The Work AI Just Came For Isn't Entry-Level

In July 2024, OpenAI shipped Presence—not a smarter foundation model for developers to tinker with, but a finished worker that resolves insurance claims, settles billing disputes, and closes IT tickets. The quiet part: it's aimed at work we've been calling "senior" for decades.

I've spent my career inside regulated firms where "senior-only" is stamped on tasks like protective classification. Claims adjudication. Billing disputes. Technical escalations. We told ourselves—and our clients—that these required judgment, expertise, institutional knowledge. That much was true. What we didn't say out loud: most of that judgment happens inside guardrails we could have documented years ago.

Presence resolves about 75% of inbound issues. It's priced like a consultant, not an API call. And it's not selling intelligence—it's selling governed labor, guardrails on, ready to deploy.

The Climb Through the Stack

The cloud did this once.

Amazon didn't stop at renting servers. It climbed through infrastructure into platforms, then into applications, until it was running the software itself. Salesforce didn't replace your CRM admin—it became your CRM, then your customer service layer, then your analytics engine. Each time, the starting pitch was "we'll handle the commodity stuff," and each time, "commodity" crept steadily upward.

AI is making the same climb through knowledge work. It started with transcription and first-draft content—intern tasks. Then it moved to research synthesis and data analysis—associate work. Now it's at claims adjudication and dispute resolution—the tasks we reserved for people with seven years of experience and a senior title.

The pattern isn't new. The target is.

What "Senior" Actually Protected

I was advising a client last month—regional insurance firm, about 200 employees—when their VP of Claims said something that stuck with me: "Our senior adjusters don't touch a claim until it's been through three junior reviews."

I asked what the senior adjusters actually do.

"They read the file, check it against the coverage rules, maybe pull one related case for context, then approve or kick it to legal."

Rules. Retrieval. A judgment call at the end.

Turns out a lot of "senior" work fits neatly into a dropdown: approve, or escalate.

That's not a criticism of the work. Billing disputes and claims adjudication are complex. They require pattern recognition, institutional memory, and understanding edge cases. But complexity isn't the same as creativity. And pattern recognition is exactly what these systems were built to do.

The uncomfortable part—and I'm watching clients sit with this in real time—is that "senior" was often protecting the workflow, not the other way around. It wasn't that the work required seniority. It's that we needed a liability boundary, and seniority was how we drew it. Experience wasn't guarding irreplaceable insight. It was naming who got to click approve.

The New Job Description

This doesn't make senior people worthless. It moves where the value lives.

When governed agents do the task work, the job becomes designing the system: the handoffs, the overrides, the escalation triggers, the human checkpoints. It's deciding which 75% gets automated and which 25% still needs a person. It's knowing when the guardrails should flex and when they should hold.

Phil Jackson never outscored his players. He built the system they won inside.

The firms that adapt fastest won't be the ones that protect the old workflows. They'll be the ones that take their most experienced people and ask: what should this system know? Where does it break? What does a human need to see before signing off? The expertise doesn't disappear—it gets encoded into architecture instead of process.

But that requires naming the work honestly. If your "senior-only" task is really retrieval + rules + approval, you're not designing a system—you're waiting for someone else to design it for you.

The Question You Should Ask Monday Morning

I've now watched this movie four times. The internet didn't kill publishing—it killed the idea that distribution was a scarce, valuable skill. Mobile didn't kill retail—it killed the idea that location was destiny. Cloud didn't kill IT—it killed the idea that running infrastructure was inherently strategic.

Each time, the professionals who survived were the ones who moved up the stack before they were forced to.

So here's the uncomfortable question for every partner, director, and reviewer in a regulated firm: Which of your "senior-only" workflows are really rules, retrieval, and an approval step wearing a title?

Not which tasks could theoretically be automated someday. Which tasks does a system like Presence already do, right now, at 75% accuracy, priced to undercut your fully-loaded labor rate?

Make a list. Show it to your team. Then decide: are you going to design the system that does this work, or are you going to review the output from the system someone else designed?

Because governed agents didn't just arrive. They're already resolving claims while you're deciding whether to pay attention.

What to do this week: Pull your three most common "senior-only" workflows. Map them honestly—how much is retrieval, how much is rules-based, where does human judgment actually matter. Then ask: if we were designing this workflow today, would we staff it the same way? That gap between "how we've always done it" and "how we'd design it now"—that's your blueprint. Or your blindspot.