Revenue Operations
AI adoption is not transformation
An AI-adopted company runs AI on top of its existing structure. An AI-native company designed its structure around what AI can do. Most companies think they're becoming the second by doing more of the first.
They're not.
I rebuild operations for B2B companies around AI and I see the same thing everywhere. A company adds meeting summaries, a support chatbot, a copilot for the sales team. The spend goes up. The org chart doesn't move.
Here's the thing nobody wants to hear. The org chart exists because information couldn't move on its own. A human had to carry it. A meeting had to exist so five people could hear what the software should have surfaced already. That's it. That's the whole reason.
And a status meeting with an AI notetaker is still a status meeting. You've made the waste faster. You haven't removed it.
The carrying layer
Think about what a middle manager actually does on a given day. They collect context from reports, compress it into a status update, carry it upward, carry decisions back down. They're a human protocol stack. That's not a dig at middle managers. It's a dig at the systems that made the role necessary.
When every internal surface is queryable, you don't need that layer anymore.
I built an agent for a B2B SaaS support team that does exactly this. Tickets come in, the agent classifies them, pulls context from the knowledge base, resolves L1 issues on its own or routes to the right person with full context attached. Before that, a human read every ticket, searched for answers, typed the response, decided whether to escalate. Fifteen to twenty minutes per ticket. The agent handles the same volume without adding headcount.
The before state: a ticket lands in JSM, a human reads the email, opens Weclapp, searches by the customer's email or number, navigates through several layers of the client page to surface account history, then goes separately to Confluence and Jira to find related docs and similar issues. All of that context lives in separate systems. None of it is in the ticket. The human is the join query.
This is how it looked.
Now with the agent, a JS script runs on a cron and does all of that before a human even opens the ticket.
I built another agent for the same company that reduced system configuration from two hours to under ten minutes. The old process was a human reading a spec, clicking through a settings UI, checking the output, fixing mistakes. Now the agent reads the spec and writes the config. A human reviews it. The bottleneck moved from doing the work to checking the work.
And here's what I keep telling people. None of this required a better model. It required writing the process down in a form the agent could actually follow.
The role nobody's hiring for
Every failed AI deployment I've seen has the same shape. The domain expert knows the process but can't turn it into instructions a machine can run. The engineer can build anything but has no idea what to build. And nobody sits between them.
That's where I work. Most of my day is turning implicit knowledge into explicit instructions, then wiring agents to run them. I watch the process as it actually happens, not as someone describes it in a meeting. I find the parts that are real judgment and the parts that are just habit. Then I separate them so the agent handles habit and a human handles judgment.
This role doesn't really have a name yet. I've been calling it agent operations but honestly even that feels too neat. If you run a B2B company and your AI projects keep stalling, it's probably because nobody owns this work. Someone needs to be the translator between how your company actually operates and how agents can operate inside it. The tooling is ready. The models are ready. The wiring is the bottleneck.
This is the first post in a series I'm calling The Post-Coordination Company. The next posts get into the practical stuff. How to design processes that agents can actually run. How to pick infrastructure that supports them. What the org looks like when you stop building around information that can finally move on its own.