AI Skills Gap: Xobin's 2026 Report Rewrites Hiring Criteria

Xobin's Human vs AI Skills Report 2026 Mid-Year Edition, published in mid-September 2026 with data collected through 30 June 2026, sets a precise threshold: 72% of the skill groups in its 2024 hiring framework are delegable to AI, fully or in part, under at least one of the models tested. The scope is stated openly: 683 skill groups, 113 leadership scorecards from 92 employers, thousands of technical assessment requests.

Read as an AI skills gap report, the document describes hiring criteria, never jobs: the text itself specifies that the study excludes any measure of roles eliminated.

What the report actually measures

The sources are the internal records of the Xobin recruiting platform. The method compares each skill group against the capabilities of several AI models and classifies it as fully, partially or not at all delegable.

Reading the 72% correctly requires a distinction. Partial delegation means a person still has to frame the task, review the output and handle the exceptions, as the report notes in coverage by HRD America^[1].

The sample has a declared limit. It reflects the clients of a selection platform, meaning organisations that already hire with structured assessments. Generalising to the whole labour market would be a methodological error.

Delegability shifts by category

The aggregate figure hides a wide spread. Analytical reasoning and problem solving come out as 90 to 100% delegable. Operations and execution drop to 28%, while leadership and interpersonal skills stay below 50%.

This distribution overturns the implicit hierarchy of many job descriptions. For years analysis was the premium skill and execution the commodity. Today the models cover the former and leave the latter exposed.

The consequence for hiring teams is direct. The person's residual value concentrates where delegation fails: in framing the problem, judging the output and handling what the model gets wrong.

Job descriptions seek direction, less execution

Across the 35 technical roles analysed, the ability to translate business requirements into technical solutions appears in 10-19 job descriptions and assessment requests. It ranks among the five most cited skills, behind AI integration and automation and analytical problem solving, each present in 30-35 roles.

The signal is structural: employers pay for the translation between need and solution, the act of direction that precedes execution.

The report reads this skill as direction: the person defines what the business needs, the model produces the solution, the person verifies it.

The distance between 10-19 and 30-35 also tells the story of a lag. Almost every organisation asks for AI integration. Around half, or fewer, explicitly ask for the skill that makes that integration useful to the business.

Leadership scorecards weight emotional intelligence at 52%

The sharpest figure concerns the selection of leaders. In the 113 scorecards from 92 employers, emotional intelligence criteria carry an average of 52% of the total score, more than all the other criteria combined.

The weighting follows a logic consistent with the delegability map: interpersonal skills stay below 50% delegability, so the market prices them as scarce.

One caution remains. A scorecard measures what recruiters say they are looking for, and interview practice can diverge. AI screening applied to criteria this subjective amplifies the selection biases this desk has already documented.

The technical test changes shape

In the first half of 2024, AI-free coding tests accounted for 75 to 100% of technical assessment requests. In the first half of 2026 the share fell to between 25 and 50%. AI-assisted tasks (generating, fixing, explaining code) rose from under 25% to a band between 50 and 75%.

The report warns that these are shares of requests, rather than absolute counts, and that the mix of employers may differ between the two periods.

Xobin's founder, Guruprakash Sivabalan, ties the ability to catch AI errors to the retention of fundamental knowledge. An analysis published in MIT Sloan Management Review^[2] describes the same risk as a "capability mirage": the apparent competence produced by the model's output outstrips the real understanding of the person using it. An assessment that measures output and ignores understanding selects for the mirage.

Redefining the criteria is an organisational decision

The evidence draws a gap between two populations of organisations with limited overlap. The first keeps testing what the model already does, and hires people who compete with AI on the ground where AI wins. The second has moved the criterion towards direction, verification and exception, and hires people who remain necessary when the model gets it wrong.

The difference between the two is a designed organisational condition, never an individual choice by the candidate. The candidate brings the skills the market has asked of them in recent years. The selection criterion is decided by the organisation.

The McKinsey^[3] paper on scaling agentic AI reaches a parallel conclusion from the operational side: the return depends on redesigning work around the systems, more than on the technology as such. An unchanged hiring criterion is the quietest form of work never redesigned.

What changes for decision-makers

For the CEO, the conversation to bring to the board concerns organisational readiness: what share of hiring criteria measures the ability to direct and verify AI, and what share still measures delegable tasks.

For the CHRO the priority is twofold. Rewrite job descriptions around the translation between business and technical solution, and pair every AI-assisted assessment with a test of fundamental knowledge that makes the capability mirage visible. The work also concerns people already on the payroll: converting internal talent towards direction skills remains the more efficient route compared with the external market.

For the CFO the useful figure is the cost avoided. Redesigning the criteria costs a review of scorecards and assessments. Hiring on the wrong criterion costs an entire cycle of selection, onboarding and replacement. For the Talent & Compensation Committee, the metric to monitor is the share of scorecards and job descriptions updated after the delegability map, with a review date.

The design question for anyone hiring in 2026 is a single one: which part of the work must the candidate be able to do independently, and which must they be able to direct and control?