I design AI agents that survive contact with a real support desk.

A prospect describes a mess — four hundred tickets a week, three systems that don't talk to each other, a team that spends its mornings resetting passwords. My job is working out which parts of that an agent can genuinely take over, which parts it shouldn't touch, and then building the demo that settles the argument.

Role

Product Manager

Focus

Enterprise AI agents, pre-sales

Integrations

ServiceNow, Atlassian, Freshdesk, Microsoft

Models

OpenAI, Gemini, Copilot, local LLMs

• Designing multi-step agents for ticket triage, knowledge search and approvals
• Building POCs against the prospect's own data, not a sanitised sample
• Working with engineering on what the agent can and can't be trusted to decide
• Writing the blueprints and docs that let someone else run it next time

Most AI projects fail on the boring parts: nobody owns the process, the data lives somewhere else, and nobody defined what "better" would look like. So I start there. The prompt is the last thing I write, not the first.

Starting point

The workflow as it really runs

Sequence

Discovery → design → validate

Who I talk to

IT ops, support leads, the budget holder

Counts as success

Time saved, errors avoided, fewer escalations

• Map the process as people do it, not as the org chart claims
• Find the 20% of tickets causing 80% of the pain and start there
• Ship the small, low-risk case first — trust is earned one use case at a time
• Measure the baseline before, or you'll never prove the after

Industrial engineering first, then four years on the implementation side for ITSM vendors. I spent that time watching enterprise software meet real teams — which is most of what I know about why automation either sticks or quietly gets abandoned in month three.

Education

Industrial engineering

Experience

4+ years in IT

Platforms

ServiceNow, ManageEngine, SolarWinds, Xurrent

Moved into

AI agents and agentic workflows

• Ran post-sales implementation across 50+ enterprise customers
• Moved from support into solution design, then into product
• Learned four ITSM platforms well enough to compare them honestly
• Now applying that to where agents actually fit in the same workflows

Planted aquariums, a guitar, a bike around Çanakkale, and games I lose too many evenings to. The aquariums are the most useful of the four — a planted tank is a slow system with delayed feedback and no undo button, which is roughly every automation project I've worked on.

Aquariums

Planted tanks, aquascaping

Music

Guitar — rock and metal

Cycling

City rides around Çanakkale

Games

Strategy, FPS, MOBA