Everyone has an AI roadmap. Almost nobody has a headcount disclosure.

An adoption roadmap is a document with dates, owners and a budget line. The thing Gartner managing partner Jackie Swanson says almost no employer has, "an honest plan for what AI is doing to their people, their pace and their pipeline of future leaders" ^[3], has none of those properties. It has no publication date, so it never has to be wrong. Staff fill the space themselves, and they fill it pessimistically.

Quiet rehiring does not close that space either. An employer that walks back an AI-justified cut ^[2] has told its workforce nothing about the next decision, and the pipeline problem is the one with the longest fuse: the entry-level work that used to season future managers can be absorbed by tooling without anything in this year's numbers recording the loss. Swanson's argument is that the cultural advantage of getting this right compounds long after the tools commoditise ^[4]. That clock runs both directions, and the debit side is invisible for years.

Torani shows what a credible answer costs. The syrup manufacturer has gone 103 years without a layoff, employs more than 500 people, and has had the same chief executive, Melanie Dulbecco, since 1991 ^[13]. It plans to grow headcount by roughly 30% by March 2027 on the back of a $60 million manufacturing expansion ^[16]. At its stated floor of 500 staff, that is about 150 additional roles ^[17], or roughly $400,000 of committed capital behind each one ^[18]. Torani is not reassuring its people with a communications plan. It is reassuring them with capital expenditure, and it also gets to do the incremental thing, testing pilots and letting individuals iterate on them ^[14]. Most employers announcing AI programmes have no comparable cheque to point at.

Superhuman, the company Grammarly became in 2025 after acquiring the AI email app of that name and folding in Coda as Superhuman Docs ^[10], substitutes authority for capital. Chief people officer Kenny Mendes says there is no top-down mandate and that adoption succeeds where teams close to the problem are allowed to pick up tools themselves ^[11]. Because the teams own the roadmap, he says, it does not read as a cost-savings exercise ^[12]. That is a real mechanism, and it is cheap, but note what it transfers: discretion, not information. A team choosing its own tools still does not know the headcount plan.

Ironclad, founded in 2014 ^[7], runs the cheapest version and the most fragile. Chief technology officer Sunita Verma, previously at Character.AI and Google ^[19], describes messaging built on upskilling, with courses, classes she taught herself, and peer learning ^[8], plus transparency about the pitfalls of the tools as well as the promises ^[9]. That works while the speaker is technically credible and turns out to have been right. It does not survive one quarter in which the tool did replace the job.

None of the three is quoted disclosing the number that workers actually want: how many people, doing what, by when. The backlash is already turning up in midterm election fights and possibly in IPO filings ^[5], which is a different audience with subpoena-adjacent habits. Sentiment sitting at 53% across demographic groups ^[1] is what an unpublished plan looks like from the outside.