The defining problem of the agentic AI era isn't building the agents-it's giving them context. Frontier models fail without the proprietary organizational knowledge that supports enterprise decisions, says Vanessa Liu of Appen Ltd.
Lesson 1: Data Is Your Moat. Companies with a clear customer problem and defensible data advantage will stand out, according to Thomson Reuters CEO Steve Hasker.
Lesson 2: Speed Is Everything. Users lose patience fast. Bright Data's Ariel Shulman notes that chatbots must now deliver responses in under 500 milliseconds.
Lesson 3: Agents Need Bank Accounts. Catena Labs is building a "know your agent" model for identity, authorization, and accountability in agentic finance.
Lesson 4: Token Lock Is the New Vendor Lock. OutSystems CEO Woodson Martin warns that betting on a single frontier model surrenders cost leverage. A platform that allows model swapping at runtime is essential.
Lesson 5: Adoption ≠ Deployment. WalkMe's Tai Carmi highlights that 80% of executives believe they provide great AI tools, but employees disagree. Contextual nudges at the right workflow moment matter more than new tools.
Lesson 6: Start Big, Then Swap. AG2ai CEO Qingyun Wu advises building with the most capable model first, then swapping in cheaper alternatives that match its performance.
Lesson 7: Production Is Where Agents Go Rogue. Monte Carlo Data's Barr Moses warns that stale data, hallucinated outputs, and blown token budgets crash many pilots. Courts hold the company behind the agent fully accountable.