Since the 2008 financial crash, the G7 has averaged about 1% annual productivity growth, half the 2% pace of the decade before. The AI Adoption Initiative set out to examine that slowdown. In September 2026 it published a consultation paper, Mapping the Missing Middle in the AI Workforce, by Lindsey Lacey, Dustin Ferrone and Christophe Combemale, with a foreword by Ed Teather. It launched on November 11, 2026, at the LSE Global Forum for AI and Social Sciences. Its central question is about adoption rather than invention: if AI is going to lift productivity, who makes it work inside real organisations?
Across roughly 94,000 AI-related job postings from S&P 500 companies, most of the demand was not for people who build AI. The research, presented at the forum on behalf of the AI Adoption Initiative, finds that facilitators account for roughly 60–80% of AI-related postings across sectors. In the paper's framing, facilitators are the people who deploy, integrate and govern AI. They form the middle layer of what the authors call the AI Labor Stack, between frontier innovators and everyday users. Most of the AI hiring the study counted sits in that middle layer.
Postings need a caveat. Bain & Company's tracking shows third-quarter postings fell year over year in seven of ten major markets, eight of ten industries and nine of ten functions, even as AI-related postings climbed. Bain is explicit that job postings are a leading indicator of labour demand but do not measure employment levels. A posting is a request for a person, not a person on payroll. The 60–80% figure shows what companies advertise, not what they staff or pay.
Deployment is the bottleneck, and the middle layer is where it gets solved
The mechanism is simple. A model never wired into a claims process or a factory line produces no measurable output. Someone must connect the model to data and software, adapt working practices, and keep the system governed. The paper groups facilitator skills into three overlapping domains: technical work that integrates models, data and software into usable systems; social work that builds trust and helps organisations adopt AI responsibly; and infrastructural work that supplies the compute, power and connectivity for inference.
The split inside the category matters. Among facilitator roles, 39% are purely technical, 32% blend technical and social skills, and nearly 20% are purely social. More than a third of postings span multiple domains. The category is not one job family. It is three families sharing a title.
New work usually arrives in categories nobody named in advance. Economist David Autor and co-authors estimated that more than 60 per cent of US employment in 2018 was in job specialties that did not exist in 1940. In May 1982, the Central London branch of Britain's Musicians' Union voted to call for a ban on synthesisers. The instrument did not disappear. Players who learned to programme it became a new trade. Facilitators are the same kind of novelty, and their job titles are still settling.
A senior-heavy hiring pattern sits under the facilitator boom
The evidence needs careful marking. The 94,000-posting analysis, the 60–80% facilitator share and the three-domain taxonomy are the paper's central claims. The 39/32/20 split and the sector demand growth since 2024 in manufacturing, financial services, technology and healthcare are reported secondhand.
A separate analysis complicates the picture. The AIDE Institute examined 161,645 postings from nearly 500 of America's largest public companies and found 8,140 AI-related roles, of which 71% are senior-level. Senior positions outnumber junior ones by more than five to one, with 16% middle-tier and 13% junior. This is a single institute's finding and has not been replicated. If it holds, the facilitator layer is large while the entry ramp into it looks narrow. The paper itself argues for a productivity agenda married to a pro-worker agenda of participation and resilience, so the new skills and wage premiums are shared. Whether the market does that is still open.
Facilitator roles will split, and I expect only one branch to hold its pay
Facilitator demand is real and will keep growing, because deployment is where productivity either arrives or stalls. But "facilitator" is a bucket, not a career, and the fault line runs through it.
My expectation over 12 to 24 months is that facilitator roles become the fastest-growing named category in corporate AI hiring, then divide. Technical facilitators, doing integration, data plumbing and governance engineering, own the surface where a system meets the business. That gives them leverage to build a durable ladder. Purely social facilitators, the adoption coaches, change managers and AI trainers, are likely to be folded into existing HR, operations and consulting budgets. Those budgets carry no output metric the social role can defend.
The reasoning runs as follows. When productivity gains fail to appear in a quarterly review, the function with the weakest link to measured output is cut first. That is the social branch. The technical branch survives because it is wired to the systems that are meant to produce the gains. A second-order effect follows: the social work does not vanish, it gets absorbed at flat rates and loses its career structure. A third-order effect falls on people. Mid-career professionals in finance, healthcare and manufacturing who can bolt technical integration onto domain expertise gain. Entry-level candidates, already squeezed by the senior-heavy hiring the AIDE data describes, lose the on-ramp, since the facilitator job assumes someone who has already run the process it is meant to change.
This position would be wrong if the next two budget cycles show purely social facilitator roles holding their own pay bands and forming their own titled ladders. I would then revise the claim that the split is a wage split.
The paper's line that the work need not concentrate in a few hubs is probably right about location. Teather's team argues the work does not need to cluster in a few coastal cities. The pay is a separate matter. A facilitator in a regional plant can do the work anywhere, but whether they earn what the engineer down the road earns depends on where integration and governance spend lands. Diffusion of location without diffusion of wage is the outcome I expect. Watch the budget line, not the announcement.
What changes for the people in the middle
For a 44-year-old claims adjuster with fifteen years of domain knowledge, the paper's logic makes her scarcer. The credential she lacks matters less than the process she already understands, and an employer who can pair that knowledge with technical integration has something a recent computer science graduate does not yet have. For a 23-year-old with that degree, the picture is harder. She enters a market where the named AI roles skew senior and facilitator postings ask for someone who has already run the process.
Consulting firms and HR departments are absorbing the social-facilitator function at existing rates. That work is real, but it is priced as an extension of old jobs. The people doing it may find the title is new and the salary is not.
The geography loosens while the pay geography does not. Facilitator work can be done in more places. The premium for doing it well stays concentrated among people who combine a technical edge with domain depth, a combination scarcer than the postings suggest.
Back to the 1%
The researchers began with a productivity figure: 1% a year across the G7 since the crash, half the earlier pace. Their answer is that this is an adoption problem, the adoption problem is a facilitator problem, and the facilitator problem is a pay-and-ladder problem. The posting count does not settle that. A posting signals demand. It is where the story starts.
Over the next 24 months, the thing to watch is not whether facilitator roles grow, since they will. Watch which half survives the first budget cycle in which productivity gains fail to show. The technical facilitators will likely keep their jobs and pay. The social ones will be the first line cut from the spreadsheet. That split, more than the headline count, will decide whether the missing middle becomes a career path or a holding pen.
