INDEPENDENT THINKING. PRACTICAL AI.
Useful systems.
Capable teams. Less theatre.
I’m Gyula Halmos. I help teams turn AI possibilities into real workflows—with strategy, agent engineering, and hands-on workshops.
01 / Context
Understanding the business goals, go-to-market strategy and product vision matters more to me than which model we use. That context tells us what is worth building—and what a good result actually means.
— Gyula
02 / Capability
I want teams to approach problems AI-first. That means designing the work around what AI can do from the start, rather than adding a tool to an unchanged process. The goal is an AI-native way of working.
— Gyula
03 / Human judgement
High-level strategy and design taste stay with people. When building a UI, people shape the experience and its details; agents write the code. I step into technical decisions when they affect those higher-level goals.
— Gyula
01 / A PRACTICAL APPROACH
Good AI work starts with your work.
Here’s where I can help.
Not every workflow needs an agent. Map the work, decision bottlenecks and ownership first—then design a team and a pilot that can prove useful change.
Bring models, tools, and your real working context together. Design for evaluation, sensible boundaries, and a human in the right places.
Build capability that stays after the session. Use real work and ambitious challenges to change how the team delegates, verifies and learns—not just which tools it opens.
MY VIEW / AI-NATIVE TEAMS
Start with the business goals, go-to-market strategy and product vision. Then design an AI-native way to deliver them—not a familiar process with AI added at the end.
MEASURE WHAT CHANGESLead time. Review waiting. Rework.
Not prompts sent or lines generated.
Small teams own the outcome and consider AI from the start. Agents implement bounded slices; people set direction and judge whether the result serves the business.
Business goals, go-to-market strategy and product vision come before model choice. Carry that context into shared instructions and skills, and improve it as the team learns.
People shape product decisions and UI details; agents write the code. Automated checks handle routine verification, leaving human attention for choices that affect intent, quality or risk.
Expand autonomy as checks earn trust. Permissions and CI enforce the boundaries—not prompts. Human attention goes to strategic choices and technical details that change the intended outcome or risk.
A challenge the old way can’t meet makes room for a new one. Missing the target is allowed; skipping checks or working longer isn’t the answer. Keep what works as shared practice—not a performance quota.
02 / THE PLAYGROUND
ILLUSTRATIVE DEMO · NO LIVE AISee how context, agents, and human judgement fit together. Three small examples. A few deliberate boundaries.
Every input is synthetic. Nothing is sent anywhere.
01 / THE INPUT
A fictional team is exploring better internal knowledge search. The scope is deliberately small.
QUESTION
How should we scope a knowledge-search pilot?
[A] People search across several approved documents.
[B] Restricted content must stay restricted.
[C] A pilot needs an owner and a baseline.Input ready. Explore at your own pace.
Good automation knows when to stop.
Local demonstration only. Real systems need server-side permissions and review controls.03 / SELECTED WORK
From an open-source agent desktop
to automation in the physical world.
Craft AgentsAt Craft Docs, I helped build Craft Agents: an open-source desktop app bringing AI, tools, and documents into a connected workflow.
At Yabune, the software meets the physical world. Smart-home systems and practical integrations built around the people who use them.
Discover Yabune Home (opens in a new tab)04 / THE PERSON BEHIND THE WORK
The interesting part starts
after the demo.
I’ve worked across cloud infrastructure, DevOps, connected homes, and AI agents. The common thread is automation that makes someone’s actual working day better.
Today, I’m an AI Ops Specialist at Polymarket. Previously, I helped build Craft Agents (opens in a new tab) at Craft Docs. Through consulting and teaching, I bring that practical perspective to other teams.
I care about who owns the workflow, how it fails, and whether the people using it can make it their own.
05 / SHARING WHAT I LEARN
Honest conversations about building with technology.
Including the parts that didn’t go to plan.
06 / YOUR NEXT USEFUL STEP
A workflow that’s stuck. A team that’s curious.
An idea worth making real.
Let’s find a sensible place to start.
Keep confidential information out. This stays in your browser until you choose to share it.
Nothing is submitted by this page.