Your Managers Decide Whether AI Helps or Hurts Your Culture
Sign Up For Daily Job Market News
Gallup published new research yesterday that settles an argument I hear from clients all the time: does putting AI into a team make the culture better or worse? The honest answer is neither, most of the time.
Among U.S. employees at companies that have implemented AI, 51% say their culture stayed the same over the past year. Of the rest, 24% say it got better and 25% say it got worse. A 24-25 split like that is closer to a coin flip than a real pattern.
At companies that have not adopted AI, 59% report no culture change at all. So bringing AI into a workplace does make some kind of shift more likely, though which direction that shift goes looks like a toss-up. Of the employees who noticed a change, 46% said AI, automation, or robotics played a real role in it.
Here is the number that stopped me. Gallup surveyed 102 chief human resources officers for its Global CHRO Roundtable, and 99% of them called AI somewhat or very important to their company’s strategy. Then Gallup asked those same CHROs about their own managers, and 50% said they are not very confident, or not at all confident, that their managers can guide an employee through using AI at work.
Every CHRO in that room has bet on AI. Half of them do not trust the people who have to make it work day to day.
The confidence gap shows up directly in the employee data. Among workers who strongly agree their manager champions AI, 33% say it has transformed how work gets done at their organization. Among workers who do not strongly agree, that number drops to 4%.
The pattern repeats on culture. Employees with a manager who actively supports the team’s AI use report improved culture at 31%, against 21% for everyone else. Ask about a big improvement specifically, and it is 9% versus 3%.
The manager decides the outcome here far more than any feature the software ships with.
Companies are responding, at least on paper. Gallup’s CHRO survey found 78% encourage peer learning and experimentation with AI, 62% are building centers of excellence or naming internal AI champions, and 57% provide AI training for managers. All reasonable moves. None of them free up a manager’s calendar.
Gallup makes this point itself: managers already handle expanding responsibilities and wider spans of control, and asking them to lead AI transformation on top of that risks turning a genuine priority into another competing demand. I would go a step further. Training and champions programs are additive. They hand a manager one more thing to learn without removing anything from their plate. We see the same pressure play out in the manager-level searches that land on our desk: the job description keeps growing, and the support underneath it does not grow with it.
If you are the one deciding whether to roll AI out to a team, put manager capacity in that conversation before you put the tool in anyone’s hands. Somebody owns adoption day to day once the rollout happens, and it is worth asking, out loud, whether that person actually has room for it. This is the same tension we flagged in our look at worker trust and AI anxiety earlier this year: the technology moves faster than the support structure around it, and employees feel that gap before leadership does.
If you are the one using the AI, the 33% versus 4% gap is worth remembering the next time you are choosing between two offers. Whether AI makes your day-to-day better or worse will likely come down to your direct manager more than the tool itself, so ask about it in the interview. Our earlier data pull on what AI is actually doing to jobs, wages, and hiring goes deeper on how unevenly this technology is landing across roles.
Staffing your team doesn’t have to be hard.
Reach out and see how we can help.
The Government’s Own Job-Openings Data Is Running on a Quarter of the Sample It Used to Have
While Gallup was measuring what AI does inside a company, a separate story broke Friday night about whether we can even trust the numbers describing the job market from the outside. The Kobeissi Letter posted a chart showing how few businesses still respond to the government’s main survey of job openings, and the government’s own published data backs it up.
The chart, sourced to Pantheon Macroeconomics, puts the JOLTS initiation rate (the share of businesses the Bureau of Labor Statistics approaches that agree to participate in the Job Openings and Labor Turnover Survey) at roughly 1 in 4. This sits near its lowest level since 2022. Before the pandemic, that rate ran between 50% and 70%.
Getting a company to agree to participate is step one. Getting its numbers in on time is a separate problem.
Pantheon’s data puts the first closing response rate (the share of firms reporting before BLS publishes its initial estimate) at about 30%, near a record low and roughly half what it was 10 years ago. The second closing rate, after late responses come in, lands around 35%.
BLS confirms the same trend in its own establishment survey response-rate tables. The monthly JOLTS unit response rate was 35.0% in December 2025, down from 66.0% in January 2016.
The high point over that stretch was 67.0% in July 2016. The record low, 29.7%, hit in September 2025.
Put plainly: the monthly job-openings number that moves markets and leads headlines now comes from a historically small slice of a roughly 21,000-establishment sample, then gets revised as slower responses trickle in. The August 4 release proved the point in real time.
June openings came in at 7.4 million, and the same release revised May down by 57,000, from an originally reported figure to 7.5 million. We covered that same report, and the low-hire, low-fire pattern underneath it, in our June JOLTS breakdown.
Kobeissi’s conclusion: “Official US labor market data is becoming increasingly unreliable.”
I would stop short of that line. “Increasingly unreliable” reads to a lot of people as “made up,” and one of the top replies under the original post says exactly that. A falling response rate widens the error bars around a number. It does not mean anyone fabricated it.
The real issue is how that first number gets used. It moves nearly every month, and a 57,000-job revision to May, buried in the same release that announced June, is routine rather than exceptional. Markets and headlines treat the first print as settled fact and largely ignore the correction three weeks later.
My advice, unchanged from what I have said about every other noisy data release this year: stop making hiring decisions off a single month’s print. Watch the 3-month direction instead, and weight what you can observe firsthand a lot more heavily than any government release.
Application volume per posting, time to fill, and how fast candidates accept offers tell me more about the market right now than a JOLTS headline does, and those numbers update in real time rather than on a monthly lag. Our recent look at the July jobs report made a similar case: one number rarely tells the whole story.
What deserves attention now is whether BLS gets the funding to fix its collection process. A voluntary survey with a falling response rate is a budget and staffing problem at the agency before it becomes a statistics problem for the rest of us.
