Inaugural Edition · June 2026
Where AI has shifted how People teams work in the first half of 2026, where it stalled, and what leaders are watching next. Drawn from 208 People leaders and a panel of operators driving AI inside their own organizations.
The picture at mid-year
The People function is in a position no other team quite shares. It has to adopt AI for its own work, and it has to help everyone else in the organization do the same. Almost half of leaders told us their team's AI work right now is either doing that double duty already, or is focused solely on supporting the organization with AI adoption.
What stands out in this first pulse is how far AI has spread inside People teams, and how little of that spread feels like leadership. AI is in employee comms, recruiting, analytics, onboarding, performance. Yet only about one in eight feel their function is ahead of the rest of the organization on AI.
The gap between activity and confidence runs through everything that follows. Most of the people who answered this survey have spent ten or more years in the field, which means they are operating AI from a foundation built well before it existed. That experience is an asset. It also means the work ahead is as much about reinvention as it is about adoption.
Who we heard from
Alongside the survey respondents, we drew on Clarinet's work with more than 70 organizations on AI adoption, and on the expertise of our three panelists driving AI inside their own teams at LifeLabs Learning, Stitch Fix, and Pyn.
Watch
The mid-year panel with Diane Sadowski-Joseph (Clarinet), Stacey Nordwall (Pyn), Stephanie Shuler (LifeLabs Learning), and Sari De (Stitch Fix).
Part One
Where AI has changed how People teams operate in the first half of 2026.
Teams are using AI in 3.6 areas of the People function on average. More than three-quarters use it in more than one area, and nearly half use it in more than three. In Clarinet's study of AI superusers, a diversity of use cases is one of the clearest signals of value from AI.
Comms is the common entry point because it's qualitative and low-stakes to test. Compensation sits lowest, where the data is most sensitive and the tolerance for error is smallest.
Asked where AI has delivered the most value, 83% named a specific area. Two stood out, tied for top place: drafting and analytics.
Results above based on the 173 leaders (83% of the full data set) who named a value area. Eight more areas — policy, recruiting, L&D, performance, automation, knowledge, coding, and other — make up the rest. 17% said it's too early or they haven't deployed yet.
"Streamlined communications — helping to refine drafts and reduce iteration has been the biggest win. Same with comparing data and streamlining document or deck creation."
"Our data tends to be all over. The thing I get excited about is when people build a dashboard, pulling together this spreadsheet and this platform, to get the data they want, how they want it. Something purpose-built for them."
A caution that came up live: analytics is also where fluency matters most. Throw raw data at a general model and ask it to "analyze this," and the odds of a confident wrong answer climb. The value is real; so is the need to know how to ground it.
The panel was clear that "leading" doesn't mean becoming the organization's AI expert. The strongest framing was about playing to the function's real strengths: change management, enablement, workforce planning, and the human side of adoption.
At the same time, the function's reach is growing. Most AI work still sits inside the People team, but 45% of leaders say their AI work already extends beyond it — supporting the wider organization, or split evenly between the two.
"It's not our job to be the experts in AI. In L&D, I'm not the expert in harassment law, but I'm still responsible for making sure everyone is trained on it. My job is to create the space that enables the learning to happen."
"AI can be the Trojan horse we've all been waiting for, for people to take the human stuff seriously. Now that everyone feels insecure, it's easier to make the case for psychological safety as a business mandate."
Part Two
Where the tooling wasn't ready, where rollouts stalled, and where teams rethought their approach.
70% of respondents named a place where AI fell short of what they'd hoped. Among those who gave a specific reason, the single largest was readiness: skills, training, and strategy. Output quality and integration followed.
The panel made the same point from the field, and from frustration. One described a scheduled task in Claude that pulls a weekly report — and routinely doesn't.
"The report runs every Monday, and that's probably when I swear the most all week. Sometimes Claude just decides not to pull the data. It flags something I fixed weeks ago, and I've told it every week since. But that's also how I learn what the tool can actually do well right now."
The reframe that landed hardest was about discernment. The early question — can AI do this? — gives way to a harder one.
"It became less about 'can AI do it' and more about 'should this work even exist?' Before we throw AI at something, does it need to exist in its current form? If we were starting today, is this still what we'd build?"
That kind of reinvention asks people to let go of processes they built and were proud of — what one panelist called the IKEA effect. The skill the function has to teach, in Sari De's words, is the willingness to start over.
"I took a pottery class. My bowl wasn't perfect, the teacher fixed it, and it was perfect. But I didn't join to make a perfect bowl. I joined to learn how. So I pulled the clay off the wheel and started again. When I think about my work in L&D and how it connects to AI, it's less about teaching people more ways to use AI, and more about teaching them to rewire their brain so they're comfortable pulling the clay off the wheel."
That sounds like churn until you look at what happened next. About 60% of them simply switched to a different tool. Almost no one abandoned AI. Tools came and went; the commitment stayed.
It's a useful thing to remember the next time a leadership meeting gets stuck arguing over which tool to standardize on. Tools are not forever — we're not still asking Jeeves. The panel's view was that purpose-built tools earn their place in some functions and not others, so a blanket "everyone uses the same thing the same way" rarely fits.
"Some tools are purpose-built and some aren't. I wouldn't expect every department to be at the same place. For engineering, the expectation might be reworking whole workflows. For another team, it's experimentation and learning. A blanket approach doesn't make sense."
