Back in June I asked here how you'd underwrite the AI line in a deal model. If you go back and read the thread, the consensus seems to be that you shouldn't be underwriting any gains from AI because you can't rely on the outcomes. I've been pondering this the past several weeks. Back in June, I was writing about my team of agents working in my software business. That was fairly easy to build effective agents because the company is new and all the information is in a database, codebase and a small repository of insights. Since then I've been building a different kind of agent, a personal AI chief of staff, and it taught me the variable I think I under-weighted. Context. It turns out that your personal life, when you consider wealth, health, relationships, and business ventures over a 20 year time horizon, approaches a level of complexity far greater than a simple software startup. Let's stop and consider what makes a human chief of staff effective. It's total knowledge of you: your calendar, your deals, your commitments, your family logistics, how you make decisions, what you dropped last week. Hire the sharpest person alive, give them none of that, and they can't do the job. Same with the AI version. My first attempts were regular agents: capable, cheap, and honestly kind of disappointing. Generic output I had to rewrite anyway. Then I wired them into a second brain, about 3,000 interlinked notes holding every meeting transcript, historical email, personal CRM, evernote note, google drive file and decision from my work and personal life. The improvement was undeniable. I measured it last week: 12 real hours saved, counting only work I would have done manually myself. Email drafting, meeting follow-ups, calendar coordination, the to-do list. Same models. The only change was what they could see. I suspect the same principle decides those corporate AI projects. For a business, context means consolidating the data silos so an agent has fresh, complete information in front of it before it acts: the CRM, the inbox, the SOPs, the financials, what was decided last quarter and why. My chief of staff was the easy version (one person, greenfield). In an acquired company, the consolidation is most of the project. But, it's doable and it will happen over time. BTW: I'm demoing the chief of staff and its second brain live Wednesday morning. Come poke holes in it. Event Link: redacted For those of you deploying AI in a business you own or operate: does this match what you're seeing? When a pilot stalled, was the model the problem, or could it just not see enough of the business to be useful? Anyone tried to build a "second brain" for a business?