Design
mock-up → codeDesign system, interfaces, brand. From mock-up to a component in code.
Product team · B2B and consumer · RU / EU
Six A-players. We close a year-long roadmap with you under a hiring freeze: one process in production in two weeks, the whole product in eight. We’ve shipped at Lenta Online, Utkonos, Atom and BridgerPay.
From discovery to traffic, community and brand after launch. Any stage or all of it — one contract, one pod.
Design system, interfaces, brand. From mock-up to a component in code.
Web, mobile, integrations. Production every week.
Your perimeter, your access. Observability, backups, audit.
Events, video, live streams — from the venue to air.
Social, media, community and brand after launch. Plan, production, metrics.
01 — where the team has shipped
15 companies
Lenta group revenue, 2025
BridgerPay payment network
invested in the Atom EV project
How big these clients are. Public sources.
Our own numbers on this page: 15 cases shipped to production · 6M+ users at Eatr · 4M rps of Hyundai Mobility telemetry.
02 — method
4 rules · visible from week one
Not promises. An operating policy you can hold us to — every rule shows up in week one.
A short screen recording every evening. No status meetings.
The unit of delivery is a week, not a sprint.
You can stop on any Friday.
Nobody splits attention between clients.
AI is in the machine, not in the deck. That’s why six people ship what you’ve budgeted a department and a year for.
What stays after us: code, documentation and a design system. Your team carries them on without us.
AI adoption scale
We don’t run training sessions — in every meeting we just show what we use and how. The teams next to us take that into their own routine.
On the scale by Boris Cherny — the creator of Claude Code — teams land on steps 1–3 after working with us. We’re on the fourth and already moving to the fifth: the scale just doesn’t go there yet.
example · iSmart: most of the team at step 3, the rest at step 2 and moving up
client teams after us
us · moving 4 → 5
step 0
Household
the odd chat
AI means “Alexa, what’s the weather?” and the odd chat. None of it at work: either it isn’t allowed in, or nobody has tried.
step 1
Assisted
you + one agent
One agent, one session, you read every edit yourself. Still, an afternoon’s task now fits between meetings.
step 2
Parallel
orchestrator
Five to ten agents at once, each on its own branch. They run tests, lint and review themselves — you look at finished diffs.
step 3
Supervised autonomy
manager of managers
Agents write nearly all the code and pick up work themselves. “Did you read the code?” turns into “what context was it missing?”
step 4
Production line
thousands of agents · outcome owner
Production is built around thousands of agents and the structure itself changes: people set the tasks, accept the result and own it — not the lines of code.
step 5
AI-native company
agents tune the line itself
The company is built around agents from day one: processes, roles, economics. For every project the agents tune the production line itself.
step 0
0agents
Household
the odd chat
AI means “Alexa, what’s the weather?” and the odd chat. None of it at work: either it isn’t allowed in, or nobody has tried.
step 1
~1agents
Assisted
you + one agent
One agent, one session, you read every edit yourself. Still, an afternoon’s task now fits between meetings.
client teams after us · 1–3
step 2
~10agents
Parallel
orchestrator
Five to ten agents at once, each on its own branch. They run tests, lint and review themselves — you look at finished diffs.
step 3
~100agents
Supervised autonomy
manager of managers
Agents write nearly all the code and pick up work themselves. “Did you read the code?” turns into “what context was it missing?”
step 4
~1000agents
Production line
thousands of agents · outcome owner
Production is built around thousands of agents and the structure itself changes: people set the tasks, accept the result and own it — not the lines of code.
us · moving 4 → 5
step 5
AI-native
AI-native company
agents tune the line itself
The company is built around agents from day one: processes, roles, economics. For every project the agents tune the production line itself.
Boris Cherny · Steps of AI Adoption · Jul 16, 2026
03 — timing
5 task types · up to 26 wks
Not “faster”. How many weeks and how many people, set against what’s already in your plan.
| Task | Market norm | Narhi | Team | |
|---|---|---|---|---|
01MVP or a new product | 24 wks | 8–12 wks | 4 ppl | |
02A major feature in a live product | 12 wks | 4–6 wks | 3 ppl | |
03Redesign with a new design system | 16 wks | 6–8 wks | 3 ppl | |
04Integration or migration | 8 wks | 3 wks | 2 ppl | |
05Automation on top of an existing process | 6 wks | 2 wks | 2 ppl |
“Market norm” is the timeline clients brought us in their plan before the start. Ours come from the projects on this page; under NDA — no name, but the number stays.
04 — pricing
4 steps · entry 0 ₽
A four-step ladder. You can stop any step on any Friday.
One process, live in production. The metric is agreed up front.
Art director, two engineers, a designer. Weekly delivery, weekly billing.
The whole product: discovery, design system, production, infrastructure.
A pilot usually fits under your internal no-tender threshold. The exact figure is in your procurement policy.
Prices are guides as of September 2026. Final pricing after the review.
You can stop on any Friday.
Paid weekly. Code rights transfer as each week is paid.
05 — work
15 cases · 14 industries
Every case: one task, one result, duration and team size.
car subscription · Russia’s first online car-subscription service
4M rps of telemetry and terabytes of traffic a week on own hardware. Apps live in the App Store and Google Play. The service went from 1,600 to 5,000+ clients in a year (Hyundai’s public figures).
consumer · recipes & nutrition
Production stable within two weeks. 0 seconds of downtime at the 10× New Year peak. A new meal-plan engine in four months. 6M+ customers.
e-grocery · merger of two products
One codebase in production after 3.5 months. Several stable releases in a row — then handover to the in-house team.
04electric vehicle · the car’s digital interfaces
The visual and UX logic of the car’s key interfaces is in place. The AR HUD and the yoke with a touch display are part of an HMI concept with no equivalent on the mass market. Design, engineering and production worked as one chain.
06 — objections
4 questions · 4 answers
The ceiling is set by architecture, not by headcount. The team has run products at Lenta Online and Atom — hundreds of staff and millions of users — and held 4M rps of Hyundai Mobility telemetry on own hardware.
You see a demo every evening and code in production every week. Payment orchestration at BridgerPay and oncology diagnostics leave no room for shallow work.
Code rights transfer to you as each week is paid. What stays: the repository, documentation and the design system.
We work inside your perimeter: your repositories, your infrastructure, your access. That is how we handled patient data inside a hospital’s secured perimeter and passed a children’s personal-data audit for a state programme.
07 — people for you
1 filter · 2 ways
Everyone here was several times faster than the rest of their previous team. That kind of filter can’t be run by hand at volume, so we turned it into a tool and a service.
Our in-house agency: we find the ones who are several times faster than the rest of their team. The same filter we used on ourselves. That is how a hospital got its team for an early cancer-diagnosis project.
discuss→Our AI hiring system: plugs into your ATS and job boards and does the candidate breakdown — the choice stays yours.
open↗08 — entry
45 minutes. No forty-slide proposal.
We’ll find where your bottleneck is and what we’d do in two weeks. Free, with no presentation about us.