Ivan Popov is the founder of Inoku, an independent research lab working on how to make AI agents tell the truth about their own work. He started it in 2026 after seeing how often agents report success they didn’t achieve, and how rarely the usual checks catch it.
Background
Ivan has worked in web and systems engineering for more than twenty years, across backend architecture, platform performance, technical SEO, migrations and search-driven growth infrastructure. The engineering came first: years of building and running web applications, backend systems, data pipelines and production infrastructure. The search work grew out of it, as technical SEO on large platforms turned into a question of rendering, internal links, architecture and deployment behaviour rather than page-level tweaks.
He is the founder and principal advisor at IvanLabs, an independent platform and growth infrastructure advisory for revenue-critical web platforms, where he works directly with founders and engineering leaders on performance, migrations, architecture and search visibility. He also builds ProxyLax, an edge security platform on Caddy and the OWASP Coraza web application firewall, currently in pre-launch engineering.
Languages and tools, for the record: Python, Rust, PHP and TypeScript; Django, FastAPI, Axum and Zola; Linux, Docker, Caddy, Cloudflare, PostgreSQL and Redis. Inoku runs on the same stack, and this site is built with Zola and served by Caddy.
He is based in Larnaka, Cyprus.
What he works on at Inoku
- Ways to measure how often an agent’s “done” is true, on tasks it can’t have trained on.
- Ways to reduce false success claims, so that agents report what happened, including an honest “I couldn’t”.
- The research notes on this site, where methods and results are published, including the negative ones.
He writes every post on this site unless the byline says otherwise, with AI models as assistants across research, engineering, writing and marketing: Claude, GPT, GLM and Qwen among others, as the About page sets out. He checks every factual claim against its source before it is published and is responsible for every page. Each page says whether the thing it describes is built, in development or still research.
Why this problem
The short answer is that the usual checks don’t catch it. An agent’s own transcript is written by the agent, and a reviewer who reads only that transcript sees only what the agent chose to record. Ivan’s view is that the record of what an agent did should sit outside the agent’s reach, and that an honest “I couldn’t” should count for an agent rather than against it. Inoku exists to test that view in public, with numbers.
Elsewhere
- LinkedIn: linkedin.com/in/iappv
- GitHub: github.com/aimagician (personal) and github.com/IvanLabs-com (IvanLabs). Inoku’s own repositories are at github.com/inoku-labs; they hold READMEs and roadmaps until the method is published.
- IvanLabs: ivanlabs.com, the advisory practice
- Email: hello@inoku.com
For press enquiries, research collaboration or anything else, use the contact page. For a short description of the lab and its mark, see About Inoku.