Boston Consulting GroupJune 3, 2026
Vinciane Beauchene, Sylvain Duranton, David Martin, Vanessa Lyon, and Jeff Walters
Why it matters. BCG's fourth annual survey of nearly 12,000 workers and leaders: 67% say AI has taken over simpler tasks, 72% say skill expectations have changed, and 36% feel adequately upskilled. Hold the last two together. The requirement moved for roughly three in four people and the support reached about one in three, and that distance is where capable employees quietly fall out of a role they used to be good at. What is left once the simple tasks go is review, direction, and decisions, which is harder work than the work it replaced and is almost never described that way in the job posting it gets added to.
McKinsey Global InstituteNovember 25, 2025
Lareina Yee, Anu Madgavkar, Sven Smit, Alexis Krivkovich, Michael Chui, Maria Jesus Ramirez, and Diego Castresana
Why it matters. More than 11 million US job postings analyzed, with roughly 72% of today's skills shared between the work that could be automated and the work that still needs a person. The overlap is the finding, and it argues against the clean split most workforce plans are built on. Skills are not being retired so much as relocated: the same abilities get spent framing the problem, interpreting the result, deciding, and handling the exception instead of producing the first draft. That makes which skills survive the wrong question to plan around. Where in the workflow a person's skill is being spent is the one that is actually moving.
OECDJune 18, 2026
OECD
Why it matters. The OECD's case for skills-first hiring arrives with its own warning attached: assessing skills is hard, and bias can quietly restore the credential gatekeeping the approach was meant to open. The number to plan around is that nearly half of surveyed employers in OECD countries with available data expect to use pre-employment tests by 2030. Demonstrated capability is becoming standard practice rather than the alternative to one, which raises the stakes on build quality. A badly designed assessment is a credential with worse paperwork. The thing tested has to be the work, and someone has to check who it screens out before it runs on a full pipeline.
Stanford Institute for Economic Policy ResearchJuly 2026
Neale Mahoney, Erika McEntarfer, and Karsen Wahal
Why it matters. Little evidence so far of significant AI-driven job losses in the aggregate, with early-career declines in some exposed occupations treated as suggestive rather than causally settled. Both halves matter. The aggregate reading is the corrective to the displacement story, and the early-career signal is the half that touches hiring directly, because entry-level roles are where the bounded tasks were concentrated. Nothing here licenses a headcount forecast in either direction. What it supports is narrower and more useful: expect the composition of jobs to move faster than the count of them, and expect the first visible movement at the door you hire through.
