As part of the UK Government's AI Champions programme an independent report AI Adoption Plan: Creative Industries was published this week. It is worth reading for the 8 recommendations as a strong articulation of an approach for accelerating AI adoption responsibly and confidently.
There's also an eye catching stat that 51% of creative businesses are using AI, with the source data indicating this is dominated by two tools (text generation/LLMs at 27.7% + visual content creation at 23.7%). However, I'd gently question measuring adoption by tool uptake at all. These figures may tell you who is experimenting, but they tell you very little about who's getting real value, and who's just moved the effort somewhere less visible.
It is common to see the value-gap in the AI "false dawn" optimism generated by one-shot pilots, happy-path-only demos, and tool dumping... (or, more generally, AI for the sake of being seen doing AI). More often than not, this is driven by senior stakeholders' demands (and subsequent expectations), rather than people grounded in the pragmatic reality of the work, with predictable results. This is probably the most common failure mode right now, exemplified by messages like "we've rolled out the tools, now get on with it".
My experience has taught me that adoption isn't a clean efficiency curve. It's jagged and messy.... the hard part is almost never the model, the tool, or identifying opportunity... it is behaviour change, mindsets, and trust. Setting up for success by re-imagining a business or a process, rather than automating or bolting AI on to an existing solution, is typically the harder path (again, mainly because it's a human and ambiguity challenge, rather than a technical one), but it also generates and sustains greater value.
The report acknowledges this in its challenges section and case studies. It's also refreshing to see the productivity mirage called out ("...struggle to distinguish genuine efficiency from displaced effort").
The lesson from my AI transformation work in 2025 and 2026 is that the context/knowledge/data preparation, the impact of new failure modes, allowing time for multiple iterations, legal and security bottlenecks, and a longer term view on correcting/sustaining outcomes can quietly absorb the time and costs the AI workflow was forecast to save. This is typically more apparent in the short-term (more "discovery" and "optimisation" required), so initial headline savings are over-estimated (with a knock-on effect that the long-term impact is then under-estimated). It also pushes use cases towards more operational and coherent use cases, which can also be seen in the data underlying the headline stats (see attached chart).
Nonetheless, I have no doubt at all, that much of this is early-adopter friction, but it's great to see the conversation happening. As a transformation optimist, I find it helpful to remember Ethan Mollick's articulation that "today's AI is the worst AI you'll ever use". I also like his thesis around moving from Co-Intelligence to Co-Existence.
A version of this post was originally published on LinkedIn.
New writing by email
Occasional pieces on product, technology and AI — and how they actually play out in practice.
