The most important problem for AI in B2B is replatforming I co-taught a lightning lesson on AI for B2B PMs yesterday. The core is that everyone's racing to ship agents but the stack to support agents is not in place. When teams try to put the stack in place they find that the stack is shifting and they have to rebuild several times. → Data + plumbing. Most teams don't even have the data captured. Before anything else, you connect the siloed systems so agents can reason across them. → Evals + guardrails. In enterprise, you don't earn trust without a way to define and enforce "good." → Orchestration. Routing, multi-agent flows, and a shift toward constrained, deterministic skills where it matters. → Governance. Non-negotiable the moment AI touches a real customer. The gap between hype and reality is vast. One anecdote was about a company still migrating to the cloud. 😂 Then security. Then maybe agents. Teams are rebuilding their agent architecture 3–6 times in six months because everything underneath them keeps moving. Two things that surprised people most: 1) Trust, not accuracy, is the adoption blocker. One team I talked to had a system stall even though the recommendations were good. The reps and compliance didn't trust how it was built. Accuracy was never the problem. 2) The moat is the feedback loop on top. At consumer scale that loop is almost free: telemetry everywhere, easy experiments, instant signal. In B2B it's data siloed across every customer, telemetry that often doesn't exist, every account behaving differently. A real moat means the data belongs to you (not your customer, or they walk away with it), it's outcome-linked, and you turn usage into improvement faster than anyone else. Instrument first. You can't compound what you can't capture. Sometimes the moat is simply trust vs data. Customers already trust you and your company has gone through all the painful work already so data moats may not be that important. The teams that win are the ones doing the unglamorous work: fixing the data, earning trust, and making their products ready for agents. Thanks @gayathri_rg for co-teaching this with me. We're going deeper in a workshop for B2B PMs this July.