Host Daria and monday.com’s roy-mann (co-founder and co-CEO) and netanel-abergel (R&D director) open a monthly series on what they’re learning about agents. monday’s internal “Sphera agents”: after everyone built their own agents (hundreds, with cost and quality out of control), they moved to team-level AI teammates with real identities (Slack, GitHub, monday). The agents are assembled on demand from shared memory, skills, and knowledge, not one machine per agent, so an improved skill updates every agent at once. More than 20% of code is now written by agents. Mann’s taxonomy: agents and people working together, either in code (an orderly world with tests and deploys) or in general work (processes must be built, audited, and kept dynamic); and agents doing the work, closer to automation and splitting into many architectures and vertical startups. Lessons: don’t mix personality and memory into task agents; replace deterministic steps with scripts; give agents goals, not micro-tasks, and tools to check themselves. “Agents don’t fully work yet”: like early 3D printers, a decade of engineering above the model is the opportunity. Try everything hands-on, and track fundamentals, not product hype.
Show: startup-for-startup
Concepts: agent-architectures · agent-workspaces · harness-engineering · workflow-automation · ai-rd-rollout