RolespectFind the role. Get the Rolespect.
Field notes

What an AI job agent can and can't tell you.

An agent can find more roles in a minute than you can in a month, and some can put your name in front of a hiring manager. What it cannot do is decide which of those roles deserves you. That part is still yours, and it is the part that decides how the search goes.

Listen to the companion episodeRun The Mission, Episode 24: What an AI Job Agent Can and Can't Tell You

Run The Mission is Bradley's podcast from Wendling Nonprofit Consulting: nonprofit operations, with a job search series beside these notes. Every episode

Two kinds of agent. One kind applies for you. It matches roles, tailors your resume to each one, fills in the forms and submits. The other kind introduces you. It talks with you about what you want, matches you against open roles, and when an employer agrees, puts you in front of the hiring manager.

Both are real, and both do things a job board does not. Neither one knows what you will not move on until you tell it, and telling it once is not the same as holding the line.

The question an agent answers is where you could apply. The question that decides a search is where you should. Those sound alike. They are not the same question, and only one of them can be handed off.

What an agent can tell you

Credit first, because it is earned.

  • That a role exists. Millions of listings a day, read faster than you will ever read them. Finding is the part software does best, and it is better at it than you are.
  • How closely your words match theirs. A match score compares the language in your file with the language in the posting. That is useful. It measures overlap, and overlap is one input to a decision, not the decision.
  • That an employer wants to meet you. An introduction the employer asked for is the strongest signal in a job search. It settles whether the seat is real and it skips the pile entirely. If one arrives, take the meeting.
  • What the room will probably ask. Practice questions built from the posting are cheap, and they beat walking in cold.

What it can't tell you

Five things sit outside what any agent can see. None of them is a flaw in the software. They live with you, and no amount of matching reaches them.

1. What you will not move on

Only you know

Your pay floor, the longest commute you will make, the lowest title you will take. An agent can ask, and a good one does. But a preference typed into a conversation gets weighed against a match score, and a 94 percent match is persuasive at eleven at night. A floor has to be a gate. Below it, the role does not clear, however well you match. Write yours down before you open anything.

2. Whether the seat is live

Check it yourself

A listing can keep rendering for weeks after the role is filled or the budget is pulled. An introduction settles this, because somebody on the other side asked for you. A match that arrives without one has not been settled, and thirty seconds on the employer's own careers page settles it. The eight ghost job signals are written out here.

3. Whether the role is worth your evening

Not the same as match

Match asks whether you fit them. Worth asks whether they fit you: the pay against your target, the level against where you are going, the commute, the work itself. A role can be a 96 percent match and a bad use of a Tuesday. Score the two separately, or one will quietly stand in for the other.

4. Anything true about you that is not already written

Nothing to add

An agent that writes your application works from what you gave it. It can reorder it, reword it and lean on the right parts. What it cannot know is the story behind a line, and anything it writes past what you gave it is a claim you will be asked about in the room. Read every line that goes out in your name. If you could not tell the story behind it, cut it.

5. What its own numbers mean

Ask out of how many

More interviews, faster offers, a match percentage. Every one of those is a count somebody chose how to make. Out of how many. Measured against what. Counted by whom. A figure that cannot answer those three is a headline, not a finding, and that rule applies to figures on this site too.

Volume gets cheaper. Attention does not.

The last field note walked through the Greenhouse benchmark report from March 2026. Applications per job went from 116 in 2022 to 244 in 2025, while the average recruiting team went from 10.4 people to 4.6.

An agent that applies for you makes each application cheaper for you. It does nothing for the person at the other end, who is now reading your tailored resume next to everyone else's tailored resume. When sending costs nothing, sending stops meaning anything. The applications that still land are the ones somebody chose.

That is the case for the kind of agent that introduces. It is also the case for deciding, which you can start tonight with nothing but a spreadsheet.

Use one, then decide

None of this is an argument against using an agent. It is an argument about order. Let the agent find. Bring what it finds back to a decision you make against rules you wrote before you saw the posting. Then send the few that clear, with your own words on them.

It is the same order that works with the job boards. A board, an agent, a friend forwarding a link: each is a place a posting comes from. The decision about it happens after, and it happens in one place, against one set of rules.

Five questions to ask any agent

  • What happens to the things I will not move on? Held as a gate, or weighed against a score.
  • Does anything go out in my name without me reading it first? If it does, how do I see exactly what was sent.
  • Where does my file live, and how do I delete it?
  • Who pays, and when? Free to you can mean the employer pays on a hire, or you pay after a trial. Both can be fine. Know which one you are in.
  • Out of how many? For every outcome it quotes.

What this will not tell you

  • Whether any particular agent is good. This note is about the category. Some of them find roles very well, and an introduction is worth more than anything written here.
  • That fewer applications produces more interviews. Not on its own. It means the same evenings go to roles that could actually answer you, which is the part you control.
  • Whether those Greenhouse numbers describe your market. They come from organizations running one applicant tracking system, which skews larger and more technical. A forty person nonprofit hiring its first operations director behaves nothing like that.

Running it without the app

None of this needs software. By hand, the order is:

  1. Before you open any agent or board, write three lines: your pay floor, your longest commute or the work arrangement you need, and the lowest title you will take.
  2. Hold every match against those three lines first. If it fails one, pass, and log the pass so you can see how many evenings you kept.
  3. Confirm the seat on the employer's own careers page. An introduction the employer asked for counts. A listing does not.
  4. Score worth separately from match. Pay against your target, level against where you are going, and whether you want the work.
  5. Read every line of anything sent in your name. Cut what you could not tell the story behind.
  6. Log where each application came from, the agent included, and which resume version went.
  7. Every fifth application, look at the record instead of asking for more matches. Ask which source has produced a conversation.

Rolespect runs that sequence, keeps the record, and prints the rule behind every call it makes. It finds nothing on its own and writes nothing in your name. It is free, there is no account, and nothing you enter leaves your browser.

An agent can hand you a thousand roles. Deciding which ten deserve you was always the job.

Run it

Let the agent find. Decide here.

Paste what any agent or board sends you. Rolespect confirms the posting is real, holds it against the gates you set once, and returns Apply, Investigate First, or Pass with the rule next to it. Free, no account, and nothing you enter leaves the device.