I use AI intentionally every working day – and quite a range of different tools depending on the task. ChatGPT and (increasingly) Gemini assist my troubleshooting and research. I use FireCut a lot for video and it’s probably my single biggest productivity enhancer.
This is no surprise given I run an AI consultancy, but I still wouldn’t say I’m one of the faithful. I haven’t appointed an AI PA or even really bothered with meeting notetakers. I don’t really want tech taking over my life quite yet. I can’t quite find the willpower to change my long ingrained habits, even with promises of productivity. Perhaps this year I will.
As our Top 100 AI Tools list shows, there is a lot for people to learn, even if they really wanted to. Following AI founders on LinkedIn, as I do, is like swimming in a fast lane of tech FOMO. Most of us aren’t so nearly as far along. I wanted to delve into how we’re really using AI at work.
Disconnecting from the frenzied world of social media, the reality is we are still early with AI adoption, and are not at the top of the bell curve of AI adoption. I actually think, based on what’s in this email, we’re still in fairly early adoption.
Based on the data below, I’d estimate we’re moving into the early majority stage for what I’d define as ‘regular’ AI usage in the workplace.
Yet because of its general purpose nature, it is difficult to determine exact numbers for AI adoption at work.
How many people are using AI regularly at work?
- An end of 2024 IDC/Microsoft study found that some 75% of respondents ‘report current usage’ of AI (compared to 55% in 2023). But it must be noted that these respondents were large companies, and getting an up to date survey of smaller companies is difficult. Meanwhile, a mid 2024 Gartner study showed high rates of abandonment for business gen AI development, largely because of unclear business value.
- Claude/Anthropic launched its Economic Index on 10 February, stating that 36% of occupations use AI for at least a quarter of their associated tasks. But the sampling was based on Claude users, who are probably the savviest bunch of all AI enthusiasts. Claude usage makes up just 3.5% of chatbot market share.
- Getting to ChatGPT, which is by far the most adopted chatbot at about 60% market share, it now has roughly 5 billion sessions a month. This equates to 300m weekly users – which is about a similar size to X (we’ll leave you decide whether that’s big or small). Despite this footprint, it also reported to have around 10m paying users, which really isn’t many in the grand scheme of things. The New York Times has 11m, for instance. Certainly at work, you’d need paid access to get much persistent use out of ChatGPT, but as mentioned, enterprise adoption just isn’t there yet.
A study of Danish adoption published in December gives us some further clues for work:
ChatGPT is widespread in the exposed occupations: More than half of workers have used it, 41% have used it for work, and 21% have used it for a core job task. Almost all workers are aware of ChatGPT.
The unequal adoption of ChatGPT exacerbates existing inequalities among workers, Anders Humluund and Emilie Vestegaard
The Danish study shows the % of job roles that have used ChatGPT in the last two weeks – but that does not mean regular use.
Often when companies refer to AI used, they are really talking about Large Language Model (LLM) / chatbot adoption. As mentioned, 60% of that market is ChatGPT, and thus a majority of people are unlikely to be using AI for work currently, and certainly not on a regular basis. Also, is an irregular basis really ‘work’? It’s very difficult to determine.
We’re left scratching our heads with this data. Adoption is increasing, but it doesn’t appear to be in the majority. It seems probable that about 40% of workers have knowingly used AI in some way at work. Daily usage of a chatbot (the most common tool), however, is probably much lower. From the graph above, this is likely to be well under 20% across all knowledge workers.
How are we using AI at work?
Despite its relatively small size, the Claude Economic Index does go into quite specific roles that people use ‘AI’ (Claude in this case) for. It’s a Large Language Model, and the top users are all writers of some kind, particularly those who write code, with 37% of all conversations coming from those in ‘Computer and Mathematical’
Top categories and professions by % of Claude conversations.
Claude has certainly carved itself a niche of having high appreciation in software development but the findings are not distant from the Danish study, of which marketers were the top profession for adoption, and journalists then software developers not far behind.
Creative work
When we move into visual design and media editing, generative AI has become hard to avoid. Adobe Creative Cloud has integrated many generative AI features into its apps, blurring the distinction between what AI tools and software really are.
Yet Adobe has also seen share price challenges due to the rise of alternative media editing AI tools.
Descript and Midjourney are amongst the leaders, but we can hardly call their user base high. Descript’s pricing page states that ‘more than 6 million creators and teams’ use the application (read: have ever used it). The number of accounts on Midjourney’s Discord server is 20m. Contrastingly, Adobe Creative Cloud has 37m unambiguously paying subscribers (that’s 27m more than ChatGPT).
A gender divide?
It has long been lamented that AI has a diversity problem. Tech is an overwhelmingly white male profession, and not only are they building the tech, but using it too. Back to the Danish study:
Finally, we document a staggering gender gap in the adoption of ChatGPT: women are 16 percentage points less likely to have used ChatGPT for work than men in the same occupation. The gender gap is pervasive in all occupations, exists in various adoption measures, and persists when comparing coworkers within the same workplace handling the same types of job tasks.
Humluund and Vestegaard
I saw a snippet in The Week related to this back in September, referencing an Economist article:
What explains this? The study concludes that “lack of female confidence” may be to blame – women were more likely to say “they needed training”. But a separate survey, delving deeper into female psychology, observed something else: that “high-achieving women appeared to impose a ban on themselves” – viewing AI tools as short-cuts for cheats. “It’s the ‘good girl thing’.” Managers place an 8% value premium on “high-performing women with AI expertise”, so women have an incentive to relax their ban.
The Week, September
So regular workplace AI adoption is relatively low, dominated by professions that are language focused, and less likely to be adopted by women.
All things considered, we’re not in a mass adoption phase at work, but the technology is changing so rapidly that what we even have now might not even seem especially relevant in a few years time. Stay with us to keep ahead.
