Your AI transformation partner.
Seeko maps how your company works, identifies where AI should change it, and helps you make it happen.
The Two‑Week Audit.
The cheapest way to find out which way your AI work is going to fail before you spend a year proving it.
A written diagnosis & a ranked list of bets.
Most AI work fails in the first six weeks, in the same handful of ways: wrong problem, wrong data, wrong people in the room. The audit names yours, on paper, with the parts you'll disagree with surfaced first.
Roughly one in three audits ends with us recommending you don't hire us. That's the point.
Start an auditEngagements.
Three ways to keep going. Most teams pick one. Some pick none: the audit is the deliverable, and that's fine.
Sprint
For one bet that's clearly worth taking, with a clean owner.
- One workstream, six weeks, fixed scope
- One operator embedded; your team owns the code
- Weekly written update; one steering call
Quarter
For two or three live bets that share infrastructure.
- Embedded operator, four days a week
- Eval harness + production playbook
- Bi-weekly steering with the exec sponsor
Steering
For leadership teams that want a second pair of eyes, ongoing.
- Monthly working session with Haroon + lead operator
- Async review of your team's biggest open bets
- First call on capacity when you do need a sprint
Principles.
Four ideas we re-read on the first day of every engagement.
The first deliverable is always the diagnosis, in prose. Slides come later, if at all. The teams that fail with AI are almost always solving the wrong problem with admirable rigour.
We optimise for the smallest experiment that proves the thesis. If the smallest bet is 'ship nothing for two weeks and write the eval first,' we'll say so and we'll mean it.
By week six of any sprint, your operator can run the thing without us in the room. We write what we build to be handed off, not held hostage. We keep our own tooling out of your stack unless you ask for it.
Every report has a kill list. We will tell you, by name, which projects to wind down and roughly when. The hard work isn't picking the next bet; it's getting honest about the last one.
Deliverables.
Things your team can use after we're gone: not slideware, not a Notion graveyard.
Written diagnosis
~25 pages. The picture as we see it, with the parts you'll disagree with surfaced first.
Ranked bet list
Every live and proposed bet, scored on cost, payback, and confidence. Plus a kill list.
Eval harness
The smallest test suite that catches the regressions you actually care about. In your stack.
Production playbook
How this thing is run on a Tuesday at 3am. Owners, alerts, what's allowed to fail.
Weekly update
One page, every Friday. What moved, what stalled, what the operator is asking for next week.
Monthly read-out
A short memo to the exec sponsor on the open bets: what's compounding, what's drifting.
Proof.
Numbers from the work and the broader practice. Happy to walk through specific engagements on a call.
They told us to kill the thing we were proudest of. It was the right call, and nobody else in the room would say it.
FAQs.
Because two weeks is small, and a year on the wrong bet is enormous. If we already know each other and the bet is obvious, we can skip it. Say so.
Yes. Mutual NDA before the first call if needed. We've worked under stricter ones; ask.
If you're under 30 people and pre-Series A, the audit usually isn't the right shape. We'll point you at a sharper-edged firm or a single advisor.
He runs every audit. On sprints and quarters, the lead operator runs the day-to-day and Haroon stays on the steering cadence.
Contact.
Tell us who you are and what you're stuck on. We write back within one business day with whether we're the right fit.
Read by a human.
If we're not the right fit, we'll say so on the first call and point you at someone better suited.
Name, work email, company, and a short note are enough. Website is optional, but helps us come prepared.