one/2/work builds the parts that carry weight: backends under load, apps on every screen, and AI that runs in production with a budget, an evaluation set and someone on call. A team of 50+, for products that cannot afford a bad Tuesday.
01
Three kinds of work. Each has a failure mode we have seen often enough to name it.
01
Products where the load graph only goes up
02
Teams shipping to phone, desktop and watch at once
03
Companies whose AI pilot impressed everyone and never shipped, and companies that need to add AI to a system already under load
02
How a typical project runs, from the first letter to the day we hand over the pager.
Week 0
Not a sales call. A written first read of your system within three business days, with the questions we would bring to a call.
You hold: a first opinion, in writing
Weeks 1–2
We read the code, the incidents and the graphs, and talk to whoever gets paged. Two people, fixed price, your repository.
You hold: a plan, ranked risks and a price
Week 3
Every two weeks something demonstrably works, demoed live, with hours reported against the plan.
You hold: working code and a burn chart
Week 8
Load rig at twice the forecast peak, failure drills, a rollback someone has actually performed. Launch day is boring on purpose.
You hold: a load report and a runbook
Week 14
Staged rollout behind flags. The same team that built it is on call for it.
You hold: a system that survives a Tuesday
Week 15+
We keep the pager, or we train your team and leave the runbooks behind. Leaving us takes a calendar invite, not a migration.
You hold: everything, in your repository
Weeks are indicative. The order is not.
03
No bench, no subcontractors. The rules below are why clients stay.
50+ people, four practices
You meet the team before the contract.
If you would not hire them, we do not staff them.
Nobody is rotated off mid-project.
Whoever designed the system is still there when it is time to scale it.
Every decision is written down, in your repository.
Nothing lives in a chat thread.
We say no before the contract, not during it.
If the deadline or the budget does not fit the work, you hear it in week zero.
The team that built it is on call for it.
Incentives stay honest when the author is the one woken up.
No account managers.
You talk to the people doing the work. That is the whole layer.
A model is a service, not a demo.
It gets an SLO, a cost ceiling, an evaluation set and a pager, or it does not ship.
After twelve months you may hire them.
With a transfer fee agreed up front. Better that than pretending the option does not exist.
04
Four ways to buy the same team. Prices are anchors; the number you sign comes out of the audit sprint.
€8k
fixed
Two weeks, two people, your repository. A plan, ranked risks and a price. If any of the three disappoints you, it ends there and you keep the documents.
Fits when you need to know what you are dealing with
€8.6k
per person · month
Five to ten people owning a product line end to end, with a named lead. Monthly, thirty days' notice.
Fits when the work is a product, not a project
€95k
per phase
A defined launch with a hard date. Phases with a fixed price each, once the audit sprint has told us what is inside.
Fits when there is a date and a scope
€54
per hour
One to three named people inside your own team, on your rituals. Weekly timesheets.
Fits when you have a team and a gap
Every model includes a lead and QA without extra billing, a security review before each release, an NDA and a GDPR data-processing agreement, and full IP transfer on payment. Rates assume 160 billable hours a month.
05
Conservative on purpose: tools you can still hire for in two years. Anything unusual arrived with a written reason and an exit.
06
All answers open. Nothing behind a click.
It is simple. OneTwoWork doo is an EU-compatible counterparty: contracts under your law or ours, invoices in EUR, VAT handled, a GDPR data-processing agreement signed with the MSA. CET hours cover the whole EU working day.
That is most of our work; the audit sprint exists for it. We have never walked away from a codebase, though we have twice recommended a rewrite the client did not want to hear about.
Per phase, yes, once the audit sprint has told us what is inside. Fixed price on an unexamined system is a guess dressed up as a commitment, and one of us always pays for it.
Then you paid for headroom you did not need. That is why the first phase is sized to the load you can actually forecast, and the rest stays a documented plan.
Rarely, but it happens. Every part has two people who can carry it, decisions are written in your repository, and the handover period of a replacement is not billed.
Because a pilot answers one question and production answers thousands, some of them hostile. Between the two sit an evaluation set that tells you when the model got worse, a cost ceiling so a traffic spike does not become an invoice, a fallback for when the model is wrong, and someone who is paged when it is.
07
Tell us what it is and what is wrong with it. You get a written first read within three business days.
Before you write — a quick fit check
Write to us if
Probably not us if
If you are unsure which column you are in, write anyway. Saying no quickly is also a service, and we do it in writing.