type is bold, that letter is a wall. Where the type is light, that letter is the way through. There is no tile and no picture behind the words. The agents drove the square through the light letters. Letters the square had already crossed turned blue, so the way they came stayed on the page. The way out they had to find is the blue opening in the bottom wall. When the square split a bold letter, the wall broke and the square was sent back to the start. The blue letters stayed, so the failed run was still there. They set off again from the start. They found the opening. They also split a bold wall, and that sent them back, so the failure was tested as well as the way through. They did this on a wide screen and on a narrow one. Both runs were in a browser, at those two layout widths. It was not tested on physical handsets. The letters in front of you are that test, still set as type.
Through.