A working prototype
Not a mockup and not a demo video — a running system that takes your real inputs and produces real output. You keep it, whatever you decide next.
Working AI prototype on your data — 2–4 weeks.
Instead of a slide deck about what AI might do for you, we build the thing and run it on your own data. In two to four weeks you have something your team can click through, argue with, and decide on — before you commit a budget to a full build.
What you get
Three things at the end of the two to four weeks.
The point of a proof-of-concept is a decision you can defend. So you finish with a working system, an honest read on what it takes to scale it, and everyone in the room having seen it run.
A plan to scale it
What it would take to move from prototype to production: the gaps we hit, what stays fragile, the effort estimate, and what it will cost to run.
A demo to your team
We walk your team through it live and answer the hard questions — so the go/no-go call is made by people who have actually seen it work.
Demos we've built
What a prototype looks like in practice.
These three are internal demos we built for ourselves — not client projects. We show them because they are the fastest way to see the shape and depth of what lands in two to four weeks.
Internal demo
Voice to structured tasks
You talk; the system turns loose speech into properly structured tasks — titles, descriptions, assignees, estimates — and files them end to end into a real task tracker. It runs live, not against a mock.
115 automated tests.
Internal demo
Booking bot for a service business
A Telegram bot that handles booking in plain conversation: reads what the customer wants, checks the schedule, confirms the slot. Built for a salon-style business where every booking is a phone call today.
106 automated tests.
Internal demo
Call quality assurance
Analyses recorded customer calls against a scoring rubric and returns a per-call breakdown — what was handled well, what was missed — so a manager reviews the outliers instead of sampling blind.
113 automated tests.
None of the three has a client or a pilot behind it. When we do have client results, we say whose and what changed.
How we build this fast
AI agents doing the typing, engineers deciding what gets typed.
Two to four weeks sounds fast because most of the writing is done by AI coding agents running under an engineer's supervision. We plan the work, the agents produce the code and the tests, and an engineer reviews, corrects, and owns every line that ships.
On one of the demos above we measured it: about two hours of agent wall-clock time against a human estimate of roughly 19 to 24 hours for the same scope. That ratio doesn't hold for everything — plumbing, glue, and well-covered patterns go fast; the parts that need judgement still take a person the time they take. It is why a prototype fits into weeks rather than a quarter.
~2 hagent wall-clock on one demo, against a ~19–24 h human estimate for the same scope
The same agent-assisted approach migrated a 44K-line legacy system in our R&D — see the cases →
How we work together
Pick the engagement that fits the job.
A proof-of-concept is small and well defined, so it usually belongs in the first option. The other two exist because not everything does — and we would rather route the work honestly than promise a fixed price on something nobody has scoped yet.
Fixed scope
The usual fit for a proof-of-concept — one prototype, one audit, one pipeline. We agree on the outcome before we start, estimate the work upfront, and do our best to stay within 20% of the estimate.
Discovery first
For bigger or fuzzier projects: a paid Discovery (3–5 days, $1,500–3,000, fixed price agreed upfront) → artefacts, plan, and a priced proposal. You keep the artefacts either way.
Dedicated capacity
For ongoing product work: a monthly retainer of reserved days for features and support — how we run our longest engagement today.
On price. We estimate each feature upfront and do our best to stay within 20% of the estimate. If something turns out bigger than we thought, you hear it while there is still time to change the plan.
Bring one idea. We'll show it running on your data.
Tell us the one thing you keep saying AI should be able to do in your business. We'll come back with what a two-to-four-week prototype of it would cover — and what it wouldn't.