Find the real bottleneck.
Talk to the people doing the job. Trace the decisions, delays and exceptions. Define one outcome worth improving.
I’m Ilhan Irem Yuce. I work directly with organisations to turn a real problem into a working system. Close to the workflow, close to the people who use it, and responsible for what happens after the demo.
A problem looks different when you stand inside it. The work starts there.
A licence does not know where your work gets stuck.
Malta’s national ChatGPT initiative brings AI to more people. That matters. Inside an organisation, the harder questions come next: Which workflow is worth changing? What data can the system see? Where must a person remain in control?
I start there. Not with a catalogue of models. Not with a promise that every task should be automated. With the work itself.
Read about Malta’s national initiative ↗The forward-deployed idea is simple: get close enough to the operation to see what a presentation cannot. My role combines fractional leadership with hands-on implementation. The title matters less than the work.
Talk to the people doing the job. Trace the decisions, delays and exceptions. Define one outcome worth improving.
Choose the right tools, connect the necessary data, and test the workflow against real cases rather than a polished demo.
Give people a way to review, correct and stop the system. Watch what breaks. Improve what actually gets used.
If the right answer is a simpler process, not another agent, that is the answer I will give you.
News Beast and Gambity are working examples of my multi-agent approach. Their AI journalists have distinct beats, source pools and editorial responsibilities. The system has to decide what belongs where, reduce duplication, pass work between roles and keep a human able to inspect the result. That is architecture, not a collection of prompts.
Each journalist has a job, an audience and boundaries. Different work calls for different instructions and sources.
Research, drafting, review and publication connect. A good story does not disappear between disconnected tools.
Automation can prepare the work. Editorial judgment and accountability do not disappear with it.
The work is easier to judge when you can open it yourself. These are live products, not hypothetical use cases.
A multi-agent newsroom organised around distinct editorial roles and daily publishing. Look at the output, not a diagram of the pipeline.
Open News Beast ↗A separate editorial system with its own journalists, beats and audience. The same operating discipline, applied to a very different subject.
Open Gambity ↗ChatGPT, Codex and agent workflows can each do different work. Sometimes one is enough. Sometimes another provider is the better choice. I choose the architecture after understanding the problem, the security boundary and the team that must live with the result.
My OpenAI Select Partner relationship is a route into that ecosystem, not a claim that I work for OpenAI or that every customer needs an Enterprise contract.
Read my Claude vs ChatGPT verdict ↗A partnership announcement says who I work with. The writing shows how I think. The products show what I build.
No group chat. No account manager. A private line to the person who will look at the problem with you. If I can help, I will tell you how. If I cannot, I will tell you that too.
Talk to Ilhan ↗