For engineering orgs at Level 2 of 5: on site, 3 months
AI adoption in 3 months: Level 2 to 3, from talent to proof.
For leaders without the time or the in-house expertise to run the change themselves. We work on site, in your company, to move things forward: Petr Glaser builds the AI-native SDLC with your teams, Marian Kamenistak brings the people along and reports to your top management every month. Miss a monthly goal and that month is not invoiced.
Our differentiator
No prepayment. We invoice monthly, and only when we hit that month's goals.
Why? Because we're confident in what we deliver. Marian has a computer science background and 35+ fractional CTO engagements behind him. Petr Glaser of BleedingDev works on AI agents and AI in engineering.
At the start we agree on a KPI for every month, tied to the value for your business. We invoice after the month, and only when that month's KPI was met. Miss it, and there is no invoice for that month. The risk sits with us.
- No prepayment
- Goals tied to value
- Missed month, no invoice
Trusted by engineering leaders from
Who it is for
- CTOs and VPs of Engineering whose teams already use AI, and nobody can say what it bought.
- Companies with an AI owner and trained people, Level 2, and no proof on the P&L yet.
- Leaders who want it done properly and have no spare quarter to run it themselves.
Who it is not for
- Teams still picking their first AI tool. That is Level 1: start with AI1.
- Companies that want a slide deck, not a change in how teams ship.
Month by month
What happens in each month when you start at Level 2.
Level 2 means the people are in place: an AI owner, trained engineers, AI in daily use. What is missing is proof. Nobody can say what AI cost last quarter and what it bought. Three months later you can, and every month ends on a KPI you can check.
- L1Nothing in production
- L2TalentYou start here
- L3ProofAfter 3 months
- L4AI-native
- L5Won the race
Baseline and the first harness
Technical, with Petr
- An audit day with your team, on site: repos, pipeline and how AI is used today.
- A baseline before anything changes: lead time from idea to production per SDLC phase, review wait as its own phase, change fail rate and rework.
- Agent rules in the pilot teams’ repos, kept short. Anthropic recommends a CLAUDE.md under 200 lines.
- A spec-before-prompt threshold: a spec for larger changes, none for bugfixes and config.
People, with Marian
- A strategy day: the timeline, with a technical and a people stream and a KPI for every month.
- Two pilot teams chosen. Petr pairs with their tech leads in the room.
- The first monthly report to top management, and the message they send to the developers.
Invoiced when
- A baseline published for every team.
- Agent rules merged in the pilot repos.
- The timeline with monthly KPIs, signed off by you.
A review gate and the next teams
Technical, with Petr
- A code review agent does the first review on every merge request in the pilot teams. People approve.
- A review gate and hooks in CI, a sandbox and branch protection. The boundary sits outside the model.
- A spec-driven flow for changes above the threshold.
People, with Marian
- Rollout to the next teams through tech-lead pairs, so AI-native work does not stay in one corner.
- AI fluency checked with a real demo, not a quiz.
- Optional: the talent evaluation, a traffic light per person.
- The monthly report, and a developer engagement plan with top management.
Invoiced when
- Over 30% of merge requests in the pilot teams AI-assisted, green in the ELC framework.
- Review cycle time no worse than the baseline.
- AI cost per shipped unit tracked every month.
Proof: one number CFO and CTO both sign
Technical, with Petr
- The harness in every team.
- Cost per AI-assisted merge request and per resolved ticket, on one dashboard.
- Change fail rate of AI and non-AI work, side by side.
People, with Marian
- A QBR with CFO and CTO: what AI cost last quarter and what it bought, as one number.
- The strategy update, and the next quarter’s scope toward Level 4: agents in the build pipeline, evals in CI.
Invoiced when
- AI cost per shipped unit tracked, with a threshold agreed at the QBR.
- Delivery stability of AI work within 20% of non-AI work.
- Over 70% of engineers ship a feature with AI, shown in a demo.
What changes
| Before | After |
|---|---|
| One team ships with agents. The other teams wait, doubt it, or quietly refuse. | The technical and the people stream move every team, so adoption does not stall in one corner. |
| Developers feel faster. Nobody can show it. | A baseline in month 1, and by month 3 one number that CFO and CTO both sign. |
| Top management hears about AI from vendors and LinkedIn. | A monthly report from Marian: what moved, what did not, what comes next. |
| You pay for days, whatever happened. | You pay for a month only when its KPI was met. |
How we work
- On site, in your company. AI adoption moves in the room with your teams, not in an online call.
