A shortlist of where AI is worth it
Your workflows ranked by value, risk and how ready the data is, with a note on how much freedom the model should have in each.
All services
AI-first applications
Work out where AI belongs in your product.
We help you decide what the model should do, what ordinary code should keep doing, and what it takes to turn a demo that impresses into a feature people can rely on.
This is for you if
- You have an AI prototype that demos well, and you are not sure it is safe to ship.
- Someone asked for “AI in the product” and nobody has said what it is for.
- The model sometimes gives wrong answers and you have no way to catch them.
- You are about to spend real money on AI and want a second opinion first.
What we look at
Does the AI have a clear job, or was it added because it could be?
Which answers need a source, a permission check, an exact calculation or a person to approve them?
After launch, can you tell when it gets something wrong, and can you fix it?
What does the user see when the model is down, unsure or wrong?
What you get
Rules for what the AI may do
What it can read, what it must cite, what needs a person to confirm, and what happens when it fails. Written so engineers can build it and managers can sign it off.
A plan from prototype to live
The stages, how you will test each one, who owns it once it runs, and what it will cost before you scale it.
How it works
Write to us
Tell us what keeps going wrong. A few rough lines are enough.
Agree the scope
We talk it through, then send a short note: what we will look at, how long it should take and what it costs.
We do the looking
We read the system, talk to the people doing the work and follow real cases from start to finish.
You get the findings
A written report in plain language, a plan in order of priority, and a session to walk your team through both.
Where this comes from
We build and run our own products. The advice comes from that work.
- Garden HRMSOur HR data warehouse, built to be AI ready. Vector search sits inside the database, and AI proposes changes for a person to approve. In the hiring app it reads CVs and explains its shortlists.
Memoirs of MidgardA live browser game we built with AI assistance. It taught us that building fast with AI still needs design, review and testing done by people.
Truyền ThuyếtA 69-chapter novel and gamebook written with AI under human direction, with the continuity checks and editing passes we had to design ourselves.
Tell us what keeps going wrong.
You do not need a tidy brief. A few lines about the system, workflow or delivery problem are enough to start.