Gallup's Q1 2026 workforce study found that in organizations which have implemented AI, 24% of employees say the culture improved over the past year and 25% say it got worse. At organizations with no AI in place, 59% report no change at all. The same technology produced opposite results among employees at organizations that had adopted it, and the difference Gallup measured sits with whoever is leading its use.
It doesn’t measure the procurement of licenses and fostering of training. What this is measuring is whether a task moves through the business differently ninety days later.
Teams are getting usable first drafts in minutes, research that used to take an afternoon arrives before the meeting, and the admin nobody wanted is lighter than it was. That is real, and the same tools left a quarter of the workforce feeling worse about the place they work.
What Gallup measured
Gallup surveyed workers and, separately, 102 chief human resources officers at global Fortune 500 companies. Among those CHROs, 99% call AI somewhat or very important to their strategy. Half of them, 50%, say they are not very or not at all confident in their managers' ability to guide employees on using it.
Workers who strongly agree that their manager champions AI are far likelier to say AI has transformed how work gets done, 33% against 4% among those who do not strongly agree. Gallup measured an association here, so the ratio describes a pattern.
The reachability problem behind that 50%
Half of those CHROs lacking confidence in their managers is, underneath, a reachability problem. The company can't get a consistent standard to all managers, see what any one of them is doing with AI in a given week, or tell which are leading and which are waiting it out. This also may speak to the departments who are seeing use cases to enhance their work with an agentic collaborator, rather than replace their work altogether. This takes development and enablement, a curriculum, a rollout schedule, and a measurement layer to find out afterwards whether any of it landed. By the time the program reports, the tools have moved.
In a firm of eight or twenty, nothing sits between the decision to use a tool and the people doing the work; you are in the same room, on the same projects, looking at the same client files. The expensive part of what a large company is trying to manufacture, somebody with the taste and the standing to say what good output looks like in this business, is already present and already doing that job on everything else you make.
What championing looks like in a firm of ten
Work in front of people. Run one real task through a model with the team watching, on live client work, including the places where the output was wrong and you fixed it. Show the override where your discernment takes control of the output. Ensuring adequate access to the software will limit the use of shadow AI and risk to your confidential data.
Give direction for where the returned time goes. A team that saves time with agentic work should know where they could invest those extra free hours. We find that in creative firms when shallow work is completely automated, it can be hard to work in 2-3 deep focus blocks. We recommend giving the team time back to go visit vendors, art galleries, or familiarize themselves with creative tasks related to business to fight overwhelm.
Run the failures in public. When an agent drifts, or a draft comes back sounding like everyone else's, walk the team through what happened and what you changed. These might feel uncomfortable to run, but we are working with new technology and understanding why instructions or methods change is key for everyone to move forward.
Where this goes wrong
Championing without provisioning. Enthusiasm for a tool your business has no agreement with moves client material into accounts nobody has read the terms on. Sort the sanctioned tools first, then demonstrate on them.
Enthusiasm read as instruction. If the owner is visibly keen and nobody has said which work is out of scope, people will use the tool where it doesn't belong, and the checking lands back on the one senior person reviewing everything. Say what it is for and where it stops, in the same sentence.
Time blindness to tool development. It can be easy to lose hours refining prompts and instructions when things break. Be cautious about the amount of time you take adding to instructions, because longer instructions can make a process more fragile. Protect your time and work by staying focused on the time you have to develop, repair, or innovate outside of your regular work.
Related questions
Does AI adoption improve workplace culture?
Gallup's Q1 2026 study found it goes both ways. In organizations that had implemented AI, 24% of employees said culture improved over the past year and 25% said it worsened, against 59% reporting no change at organizations with no AI in place.
Why did our AI rollout change nothing?
Often because the licenses arrived and the leadership didn't. Gallup found that workers who strongly agree their manager champions AI are far likelier to say it transformed how work gets done, 33% against 4%.
What does it mean for a manager to champion AI?
Using the tools on real work where the team can see it, saying what the returned time is for, and walking through the failures as openly as the wins.
Is a small firm at a disadvantage here?
It runs the other way. Half of the Fortune 500 CHROs Gallup surveyed say they lack confidence in their managers' ability to guide AI use. In a firm of ten there is no layer to reach, because the owner is already the manager Gallup is describing.
Do we need training first?
Training tells people what a tool does. Watching you use it on the firm's own work tells them what good output looks like in your business, which is the standard they will be judged against on the next project.
How we read this
A flat result after a rollout usually gets told as a story about the technology being oversold. We read the Gallup split the other way. The same tools produced both halves of it, so what varied was the human layer around them, and that is a fixable thing to be short of.
The role in question is unglamorous, and it needs standing in the business more than it needs technical skill. Somebody has to decide what good output looks like here, show people that standard on real work, and keep deciding as the work changes. That has always been the job. What AI changed is how quickly its absence becomes obvious.
A large company answers this with an enablement program, because it has no other way to reach two thousand managers. You can answer it on a Tuesday, with an hour of live work in front of your team.
What you can do this week
Pick one task you have genuinely run through a model yourself, and run it again with the team watching. Show the prompt, show the output, and show the place you overrode it.
Run it on live client work, because a demonstration on a toy example teaches nothing about your standard. Run it on an account the team can also use, so what they copy is something they can copy properly.
Then leave it alone for two weeks and watch who picks it up. Expect one person to copy it wrong. Ask them to walk you through their version rather than correcting the output, because the method is the thing you are trying to spread, and the people who copy it well are where a shared skill library starts.
Working together
Flow State Found works with a limited number of businesses to make their best work their baseline. You rolled AI out, two people took to it, and a year later nobody could say what had changed. We put the standard in front of the team on live work, so how you judge good output becomes something the practice runs on.
We take on limited engagements, so it starts with a conversation.
Start a conversationFor Deeper Context
- Gallup, AI's Effect on Workplace Culture, Morgan Meinen and Megan Mulherin, August 16 2026
- Gallup survey of 102 chief human resources officers at global Fortune 500 companies, fielded February 10 to March 16 2026, reported within the same research