- Start, 4 person-days with Marian and Petr: an audit day with your team and a strategy day that produces the timeline, with a technical and a people stream and a KPI for every month.
- Petr Glaser, senior AI engineer, 1.5 days a week: the technical build and adoption in the teams, pairing with your tech leads.
- Marian Kamenistak, 2 days a month: people and communication, the strategy update, and the monthly report to top management, including help with how they talk to developers about AI.
- Renewal by the quarter. Scope and KPIs of the next quarter are agreed at the end of month 3.
Price
Invoiced monthly, after the month, and only for a month whose KPI was met.
A: 3 months, recommended
29,500 EUR / quarter
29.5 person-days at 1,000 EUR. Month 1 is 12,500 EUR including the start, months 2 and 3 are 8,500 EUR each. Invoiced monthly, after the month.
A month whose KPI was not met is not invoiced.
B: On demand
1,200 EUR / person-day
No reserved capacity, so a date can move or be declined, and we cannot commit to a monthly goal. The same 29.5 days cost 35,400 EUR.
Add-on: talent evaluation
1,200 EUR / person
Marian assesses each person against the AI-native way of working: a traffic light, green, amber or red, with a firm recommendation for every person.
All prices excl. VAT. Travel outside Prague at cost. Renewal by the quarter.
Why it works
We carry the risk.
Every month has a KPI in the timeline. Miss it and we do not invoice that month. We know what we are doing, so the risk sits with us.
Technical and people, together.
One team going AI-native while the rest refuse or move slowly is how this usually fails. Petr moves the code, Marian moves the people, in the same weeks.
Top management stays close.
Marian reports every month and helps your leaders talk to developers about AI, so engagement does not hang on one enthusiastic team.
Measured, not felt.
In a randomised study, experienced developers with AI were 19% slower and believed they were 20% faster (METR, 2025). We baseline before we change anything.
Who does the work
BleedingDev
The technical stream: builds the AI-native SDLC with your teams, 1.5 days a week.
Engineering Leaders Community
The people stream: communication, the strategy update and the monthly report to top management, 2 days a month.
Questions
How do you measure whether AI helps?
Not by how fast people feel. In a randomised study, experienced developers with AI were 19% slower and thought they were 20% faster (METR, 2025). We measure lead time from idea to production per SDLC phase, with review wait as its own phase, and add change fail rate and rework as the safety net, because higher throughput with AI tends to come with lower stability (DORA 2025). Baseline first, then each team against itself. We do not measure individuals.
What are the 5 levels, and how do you know we are at Level 2?
The ELC 5 Levels of AI Adoption: 1 nothing in production, 2 talent, 3 proof, 4 AI-native, 5 won the race. Each level has two or three KPIs with red, amber and green thresholds. The audit day checks your evidence against them, so the level is shown, not felt.
Do you use spec-driven development?
Above a size threshold. A spec for every small bugfix slows teams down and goes stale, so bugfixes and config skip it. We measure spec-driven work like any other area, because we have not found a controlled study showing it shortens lead time.
How do you keep autonomous agents safe?
We assume someone will talk an agent into something, so the boundary sits outside the model: hooks and a sandbox, its own branch with branch protection, network off or allowlisted, short-lived keys per workflow, and a budget and step limit per run. In one session an agent gets at most two of three things: untrusted input, sensitive data, the ability to change state or talk to the outside (Meta, Rule of Two, 2025).
How do agents get shared company context?
Agent Skills, a shared plugin marketplace in Git rolled out through managed settings, an organisation-wide CLAUDE.md, shared MCP servers, and common rules in AGENTS.md. Kept small and owned: every skill has an owner, changes on an observed error, and gets cleaned up once a quarter.
What happens if you miss a monthly goal?
That month is not invoiced. The KPIs are fixed in the timeline from the strategy day, so both sides know in advance what counts.
Is this online?
No. Adoption moves in the room, so Petr and Marian work on site with your teams. Travel outside Prague is billed at cost.
What is the talent evaluation?
An optional add-on at 1,200 EUR per person. Marian assesses each person against the AI-native way of working and gives a traffic light, green, amber or red, with a firm recommendation for every person.
Start with a 30-minute call.
We look at where you are on the 5 levels and whether this quarter is the right one to move.