Context, what I did, what happened. Numbers where I have them.
Creating an AI-first engineering culture while the product stayed live
A legacy CX platform, rewritten Elixir to Python and Ember to React by a team that shrank from 70+ to under 10, with almost no disruption.
CONTEXT
zenloop was a live customer-experience platform with a decade of Elixir and Ember behind it, a 70+ person org, and a saas.group mandate to become profitable fast.
The usual playbook, hire more and rewrite in a year, was not available. The rewrite had to happen inside a shrinking team while customers kept using the product.
WHAT I DID
I wrote the tooling before I asked anyone to change how they worked: maggy, an agentic harness with 13-tier model routing, plus the agent-driven migration that drove the rewrite. Every PR went through a multi-model council review.
Then I made agentic development infrastructure: ADRs, specs, TDD, functional and visual validation, PR gating, parallel isolated worktrees. The process was encoded so agents could build safely and people could review instead of type.
OUTCOME
The platform moved to Python, React, ClickHouse and Postgres while it stayed in production. The team went from 70+ to under 10. An independent 2026 review rated the company as leading in AI enablement.
The harness left with me as open source: maggy, 706 stars and 887 tests, with a real migration behind it.
I build the hard modules, then hand ownership to the team so there is no single-person dependency.
Going from enterprise sales to product-led growth
From winning AXA and Allianz on full RFPs to shipping a gateway developers adopt from a dashboard, and what had to change in me along the way.
CONTEXT
At eKomi I repositioned a smaller product against Qualtrics and Medallia, won AXA and Allianz, rolled out 15 countries in 3 months and ran the insurance P&L. Enterprise motion: long cycles, RFPs, executive sponsors.
srooter is the opposite: a commercial LLM gateway that sits between Claude Code, Codex, Cursor and 10+ providers. Developers try it in an afternoon or never.
WHAT CHANGED
The instinct I had to unlearn was the long qualification cycle. In PLG the product qualifies itself, so the first self-serve funnel and the pricing had to carry the weight the sales conversation used to carry.
What carried over is that I can still talk to the exec and whiteboard the architecture in the same meeting. That combination sells a developer tool to the people who sign for it.
OUTCOME
srooter routes across 10+ providers with per-team budgets, policy and audit, and it runs my own daily work. Integrations that were quoted at $40k+ shipped in days once the gateway was in place.
I can talk to the exec and whiteboard the architecture in the same meeting.