About
Bartłomiej Szczygło
I have spent the last decade taking things apart to see how they work — game mechanics, renderers, and lately the systems that let a language model act on its own. That path ran through VR startups, Samsung, real-time 3D and VFX, and it landed somewhere specific: the unglamorous plumbing that decides whether an AI agent is a demo or a tool.
Most people writing about AI stop at the model. I care about everything around it — where the agent runs, what it is allowed to touch, what happens when a web page tries to talk it into something, and how the whole thing behaves at 3am when a provider goes down. That is what I build, and that is what I write about.
I also speak business. I ran client engagements, requirements workshops and process work long before agents were a category, which is why I tend to ask what a company actually gains before writing a line of code.
Right now
- ARP — Agentic Runtime Platform
- My main build: a runtime that gives agents persistent desks, secretless access to real accounts, and tools they can drive without a human in the loop.
- AI Engineer at PwC
- Contract work on the R&D and emerging-technology side: taking AI from research and proof-of-concept through to production multi-agent and LLM systems, MCP servers, voice pipelines and evaluations.
- Atos Software
- My own company. Custom AI automation for clients — document and invoice pipelines with OCR and LLMs, scraping, decision support, voice agents — plus the products under my own name.
What I actually do
Agent infrastructure
Orchestration, persistent workspaces, tool surfaces (browser, shell, SSH, HTTP), credential isolation, recovery when an agent gets stuck.
LLM operations
Model routing and fallback, cost control, local and open models, embeddings and RAG, OCR and vision pipelines, evaluation harnesses.
Agent security
Prompt injection defences, fail-closed guardrails, handling untrusted content, keeping secrets out of model context.
Business translation
Requirements workshops, process design, and honest calls about what AI will and will not do for a given company. Most engineers skip this part.
How I got here
- 2026 →Building ARP. Writing about agent infrastructure in public.
- 2025 →Founded Atos Software. Shipped LegalnyDeweloper, KayakBooking and KayakSystem, SuperLango. Started the PwC engagement.
- 2024CTO at Chatino.ai — custom AI chatbots and LLM automation for e-commerce.
- EarlierVR startups, then Samsung. Unreal Engine and Metahuman avatars, real-time 3D, VFX and display software. 20+ hackathons and game jams since high school.
- StudiesComputer Science at Warsaw University of Technology; Fachhochschule Salzburg; postgraduate sales and marketing at Asbiro.
Away from the keyboard
Triathlons, obstacle races, road and mountain biking, snowboarding, climbing, swimming, squash, and long walks in actual mountains. Occasionally 3D printing or cutting video with too many effects. I work in Polish and English, get by in Czech, and am slowly repairing my German.