The pattern is familiar to every enterprise leader: the AI demo works perfectly, the enterprise deployment disappoints, and the CIO can't explain why. Ahmed Bashir, Chief Technology Officer of DevRev, joins Craig Smith with a specific and well-supported answer: the problem isn't the model. It's that enterprise data is organizationally complex, fragmented across dozens of tools, locked behind conditional permissions, organized by custom schemas, and every prominent AI benchmark tests models against flat files with pre-packaged data, which has nothing to do with how enterprise data actually lives. DevRev's Enterprise Bench is the first benchmark to test complete agent harnesses against a realistically distributed enterprise environment where the agent must find, relate, and validate information before performing any task, and the results expose a gap that pre-packaged benchmarks hide entirely. The most striking finding: DevRev's system achieves 50% higher accuracy than Claude Code using
Andrew Barto on Rewards Gone Rogue Andrew Barto helped found reinforcement learning. He and Richard Sutton wrote the field's standard textbook, and they shared the Turing Award for that work. Reinforcement learning now trains today's reasoning models and AI agents. It also sits behind some of their strangest behavior, including the recent cases of agents breaking out of their sandboxes. Craig visited Barto at his home on Cape Cod. They talked about where rewards come from and why they are so hard to get right. Barto traces the problem back to Norbert Wiener's warnings in the 1950s. He explains how the dopamine system in the brain mirrors the learning algorithm he and Sutton developed. And he admits a weakness in their own framework: boiling every goal down to a single number. He also makes the case for engineering discipline. Ship designers learned their codes from wrecks and drowned sailors. Barto hopes AI can learn its lessons without that kind of cost.