If you spent this week accepting AI drafts without pushing back, or fixing errors nobody asked you to catch, you may be doing invisible work that nobody counts as a skill. On September 21, IBM’s Institute for Business Value published a global study of 1,500 chief human resource officers and 8,800 employees. HR Dive covered the findings on September 23 under a blunt headline: the human skills CHROs call essential are the same ones workers say are eroding fastest.
The gap is not about learning another tool. It is about critical thinking, problem framing, and the judgment to supervise, validate, and override outputs when they are wrong. That is a soft-skills story with hard career consequences, especially for remote workers and freelancers whose value is measured in decisions, not seat time.
What CHROs want versus what workers prioritize
IBM’s study lands at a moment when many teams already use AI for first drafts, research summaries, and routine analysis. Leaders and individual contributors are reading the same dashboards and drawing opposite conclusions about what matters next.
| Finding | CHROs | Employees |
|---|---|---|
| Supervising, validating, overriding outputs is essential | 71% | 29% rank judgment as important |
| Critical thinking matters in an AI-enabled workplace | 57% | Worries about erosion cite it most often |
| Human judgment is a top capability | 48% | 3 in 4 who worry say skills are already declining |
Read that table carefully. The capability leaders treat as non-negotiable is the one workers rank lowest. That is not laziness. It is a signal that daily workflows reward speed over scrutiny.
IBM senior VP and CHRO Nickle LaMoreaux put the framing in the company’s release: as AI takes routine tasks, uniquely human capabilities become more important. CHROs, she argued, must redesign work so people focus where they have the greatest impact. The study suggests many organizations have not done that redesign yet. Nearly half (46%) do not involve the CHRO when AI strategy is defined, which helps explain why judgment stays off job descriptions even when leaders say it is vital.
Invisible work and the thoughtload problem
The IBM numbers align with a separate piece published September 23 by organizational psychologist Liane Davey in The People Space. Davey argues that faster AI output does not automatically reduce workload. Prompting, checking, integrating, and taking responsibility for results create what she calls thoughtload: the cognitive and emotional burden of staying accountable when production speed jumps.
IBM found that 80% of CHROs believe AI adoption creates invisible work, validating recommendations and fixing errors. Nearly half of employees agreed AI added to their workloads. Significantly fewer said managing those outputs would be an important skill in the future.
That mismatch shows up everywhere remote teams work. A developer merges AI-generated code without a review checklist. A marketer publishes a client newsletter after a light skim. A freelancer sends a scope document that reads well but misses a constraint the client mentioned on a call. The work looks done. The judgment layer was skipped.
We covered a related pattern in work stress and role clarity on September 18. Unclear roles make people redo work because success was never defined. Skipped judgment makes people redo work because nobody built verification into the workflow. Different root cause, same exhaustion.
Where confidence rises or falls
IBM’s study offers one operational clue. In workplaces where judgment is built directly into how work gets done, 62% of CHROs reported growing employee confidence in AI-enabled decisions. Where judgment was not integrated, 57% reported confidence declining.
Organizations that clearly define workflows as human-led, AI-assisted, or AI-executed reported 18% risk reduction and 20% quality improvement. Those are leadership design choices, not individual willpower exercises. Still, individual contributors and freelancers can copy the pattern at smaller scale.
Human-led: You own the outcome and use tools for prep only. Final call stays yours.
AI-assisted: You generate options, then apply an explicit review standard before anything ships.
AI-executed: The tool runs a bounded task with fixed inputs and you spot-check against a sample.
Without those labels, everything blurs into “use AI when helpful,” the same vague instruction that fueled ambiguity in September’s stress research.
Why younger workers face sharper risk
HR Dive’s coverage notes that Cangrade’s 2026 report found younger workers scoring below average on critical thinking, attention to detail, and problem solving. IBM recommends rotations, supervised decision-making, and coaching for early-career talent. Freelancers can approximate that: document your reasoning on one project per quarter and ask a peer to challenge a major decision before you finalize it.
Soft skills that protect judgment
The IBM study and Davey’s thoughtload frame point to habits you can start this week without waiting for HR to rewrite workflows.
Name the decision, not just the deliverable. Before you send work, write one sentence: “I decided X because Y.” If you cannot fill in Y, you have a judgment gap, not a formatting problem. Clients and managers notice when you explain tradeoffs. That behavior connects to the reliability employers ranked highest in personality and hiring research earlier this month.
Build a personal review checklist. Three to five questions you run before every outward-facing output: accuracy against source material, audience fit, risks if this is wrong, and what you would override if you had more time. Checklists sound mechanical. They are how judgment becomes repeatable instead of heroic.
Cap concurrent threads. Davey warns that one person cannot stay fully accountable for too many disparate AI-assisted streams. Freelancers juggling four clients feel this first. Limit how many open AI drafts you will validate in a day. Finish scrutiny on one before starting generation on the next.
Make override normal. In structured hybrid teams , we wrote about writing decisions down so remote colleagues stay aligned. Post in shared channels when you rejected a suggestion and why.
Ask for judgment criteria in discovery calls and interviews. Our guide on where to apply for remote work helps you find listings. This week’s IBM data tells you what to ask next: how does this team define human-led versus AI-assisted work, and who validates outputs before they reach customers?
What not to do
Do not treat speed as proof of competence. Finishing faster while skipping review creates the invisible work CHROs describe, and you rarely get credit for the fixes.
Do not assume judgment will stay sharp by osmosis. IBM found 60% of employees worry about skills erosion, and three in four of those say decline has already started. Skills atrophy when workflows skip the thinking steps.
Do not outsource accountability because the draft looked confident. Tone is not evidence. The gap between CHRO priorities and employee rankings exists partly because polished outputs feel finished before they are verified.
A practical playbook for this week
Pick one recurring deliverable: weekly report, client email batch, code review, design revision. Label it human-led, AI-assisted, or AI-executed. Write a five-question review checklist and use it on every instance this week. After two cycles, note one override you made and share the reasoning with a teammate or client contact.
If you manage people or lead a client engagement, audit one workflow for invisible validation work. Ask who fixes AI errors today and whether that effort appears in timelines or performance expectations. If it does not, you are burning judgment without building it.
If thoughtload feels high, cut one concurrent AI stream before you cut hours. Quality of scrutiny matters more than volume of drafts.
The bottom line
September 21 through 23 brought one of the clearest employer-employee disconnects of the month. CHROs say critical thinking and judgment define success in an AI-enabled workplace. Most workers do not rank those skills as important, even while many report they are already eroding.
For remote workers and freelancers, the lesson is immediate. Your edge is not generating more output. It is the verified decision on top: what you checked, what you rejected, and what you stand behind when something breaks.
Judgment is a soft skill with measurable outcomes. IBM linked it to confidence, risk reduction, and quality. Davey linked its absence to thoughtload and burnout. Practice it on purpose this week, in writing, before the gap between what employers need and what you deliver quietly widens.