Post by .406 Ventures

Last week, our Data + AI team hosted a dinner in San Francisco with CDOs, CIOs, and founders working on enterprise AI across industries. Graham Brooks led the conversation, joined by Liam Donohue and Randy Bean (.406 EIR and a leading voice on data and AI leadership). It was a few hours with a room full of leading practitioners. A handful of things stood out: On cost: The AI-native companies aren't capping token spend, they're forecasting it. Several are seeing it climb toward 15% of salary spend and choosing to let it run, because they believe the ROI is there. That said, ROI is still critical, and a lot of CDOs continue to work directly with their CFO's office to quantify and validate the ROI on individual AI initiatives. On agents: Healthcare and financial services keep coming up as the early adopters of agentic LLMs, which surprised some people in the room. One CDO had a good theory for why: regulated industries are deeply process-oriented, and process is what agents automate well, giving them a leg-up in the agentic race. On continual learning: The teams getting the most out of their models aren't just using them to answer questions, they're using them to surface context and knowledge the business didn't know it had. Figuring out how to operationalize that loop feels like the next real piece of work. Thanks to everyone who joined us, and looking forward to the next one. Ahmed El-Kishky, Alex 'Sandy' Pentland, Arun Kumar, Eric Shoemaker, Faraz Shafiq, Lee Davidson, CFA, Parker Phillips, Pär Edin, Smita Nair, Stephen (Steffin) Harris, Tommy Dwyer