Umut Vural — Artificial Intelligence

He keeps a persistent knowledge base in Obsidian using Karpathy's 'LLM Wiki' approach: he chooses the sources, AI writes and updates the pages.

• Raw sources are stored untouched; AI turns them into concept, organisation, source and question pages.
• Every page is linked; a catalogue (index) and a chronological log are maintained.
• Knowledge is not lost in chat; it accumulates with every task, and every fix is recorded the same day.

He has handed recurring information and communication work to scheduled AI routines.

• Morning brief: calendar, Türkiye headlines, marketing and market research news (ESOMAR, Nielsen, Ipsos) and a strategic suggestion of the day on one screen.
• Weekly summary, CRM email flows, a LinkedIn engagement routine and blog automation.
• Regular sync of the knowledge base; the automations' health is logged and every failure becomes a new rule.

He verifies before he trusts. The working rules in his records make this concrete.

• Nothing is called 'working' until it has been tested against the live system.
• If a result looks too good, the measuring tool itself is suspected first.
• Missing data is never counted as zero.
• Automated test suites, security audits and checklists run before launch; security and testing are assigned to separate specialist personas.

He turns business requirements directly into working tools with Claude Code, and he does it through a written process rather than an ad-hoc chat.

• First a written development plan; the work is split into numbered milestones and each stage closes with automated tests.
• A separate security audit before launch; security and testing go to separate specialist personas.
• Critical rules ('invariants') and handover notes are written down, so different sessions and agents work with the same rules without losing context.
• Systems are built from day one to be adaptable to more than one brand (white-label).
• Success and stop criteria are set up front; when they are not met, he does not hesitate to stop the work.

Curious but critical: he welcomes AI taking over drudge work, but argues the user must draw the line. “Make my work easier, but don't live my life for me.”

• He stresses that there is still a gap between impressive demos and business processes that work reliably every day.
• He argues that the use of AI and synthetic data in research should be openly disclosed, and sees this as a source of trust for Adgager.
• He writes regularly on LinkedIn about AI safety, multi-agent systems and robotics.