Cheap software made your PM job harder, not easier. Here's the new job.

TL;DR

• 1

  The bottleneck moved from building to judgment — AI makes first versions cheap, so PMs are no longer gatekeepers of scarce engineering but classifiers of software abundance into market value, internal utility, or deletion.

• 2

  Microsoft’s scale shows the new reality — Inside Microsoft there are 1 million+ Power Platform assets, including 170,000 Power Apps, 50,000 flows, 18,000 agent environments, and 1,200 chatbots, forcing governance around inventory, telemetry, permissions, and data policy.

• 3

  Non-technical PMs are running out of road — Jones says PMs now need to reason about model behavior, evals, retrieval, latency, cost, permissions, workflow boundaries, and failure modes because those technical choices now shape the product itself.

• 4

  Broad experimentation should be allowed, but not left unmanaged — The right posture is not 'only PMs prototype' but 'let everyone build, then apply product judgment' so useful internal tools don’t stay hidden and risky ones don’t quietly spread.

• 5

  A production class ladder beats demo chaos — Jones proposes four classes—personal tool, team beta, supported internal product, and customer-facing product—with different ownership, support, monitoring, and governance expectations for each.

• 6

  Dead internal software is the new tech debt — If every prototype gets promoted, companies inherit support burdens, access risk, and zombie products; GitGuardian’s report of 1.2 million exposed AI secrets on public GitHub in 2025 is his warning sign.