Bottleneck Shift to Enterprise Plumbing

The bottleneck has shifted from intelligence to enterprise plumbing. Factory already has the hard part of an agent product, a runtime that can inspect files, use software, reason across steps, and make controlled changes. Moving from engineering into finance, sales, or product mostly means teaching the same runtime which systems matter, what actions are allowed, and what a good output looks like for each role, rather than rebuilding the core agent stack.

• The new Personas and managed Connectors launch shows the expansion path clearly. A finance Droid can pull revenue from Stripe, pipeline from Salesforce, and plan data from Sheets or Anaplan, then write back only through admin approved paths. That is a permissions and integration problem more than a reasoning problem.
  1 sacra 2 factory 4 factory
• This mirrors how other enterprise AI products have evolved. Writer organizes its product around playbooks, routines, connectors, and admin controls for apps like Microsoft 365, Google Workspace, HubSpot, Gong, PitchBook, and FactSet, which suggests the winning layer is governed system access wrapped around a general model runtime.
  5 sacra
• The demand signal is already there outside engineering. By June 2026, OpenAI said knowledge workers were about 20% of Codex users and growing more than three times as fast as developers. Factory is effectively betting that once the runtime exists, each new function can be unlocked by role specific onboarding and trusted app access.
  3 openai 6 openai 7 openai

The next phase is a land grab for regulated workflows, where connectors, access policy, auditability, and data residency determine who can sell into large enterprises and sovereign buyers. In that market, the advantage goes to platforms that turn one general agent runtime into many role specific workers without forcing customers to bet on a different architecture for each department.