Kraków, Poland
I took over a Partnerships team that had a reputation for missing its commitments. It is now the team that closes green, cycle after cycle. Before that I built a Poland engineering office from nothing to fifteen engineers.
15,000+
merchants served by the platform my team owns
0 → 15
engineers hired while building a new engineering office from nothing
99.5%
SLA compliance on order injection, held as partner traffic grew
1.7% → 0.8%
integration error rate, more than halved
7+ → 1
backend domains folded into a single external API for partners
92%
agreement between my AI pre-estimation workflow and the team’s planning-poker votes
The turnaround was not a process change. It was co-owning the roadmap with product instead of receiving it, refusing a commitment when the estimate said no instead of absorbing it quietly, and doing the hard work of breaking scope down with the team. Pushback only lands if you bring an alternative and the data behind it. A flat no makes you the obstacle.
AI inverted the ratio of reading to writing on my team, so anything that inflates length moves cost from one author onto every reader. Decision records lead with the value and the outcome in the first sentence and carry the decision in the first paragraph. No line-level code references. They rot on the next commit. Nobody gets to blame the model for a paragraph with their name on it.
It earns its place wherever the answer can be checked against something that already exists: reading a repository and reporting state, doing bulk work with a predefined shape, writing up what was decided. It fails on judgment about people, process and priority. My pre-estimation workflow matched the team’s own planning-poker votes 92% of the time, and the credit belongs to the context I fed it rather than to the model.
Burn-up over burndown, because burndown shows work added mid-sprint as work the team failed to deliver. A burn-up chart separates scope from progress, so added work reads as a scope change rather than a miss. WIP limits per column. Unplanned work logged as data rather than quietly absorbed. Do not optimise the chart. If a step in it reads as failure, that is a reading problem, not a delivery problem.
A caveat costs one sentence when you say yes. The same caveat during someone else’s crisis costs a negotiation with a person who is already blocked and already believes the answer was unconditional. That asymmetry decides most of the partner-facing calls I make, and it is why I answer with a narrower yes rather than a no.
I have closed incidents where the defect belonged to another team and the real finding was ours: production errors firing for hours with no alert, caught only because someone happened to look. Closing that coverage gap matters more than the fix. Every incident ends in a blameless review that produces a change, not a document.
Jul 2024 – Present
I lead the Partnerships domain: six engineers across two squads, owning the external API platform that partners integrate against.
Feb 2020 – Apr 2023
Built the Poland engineering office from nothing to fifteen engineers and owned technical strategy for those teams.
Jul 2015 – Feb 2020
Joined as the third engineer, working directly with the founders. Grew into leading an eight-person team as the company scaled.
2014 – 2015
A team distributed across Poland and Canada. I split my time between building software and running projects.
Head of Engineering, Director or VP. Kraków, distributed teams, any European time zone. The full history is on the CV.