Policy is the soft spot. Only about 1 in 4 leaders has an AI policy that covers the People function and is working; the rest have one that needs work, are building one, or have nothing yet. But an effective AI policy is part of a broader system — training, governance, and safe tool access — not a standalone fix. In the field, the policy is too often asked to do every job at once, when what it needs is supporting instruments around it.
Roughly three in four (74%) describe a policy that isn't fully working.
Part Three
The top worries and where leaders are putting their focus for the rest of 2026.
Asked for the one thing they're most worried about, leaders pointed at data and the pace of change. Job loss came in far down the list, at 6%, despite the headlines.
92% named a worry; 8% said nothing or weren't worried. Five more areas — governance, ethics, cost, time, and other — make up the rest.
"Too many different systems and privacy of our sensitive data."
The panel's read on data risk was consistent: the biggest exposure isn't the enterprise tools a team pays for. It's shadow use — someone reaching for a personal account because the sanctioned path is too slow. The answer that worked for them was safe containers, not blanket bans.
"Shadow AI is the biggest risk factor for data leakage. So get specific about what you're actually worried about. People sit at two ends — everything is dangerous, nothing is dangerous — and the truth is in the middle."
Asked what they're most focused on getting right in the second half of 2026, leaders pointed past experimentation toward making AI stick. The top three are about scale, fit, and what comes after the first wins.
Of all 208 respondents. 92% named a focus; 8% weren't sure or had no specific goal. The rest splits across policy, governance, training, strategy, and tool selection.
"Shifting to an AI-first mindset — what does that mean, what does it look like day to day, how do we measure it?"
That question — what does good AI usage and impact look like day to day, and how do we measure it — is the through-line into the second half of the year. The panel's closing answer was to measure the right thing.
"AI is as much about improving outcomes as improving processes. We're getting teams to focus on impact — not just that the process is better, but what changed because of it."
The way forward
The data and the discussion point in the same direction. The work that moves AI forward in the People function is less about the next tool and more about the conditions around it. Four places to put your attention.
The function's edge is change management, enablement, and workforce planning — not being the org's authority on prompts. Create the conditions for learning, and measure the same outcomes you always did. As Sari mentioned, when digital advertising was new, teams still measured conversions and churn, not how many banner ads someone made.
A small shift in how you ask makes this concrete. "How are you using AI in your function?" invites a conversation you can't lead. "How are you setting your people up to succeed in a function where AI is part of the work?" is one you can.
Early adoption asks whether AI can do a task. The higher-value question is whether the task should exist at all, and whether the workflow around it should be redesigned rather than sped up. The trap is one-for-one swaps — letting AI do steps two and seven of a process no one has re-examined. Don't build faster horses.
Readiness was the top reason AI fell short, and readiness is mostly about people, not platforms. That means investing in leader fluency so they can show judgment, and helping experienced teams let go of processes they're attached to. It also means giving people more ways in, not just more time. A simple anchor that works: ask people what they'd do with 30% more capacity, then work backward.
The real data risk is shadow use, not the tools you've sanctioned. Bans push people to personal accounts; clear, approved paths bring them back. Build containers that let people experiment within defined limits, invite "I want to use this" requests instead of punishing them, and get specific about what's actually risky rather than treating everything or nothing as dangerous. And stop asking the policy to do every job — it needs supporting instruments around it.
In closing
This first pulse shows a function already deep in the work: using AI across the function and, for many, reaching beyond it. The opportunity now is to turn that activity into impact. Where AI disappointed, the cause was rarely the model — it was readiness, discernment, and the conditions for adoption, all of which the People function is well placed to build.
The shift we encourage is from individual AI use to systematized use: collective efficiency rather than everyone building their own version of the same tool. The People function is positioned to lead that shift, not by becoming the technical expert, but by doing what it has always done well — bringing the humans within a workforce along, and measuring what changed because of the work. We'll take this pulse again at the end of the year.
About Clarinet
Clarinet helps teams and organizations operationalize AI adoption. Through hands-on training and strategic advisory, we help people embed AI into the flow of work and adapt as technology evolves. Our team of experts blend AI, behavioral change, and operational experience to deliver solutions that are practical and tailored to each company’s unique context. We've worked with 70+ companies on AI adoption, and we built this survey to add data behind the conversations we have with People teams every week.
Get in touch →Methodology
This report draws on a survey of 208 People and HR professionals, fielded alongside Clarinet's "State of AI in the People Function" webinar on June 17, 2026. Respondents came from Clarinet's network and the panelists' communities. This is a community sample, not a probability panel, so figures describe the people who responded rather than the field as a whole. We report it that way on purpose.
Closed questions are reported as a share of respondents. Open-text questions were hand-coded into a single, mutually exclusive bucket per response, so the coded shares sum to 100% within each question. Where a question had a meaningful share of non-answers or "too early to say" responses, we quantify that group and show the base the percentages are calculated on, rather than dropping it. Per-question bases are noted on each chart. Figures are rounded and may not sum to 100%.
The survey was confidential, with quotes used only where respondents opted in to attribution. Quotes shown here are anonymous. The panel quotes come from the live, recorded session. This pulse is designed to be repeated and to evolve: questions will build on these findings rather than stay fixed, so the series can track how the field actually moves.
