Every staffing tech vendor now claims to have agentic AI. Most of them are lying, or they do not understand the word they are using.
I have sat in a dozen vendor demos this year where the rep said agent every third sentence. When I asked what the agent actually did on its own, the room went quiet. What they showed me was a chatbot with a fresh coat of paint. A feature that summarizes a resume. A button that drafts an email. Useful, maybe. An agent, no.
The gap matters because you are about to pay for it. Vendors are pricing wrappers like agents, and staffing VPs and PE ops leaders are signing contracts for tools that do a fraction of what the pitch promised. This is how you tell the difference before the money leaves your account.
What a wrapper actually is
A wrapper is a thin layer on top of a large language model like GPT. The vendor sends your data to the model, gets a response, and shows it to you inside their software. That is the whole trick.
Wrappers do one thing at a time and they wait for a human to push each step. Summarize this resume. Rewrite this job description. Suggest a reply to this candidate. A recruiter has to click, read, and decide every single time.
None of that is bad. Some of it saves real minutes. But it is a feature, not an agent, and it should be priced like a feature. When a vendor charges you a per-seat premium and calls it agentic, you are paying agent prices for wrapper work.
Here is the tell. If the tool stops working the moment a human stops clicking, it is a wrapper.
What a real AI agent does
A real agent completes a multi-step task on its own. It makes decisions, takes actions, checks its own work, and only pulls in a human when it needs a call it cannot make.
Picture sourcing. A wrapper writes you a Boolean string. An agent searches your ATS and LinkedIn, ranks the matches, sends the outreach, handles the first two replies, books the screen on the recruiter's calendar, and flags the three candidates worth a human conversation. The recruiter walks in to a full calendar, not an empty search box.
That is the difference. One helps a person do a step faster. The other does the whole chain and hands you the result.
Real agents share a few traits. They run a sequence without a human clicking between steps. They react to what they find and change course. They know when they are stuck and escalate. And they leave a trail so you can see what they did and why.
If a vendor cannot show you all four of those in a live demo, they do not have an agent.
The five questions to ask before you sign
Do not ask if it uses AI. Everything uses AI now. Ask these instead and watch how they answer.
- Show me one task the agent finishes start to finish with zero human clicks. Not a slide. A live run. If they can only show single-step features strung together in a video, that is a wrapper edited to look busy.
- What happens when the agent gets something wrong? Real agents fail sometimes. Honest vendors have a plan for it: a confidence threshold, a human checkpoint, a rollback. If the answer is it does not get things wrong, they are selling you a fantasy.
- What has to change in my operation for this to work? This is the question that separates the pros from the pretenders. A true agent replaces part of a workflow, so someone's job changes. If the vendor says nothing changes, just plug it in, they either have a wrapper or they have never deployed the thing at scale.
- Who owns the output and who checks it? If an agent sends candidate outreach, who is responsible when it goes sideways. A serious vendor has thought this through. A wrapper vendor has not, because a human was always in the loop clicking send.
- Can I see the audit trail? Real agents log every action. You should be able to open a record and see what the agent decided, when, and on what basis. No trail means no accountability, and no accountability means you should not trust it with your candidates or clients.
The third question is the one that costs vendors deals, which is exactly why it works. Agents require operational redesign. Wrappers do not. A vendor who tells you their agent needs no change on your side is telling you it is not really an agent.
Why the redesign is the whole point
Here is what most buyers miss. The value of a true agent is not the software. It is the work you stop doing.
If your recruiters still do every step and the agent just makes each step a little faster, your cost structure barely moves. You saved a few minutes per req. That is a wrapper outcome, and you should pay wrapper money for it.
A real agent lets you take a whole function off human hands. That changes headcount math, quota math, and how you scale a desk. That is worth real money. But only if you redesign the work around it. Bolt an agent onto your old process and you get the cost of an agent with the results of a wrapper. The worst of both.
This is why I keep telling PE ops leaders to stop asking does it have AI and start asking what work does this remove. The first question gets you a demo. The second gets you the truth.
Do this before your next vendor call
This week, pull up the contract or proposal from the AI vendor you are closest to signing with. Find where it says agent or agentic. Then email the rep one line: Show me one task your agent completes start to finish with no human clicks, on a live call.
Their answer tells you everything. If they book the call and show you, you might have a real agent. If they send a recorded video or start explaining why that is complicated, you have a wrapper priced like an agent, and you just saved yourself a bad contract.