Aaron Levie described what it actually takes to deploy AI agents in the real world this week. Setting up unstructured data. Learning the workflow. Creating plans. Connecting systems. Redesigning the process. Deciding where humans oversee the work. Figuring out how to validate it.
Then he called it a job posting.
He’s right about the work. Every word of it. The part where we see it differently: he thinks you need to hire for it. We’re building an intelligent system that guides it.
Here’s the pattern nobody’s connecting:
Infrastructure teams are building enforcement at record speed. Agent identity frameworks. Policy engines. Discovery tools. Operating systems. All of it assumes the governance content exists — that someone already figured out what rules the agents should follow, for which workflows, with what boundaries.
Meanwhile, the people who actually have to figure that out are Googling "AI governance framework" and getting PDFs from 2023.
$3.6 billion in RSAC announcements built the alarm system. Hundreds of millions more are funding agent operating systems, identity frameworks, and enforcement platforms. Who’s building the intelligence that tells any of them what to look for?
That’s the work. Not the tooling. Not the hiring. Sitting with a team, picking one workflow, designing how agents and humans share that work, encoding the rules in formats machines can execute, and moving to the next one.
Rules compound. The fifth workflow is faster than the first.
The infrastructure is ready. The intelligence layer is not. That gap is where the next wave of value gets created.