Only 5 percent use AI in ways that change their job
Most companies shout about AI. Almost none redesign the work.
Welcome to FullStack HR, and an extra welcome to the 17 people who have signed up since last editoon.
If you haven’t yet subscribed, join the 10000+ smart, curious leaders by subscribing here:
Happy Saturday,
If you want to see how things actually get built, come hang out with me on Monday the 24th of November.
I’m doing a live webinar with 50skills where we build real stuff, in real time, with real chances of things going wrong. Which is half the entertainment.
It’s at 12 PM CET and it’s free. If you want to watch me either impress you or accidentally break something, click here.
🧠 Topics I’m engaging with
EY Work Reimagined 2025, AI everywhere but talent strategy is the bottleneck
EY has released its new Work Reimagined survey. It covers more than 15,000 employees and 1,500 employers in 29 countries.
The topline message: 88% of employees say they already use AI at work, mostly for simple tasks like search and summarising. Only about 5% say they are using AI in more advanced ways that actually change how work is designed.
EY estimates that organisations could unlock up to 40% more productivity if they combine AI investments with stronger talent, culture and learning strategies. Yet only 12% (!!!) of employees feel they receive enough AI training, and only a minority of companies are on track to reach what EY calls a “talent advantage” where people and technology are aligned.
Why it matters for HR:
A lot of “AI adoption” is shallow.
People tick the box by using AI for search or copy tweaks, but core workflows, roles and skills stay the same. The real gap is not tools, it is job design, training and clear intent. For HR, this means: stop measuring AI success by number of licences and start measuring how roles, skills, performance and workloads change. In Sweden the SCB numbers suggest many companies have bought AI but do not know why. That is a classic HR problem, not just an IT problem.
PwC Hopes and Fears 2025, daily GenAI users are pulling away from everyone else
One more survey? One more survey!
PwC’s new Global Workforce Hopes and Fears Survey 2025 is based on almost 50,000 workers in 48 economies. It finds that 54% have used AI in their job at least once in the last year, but only 14% use generative AI at work every day. (This is vastly different than EY above, I’ll get to that shortly.)
Daily users look very different from everyone else. They are much more likely to report that, in the last year, their productivity improved (92% vs 58%), their job feels more secure (58% vs 36%), and their salary increased (52% vs 32%).
At the same time, only about half of non managers say they have the learning and development support they need, compared with around two thirds of managers and more than 70% of senior executives.
Why it matters for HR:
Interesting that we get two studies the same week, with vastly different results but my two cents is that they look at this from totally different angles.
EY focuses on structures, culture, tools, skills, and why some companies get real ROI from AI. They look on hours saved, capability gaps, and the tensions that appear when AI hits the workplace. It’s very “how do we build the system so AI actually works”.
PwC studies the people. They loo at how workers feel, what motivates them, what they fear, and whether they trust their leaders. I think that explains some of the differences (even though I’m surprised by the large difference in daily use!)
But what does the PwC studie tell us then? That people with access to AI, psychological safety to use it, and good learning support are getting compounding advantages in skills, output and pay. Everyone else falls behind.
If we leave AI upskilling to “early adopters” and senior leaders, the gap will widen fast. The priority is to design inclusive AI capability programs, especially for frontline and non managerial workers, and to link those programs to clear career paths, not just one off training.
Learning AI feels like a second job, and that’s the real adoption problem
Matteo Cellini writes about something many people (or at least people I know and I also include myself in this) feel but rarely say out loud.
Learning AI is (hard) work. Not “fun extra learning when I have time” but actual cognitive load. He describes the pressure to stay updated, the constant stream of new tools, and the sense that everyone else is moving faster. It’s relatable because it matches what people I meet every week tell me: they want to use AI, but the effort it takes is invisible, and most companies don’t account for it.
Why it matters for HR: If AI learning feels like a second job, people will avoid it. And adoption will stall. The answer isn’t to push more tools but to make learning intentional. I’ve been talking about this in my AI Adoption series. Small routines. Clear expectations. Less “figure it out” and more guided paths. I still enjoy experimenting and staying current, but the key is to make it a choice, not an endless chase.
One third of companies say they will replace HR with AI
Ok, one final survey. But with a slightly different angle. And it’s a survey from AI Resume Builder, based on 1,250 business leaders and it finds that 30% of companies say they plan to replace some HR roles with AI in 2026, and another 21% say they have already replaced at least some HR roles with AI this year.
Why it matters for HR:
I’m torn on this one. The methodology is a big vague and one would like to know the current state of HR in these companies. Is it one person working in HR? Two? A whole department? Is business leaders = CEOs with mandate or is it leaders in the org wishing that they would need less HR?
That said, this supports a claim that I’ve been doing for the last 2,5 years - HR is not guaranteed to stay the same in all of this AI transformation that is happening and we need to re-think and re-position ourselves to see where we provide value (which we always have had to do!).
⚡ Important updates
GPT 5.1 rolls out in ChatGPT and the API
OpenAI has released GPT 5.1, with two main variants: GPT 5.1 Instant and GPT 5.1 Thinking. Instant is the default “fast” model, now more conversational and better at following instructions. Thinking is the higher reasoning model that adapts how long it “thinks” based on task complexity, so it can spend more time on harder work while staying faster on simple questions.
NotebookLM launches DeepResearch
NotebookLM (by Google Labs) has launched a new Deep Research feature that acts like a smart assistant: you give it a query, it builds a research plan and browses multiple sources to draft a detailed, source-rich report. It also now supports a wider range of file types including Google Sheets, Microsoft Word, images of handwritten notes, and links to Drive files, making it possible to analyse structured data and varied content directly.
Zoom acquires BrightHire to boost interview AI
Zoom has signed an agreement to acquire BrightHire, a company that provides AI tools for interviews. BrightHire will bring interview planning, automated notes, real time coaching, and analytics into Zoom’s platform. The deal is expected to close in the coming weeks.
The gap between buying AI liences and actually redesigning workflows is massive. Most orgs treat AI like a new spreadsheet formula when it requires rethinking entire job structures. That 5 percent stat from EY is telling becuase it shows that real adoption isn't about tech access but about courage to redesign roles. The differnce in daily GenAI users from PwC is striking too, those people are compounding advantages while others wait for perfect training that may never come.