
Andrew Molyuk
Today at CGS Tower Networks: leading AI across the company, and the architect of its WebUI.
Building bunderlog ↗: Claude Code sessions that know what the project decided and check every change against it.
Looking for: Head of AI Engineering, CTO or VP R&D · consulting on agents in production · Israel (UTC+2), remote
Experience
I've been writing software since 1991, and for 20 of those years I also ran the teams that wrote it.
Video management and a CDN that brought every Israeli media service under one roof, and the R&D department that built them.
Floating-shift management and a mobile app that took the product into the US market, to enterprises of 65,000 employees and more.
Pulled the company back from collapse: hired R&D from zero, set up the process, delivered projects for Isracard, Visa, Mizrahi Tefahot and Tiv Taam.
Live voting on air: Hillary Clinton's debates, Ricky Martin's La Banda on Univision, ABC's Rising Star, at peaks of 3M requests a second.
Automatic ad-campaign tuning and anomaly monitoring: $17M in its first year, around 23B events a day, releases 75% faster.
The WebUI for every device the company makes, high-load central management for all of them, and agents that take tickets to pull requests in production.
Writing code since 1991, leading teams since 2006, helping AI write correct code since 2025.
How I Think About Systems
The principles I work by weren't picked up from a book; each came out of a place on the line above.
Infra follows requirements
Self-hosted or managed is a call per project: load, cost, compliance, and who will run it at 3 a.m. Neither is right by default.
LLM × Determinism
Keep model reasoning apart from deterministic checks. The model proposes; the system verifies.
Depth over breadth
One problem space since 1991: systems that aren't allowed to fall over. Tools change; the problems don't.
Agents are good at real work, mostly. What they're bad at is remembering: every session starts from zero, and that annoyed me enough to start bunderlog. I still write code most days and I don't plan to stop.
Technical Focus
AI-native engineering is the focus now; it stands on everything on the timeline above, software that had to stay up. The disciplines last; the tools orbiting them change. Hover or tap one to see what it covers.
Multi-agent pipelines that take real work from a ticket to a pull request, with scripts and tests deciding what ships.
High-load, high-availability systems that keep serving through failure.
Whichever cloud the requirements point to: managed services, the edge, or one rented box. Infrastructure as code throughout.
Docker from laptop to production, with orchestration sized to the load: a single host or a Kubernetes cluster.
Pipelines and hooks that make a release routine, not an event.
Interfaces for the people who run the system, in the framework the product already has.
Microservices and the migrations between them, built to be tested and run for years: Go where throughput matters, TypeScript where the product lives.
Systems that talk to other systems — HR platforms, network devices, issue trackers — each behind its own adapter, retried and audited.
What I'm Building
bunderlog is the current focus. The rest are live products I built and run; instrumenta went into internal production at the company I work for.
Four free tools for Claude Code that work as one. Every AI coding session starts from nothing; bunderlog gives it what a team gives a new colleague — one written record of what the project decided, kept as plain files in the repository, and the tools that keep every session, change and commit true to it.
Agent pipeline that takes a Jira ticket to an open pull request, unattended, at the company I work for.
Supply-chain check for ad systems: each seller the publisher doesn’t confirm, with the exact ads.txt line that fixes it.
Every endpoint an application talks to, in one container: readable from automated tests and reset between runs.
Cron, uptime, SSL, domain and port monitors with Telegram, Slack and SMS alerts, and white-label status pages.
› More projects (4)
Finds missed time-clock entries and routes them to the right manager before they pile up at month-end.
Onboarding, role changes and offboarding across a company’s systems, each event queued, dispatched and audited.
One-time links to a secret encrypted in the browser: the server never sees the secret, the key or the PIN.
Everyday developer tools in one place: IDs, hashes, JWT and X.509, regex, cron, CIDR, TOTP, diff and more.
Recommendations
“When Andrew came to our team, we were under terrible overload, but in just one month, he has helped us to establish a working process and we were able to breathe. … I would be very happy to work with him again!”
written on LinkedIn, 2016
“I had the opportunity to work with Andrew for over five years, and I can say without hesitation that he is one of the best leaders I have ever worked with. … I highly recommend Andrew for any CTO role in the tech industry, as he is an asset to any organization.”
written on LinkedIn, 2023
“Andrew is a calm and wise leader, and I can say only good things about working with him. … Working with Andrew, you feel both freedom and support: he does not micromanage, but is always available for advice, guidance, and a broader strategic view.”
written on LinkedIn, 2026
Let's talk
Open to opportunities and consulting: architecture, systems that have to stay up under load, and AI, from the first agents to a team that ships with them. Anything else worth building, too.
Telegram is the fastest way to reach me.