The AI Skills Gap Is A Collaboration Gap, Not Technical

‍

New 2026 hiring data shows collaboration and change-management skills climbing faster than technical ones on AI roles. Here's why that changes who you should be hiring.

What the job postings actually show

The instinct when an AI rollout stalls is to assume the organisation lacks technical depth, not enough people who can build or fine-tune a model. New research from the Josh Bersin Company's 2026 talent work complicates that instinct. Across AI-titled job postings, the fastest-climbing skill requirements are not model-building or data engineering. They are cross-functional collaboration and change management (Josh Bersin Company), skills that have nothing to do with the technical layer everyone assumes is the constraint.

Where AI rollouts actually stall

Anyone who has sat inside an organisation trying to operationalise AI recognises the pattern this data is describing. The model gets built, or the vendor tool gets licensed, inside a matter of weeks. What takes months, and often never fully resolves, is getting three departments to agree on whose workflow changes first, who owns the exception cases the model gets wrong, and who is accountable when the tool's recommendation contradicts a manager's judgement. None of that is a technical problem. It is a coordination problem wearing a technical costume, and it is the actual reason so many AI pilots stay pilots.

Why this is a builder's advantage, not an operator's

This is where the distinction between builders and operators in an organisation starts to matter more than most hiring processes account for. An operator executes the workflow as designed. A builder notices where the workflow breaks, redesigns the handoff, and removes the friction before it compounds. As AI absorbs more of the execution layer of work, the remaining differentiated value concentrates in exactly the group willing and able to do that redesign work, not the group fastest at using the tool day to day.

That is a specific, testable claim, not a general one about soft skills mattering. Cross-functional collaboration in this context does not mean being pleasant in meetings. It means being the person who can sit between an engineering team's model output and a frontline team's actual workflow and translate one into the other without losing what matters in either direction. Change management does not mean running a training session. It means anticipating which team will quietly route around the new tool in three months if nobody addresses the incentive that makes routing around it rational.

What this means for how organisations hire and train

Most enterprise AI hiring is still built around technical screens: can the candidate build or evaluate a model. Fewer organisations are screening for the builder pattern described above, largely because it is harder to test for in a structured interview than a coding exercise is. The Bersin data suggests that gap in screening is becoming the more expensive one to leave unaddressed.

For training providers and L&D functions, the implication cuts the other way. AI literacy programmes built purely around tool proficiency are training people for the wrong half of the bottleneck. A programme that pairs technical AI fluency with structured practice in cross-functional negotiation and workflow redesign is training for the part of the job that is actually scarce.

The open question

Whether it is better to hire the strongest technical AI talent and hope they can learn to carry a room, or hire someone who already carries a room and teach them the technical layer, is not a question the market has settled. The evidence so far points toward the second path being underweighted relative to how much organisations are currently paying for the first.

‍