uri-eliabayev talks with nadav-erez, CTO and co-founder of Twine, which builds “digital employees” for security, starting with “Alex” for identity and access management. Not a copilot: Alex owns a KPI (everyone has exactly the access they need, when they need it) and works continuously, replacing quarterly review cycles with daily ones. It runs on off-the-shelf LLMs, with the value in a rich, constantly enriched data layer and a knowledge layer of org policies and exceptions. Where no written policy exists, it infers practice from a year of tickets. Scoped per-page agents didn’t survive real users, so they moved to one agent loading skills, including awareness of what’s on the user’s screen; evals improved immediately. Trust is earned gradually: first read-only data gathering on escalated tickets (e.g. why a user keeps getting locked out), then a “control membrane” mapping each action and context to an approval level. A deterministic plan is executed with no AI between approval and execution, alongside audit logs, undo, an honest “I don’t know”, and a “raise issue” tool for async tasks. Vision: sibling agents (vulnerabilities, network) sharing one data fabric, and a CISO who sets objectives that cascade into KPIs and tasks.
Show: hidden-layers
Concepts: agent-workspaces · ai-guardrails · skill-engineering · agent-memory