Guides
·
12 Oct 2026
·
8 min
read

Leo Marques
Founder, Otto

Most AI projects die in a slide deck. Ours ship in two weeks — not because we cut corners, but because we pick a small enough problem and refuse to let it grow.
Here’s the exact playbook we run with every client, from the first call to an agent running in production.
Week 1 · Find the boring
We spend two days shadowing the team. Not interviews — actual screen-sharing while people do their job. The best automation candidates are always the tasks nobody mentions because they’re too dull to talk about.
Repetitive, rule-based, high volume
Clear input and output
Mistakes are cheap and easy to spot
“If a task needs a meeting to explain, it’s not the first thing to automate.” — Daria Koval, design lead
Week 2 · Build, test, hand over
By day eight there’s a working agent running on real data in a sandbox. The team reviews every output for three days, we fix what breaks, and on day fourteen it goes live with a human approving anything risky.
Two weeks later we look at the numbers together and decide what to automate next. That’s it. No roadmap theatre.


