I build Saphan — the control plane that makes a fleet of coding agents answerable: who, on what evidence, allowed each change. Before that, twenty-nine years of carrier-grade security. This page is the fixed point; the work moves.
Twenty-nine years in internet infrastructure and enterprise security. It started in 1997 at a dial-up-era ISP — a Linux RAS platform and one of Poland's first free e-mail services — and ran through a decade of identity & access management: CompFort Meridian, Microsoft (Identity & Access, Central & Eastern Europe; Best TSP FY10 EMEA), CA Technologies, Oracle. As Microsoft's technology strategist for telco & media, Poland's main operators were my customers a decade before they became my partners. CISSP, on top of a stack of security and vendor certifications collected along the way (CompTIA Security+, Microsoft, Oracle, RSA security tracks).
Founder of SafeKiddo — a parental-control platform live at a European mobile operator since 2016, with national-scale deployments in the Middle East — and of the network-level DNS protection that followed it, deployed on carrier-grade DNS.
These days I run the company from Hua Hin, and so does the team: a fleet of coding agents that never asks for a day off, never remembers yesterday, and never ships anything without my signature. Everything runs on written orders, hard gates and an asynchronous protocol — the same discipline that once kept a distributed telco team honest, now applied to workers that read every word and remember none of them. That protocol became the product: Saphan Systems, founded 2026, U.S. patent pending. Saphan Studio is what I build now, and the fleet that builds it works under its own rules.
The control plane for fleets of coding and DevOps agents: written orders, evidence-bearing returns, human gates, and a signed record an auditor can verify without trusting the machine that produced it. The methodology, as software — saphan.ai. Self-hosted. Solo preview ships next; Team and Enterprise follow.
Around it, a family of add-ons. Each comes in two forms: a free, open-source tool that works on its own, and a paid add-on inside Saphan Studio for larger teams and enterprises.
A pull-request reviewer that may not post a finding without the evidence behind it. Built by my fleet in six days at the Open Agent Hackathon 2026; the effort report compares that with what a team of people would need. github.com/saphan-ai/proven
Reads a repository without changing it and tells you what is in it, with a citation for every claim. Offline, no account, and it never runs the code it reads. github.com/saphan-ai/discover
Agents develop and test against mocks instead of real systems, and their network opens to the mocks and nothing else. It started as TinyTelco, an operator in a box, and grew into mocks for any system an agent has to talk to.
Testing orchestration for teams that build with agents: tests stay in the team's own framework, every run is paired with Mocks Studio, and AI roles explore the app, explain failures and check visuals.
Network-level cyber-protection platform for mobile operators — a new product, built by a Saphan-governed fleet on top of a decade of partnership with telcos.
Six days, one person and a fleet of coding agents: Proven, an open-source pull-request reviewer that reports only what it can prove. Three model families on one commit, a secret one reviewer did not see, and a number I had to take back.
From the record — notes straight from the fleet that builds Saphan Studio, working under it. Started October 2026.
Shipping with agent fleets — a commercial telco-security product, live at a carrier six months after development started. Real incidents, the rules they produced, and numbers that survived a hostile audit of the git history. Six posts, July–September 2026, closed.
NVIDIA and NVIDIA Inception are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries.