In short
757 of 1,719 live agentic AI listings disclose a salary range, a rate of 44%. In a market where advertised bands span more than $200K between mid and staff level, the missing majority costs candidates real money and costs employers applications they never see.
We can only calculate salary benchmarks from listings that publish one. Across our live corpus of 1,719 agentic AI roles, 757 disclose a range. That is a disclosure rate of 44%.
The other 962 listings ask candidates to enter a process without knowing whether it is worth their time.
Why this matters more in agentic AI than elsewhere
In a mature market, an undisclosed range is an inconvenience because candidates can estimate. In agentic AI they cannot, for two reasons.
The bands are unusually wide. Median advertised pay runs from $201K at mid level to $259K at staff, and individual ranges within a level routinely span $100K. Title alone does not tell you what a role pays, so the guess a candidate makes is close to arbitrary.
Titles are not standardised. The same work is advertised as AI engineer, agent engineer, LLM engineer, forward deployed engineer and machine learning engineer depending on who wrote the posting. When titles do not map to bands, an undisclosed range removes the last usable signal.
The large employers disclose least
The intuitive assumption is that big companies publish ranges because they have compensation bands and legal departments, while scrappy startups keep it vague. The data says the reverse.
The ten largest employers disclose on 34% of their listings. Everyone else discloses on 52%.
The gap is eighteen points in the opposite direction to expectations. Large employers in this market are the least transparent about pay, and because they account for 44% of all listings, they pull the market-wide rate down to 44% on their own.
Several things plausibly drive this. Big companies hire across many locations with location-adjusted bands, and a single advert cannot easily express that. They also run structured levelling processes where the band depends on where you land in an interview loop rather than on the job title. Both are real constraints. Neither is invisible to a candidate deciding whether to spend four weeks on a process.
Disclosure by level
Within the disclosing half, the rate varies by seniority in a narrow band with one clear outlier:
- Senior: 49% disclose, 401 listings
- Lead: 45%, 165 listings
- Mid: 44%, 754 listings
- Staff: 42%, 322 listings
- Executive: 25%, 52 listings
Executive roles being opaque is expected and largely uninteresting. The more useful observation is that staff level, which commands the highest median pay at $259K, is also slightly less likely to publish it than senior or mid. The roles with the most money attached are marginally the least legible.
The single-opening employers are the least transparent of all
Companies with exactly one open role disclose on just 23% of listings, roughly half the market rate.
This is worth knowing because that segment is otherwise the most accessible part of the market. Small employers are nearly twice as likely to offer remote and hold two thirds of all junior openings, but they are also the least likely to tell you what the job pays before you apply. If you are working that part of the market, raising compensation early is not optional.
What non-disclosure costs employers
The usual argument for withholding a range is negotiating position. In a market this tight it mostly costs applications. Strong candidates in a supply-constrained market self-select out of processes with unclear compensation, because they have alternatives that told them upfront.
There is also a measurement cost. If you do not publish, you do not appear in the benchmarks candidates use to calibrate expectations, which means your offers get compared against a distribution you did not contribute to. Every salary article written about this market, including this one, is describing the 44% who published. The other 56% are being benchmarked in absentia against numbers they had no part in setting.
How to negotiate when there is no range
Ask before the first technical round. Not in the first thirty seconds, but before you invest a take-home. Framing it as a calibration question rather than a demand works: you want to check the band is in the same territory before either side spends more time. Almost nobody reacts badly to that framing, and the ones who do have told you something useful.
Anchor on the level, not the title. Establish what scope the role carries and map it to the seniority bands rather than to the words in the job advert. If the responsibilities are staff-level, the comparison is $259K rather than whatever the title implies. Ask what the role owns, who depends on its decisions, and what failure would look like. Those answers place the level more reliably than the title does.
Use the published half of the market. The 757 listings that do disclose are a legitimate benchmark, and citing a distribution is more persuasive than citing a personal expectation. There is a meaningful difference between saying you want $240K and saying that median advertised pay at this level is $216K with senior ranges routinely reaching into the $250Ks.
Expect less transparency from the big names. If you are interviewing at one of the ten largest employers, non-disclosure is the norm rather than a signal about you, and the number will come from levelling rather than the advert. Plan for a later conversation instead of reading it as evasion.
Method and caveats
Disclosure is counted as any listing publishing a minimum salary. Ranges quoted are the median of the midpoint between advertised minimum and maximum, calculated per seniority level and only where at least 50 listings support the figure. Non-US listings and non-annual pay periods are normalised where possible and excluded where not, which means these figures describe the disclosing, mostly-US portion of the market rather than all of it. Underlying counts are in the open dataset, which refreshes daily.
Frequently asked
What share of AI agent job listings show a salary?
44%. Across 1,719 live agentic AI listings, 757 disclose a range and 962 do not, so a candidate faces a coin flip on whether they can assess a role before applying.
Why does salary disclosure matter more in agentic AI?
Because the bands are unusually wide, running from $201K at mid level to $259K at staff with individual ranges often spanning $100K, and because titles are not standardised, so the same work appears under five different names.
Does withholding a salary range help employers?
Mostly not, in a supply-constrained market. Strong candidates self-select out of processes with unclear compensation because they have alternatives that were upfront, and non-disclosing employers also stay absent from the benchmarks candidates calibrate against.
When should I ask about pay if the listing has no range?
Before the first technical round and certainly before investing in a take-home. Framing it as calibration rather than a demand works: you are checking the band is in the same territory before either side spends more time.
How do I benchmark a role with no published range?
Anchor on the level rather than the title. Establish what scope the role actually carries, map it to the seniority bands, and cite the distribution from the 757 listings that do disclose rather than a personal expectation.