For twenty years, businesses bought what I call dumb tech: CRMs, HRIS platforms, ATS systems, and learning platforms. Every one of them held the data. None of them could act on it. The job of connecting them fell to a person: pull the report, update the record, copy the same answer into three different tools because the tools don't talk to each other. We built an entire layer of skilled, expensive humans just to be the glue between our own software.
Agentic AI is driving straight through that layer. I call it the AI Wedge: AI that stops assisting and starts running the tactical work itself, without a person moving the data by hand first. It's already splitting the human from the software they used to operate directly.
This is already showing up in the data, and not where most people are looking for it. Stanford's Digital Economy Lab has tracked a real hiring gap: workers aged 22 to 25 in the occupations most exposed to AI now sit 19% below where employment would be if it had simply tracked their less-exposed peers, the gap coming from roles that quietly stopped getting filled. MIT's Iceberg Index shows why that's easy to miss. The visible disruption, the tech-sector story everyone's writing about, is only 2.2% of wage value, about $211 billion. The exposure nobody's covering sits in administrative, financial, and professional services work: five times larger, an estimated $1.2 trillion. The Wedge is moving through the accounting department faster than it's moving through the engineering org.
Whether AI can do the work isn't really the question anymore. It obviously can. The real question, the one most leaders skip past, is what a person should be doing instead once it does.
Start with where the time actually goes today. Deloitte's research with Zoom found that employees spend nearly 25 hours a week preparing for, sitting in, or following up on meetings, three full days on coordination instead of the work they were hired to do. That's the tactical middle. That's the layer the Wedge removes, and it's most of the answer to the question above: give someone back those three days, and the job changes shape.
Wharton professor Ethan Mollick calls the model for what fills that time co-intelligence: a human and an AI agent working as one team, the human directing, the agent executing. A business built entirely on AI won't have the judgment or relationships to run itself. A business that refuses to change won't move fast enough to keep up. The leverage sits in co-intelligent teams, built on purpose.
There's a second effect of co-intelligence that gets less attention than it should: it doesn't just free up time, it compresses how long it takes to get good at a job. Stanford economist Erik Brynjolfsson and his coauthors found customer support agents using an AI assistant resolved 14% more issues per hour on average, and the newest, least experienced agents improved by 34%, with almost no measurable change for the most experienced staff. The floor rises. The gap between a rookie and a veteran shrinks.
That changes what businesses look for when they hire, because the floor already moved. Korn Ferry's 2026 talent acquisition research found 73% of TA leaders now say the skill they need most this year is critical thinking and problem-solving, AI technical skills ranked fifth. The tactical skills that used to get someone in the door are worth less. The judgment that used to sit with the most senior person on the team is the baseline expectation now.
Move too fast on that hiring shift, though, and you lose people before you ever find out if the model works. Greenhouse's 2026 Candidate AI Interview Report found 38% of U.S. candidates have already walked away from a hiring process because it included an AI interview, and another 12% would if one were required. People can tell when a company inserted AI to save itself time without thinking about the person on the other side of it. The Wedge creates leverage. It also creates a way to lose someone's trust faster than any process it replaced.
We ran our first real test of this on our own Client Success team, before we ever considered it for a client. CSMs used to spend up to thirty minutes on account research before every client conversation, more once you factor in the context switching to get into it, and the depth varied CSM to CSM depending on everything else on their plate that day. We built an agent to do that research instead: pull the account history, review the call transcripts, read the news, and hand back a real playbook. The agent took the tactical middle off the CSM's plate, freeing up time for the conversations that actually decide whether a client renews or expands.
We didn't build that agent by turning it loose, though. Before any workflow gets handed to an agent at Lever Talent, it has to clear four checkpoints we call the 4Cs: Context, Constraints, Connections, Control. We wrote the full breakdown, including exactly how we applied it to the Client Success agent above, here: [LINK: The 4Cs: How We Decide What Actually Deserves an AI Agent]
The Wedge is going to keep moving through the tactical middle of every business, whether or not anyone names it. The leaders who get ahead of it won't be the ones who bought the most AI. They'll be the ones who decide, on purpose, what an agent should own and what stays with a person, then build the governance to make that decision hold. We call that being a Human Orchestrator. It's the same discipline behind everything we're building at Lever Talent right now.
Want to see exactly how we vet a workflow before handing it to an agent? Read the 4Cs.
Want to run your own team through this? Register for our AI Fluency for HR masterclass, available in October and November.
Want help finding where the Wedge is already moving through your business? Book time with us.
Let's Orchestrate.