Human Creativity vs The Machine: Six Lessons from our GenAI Pilot

As a studio we are committed to Human-First Authorship. Practically this means anything released to the public or to partners will be crafted by a human. This human-first approach is central to our craft and we don’t expect it to change.

For the Pilot we imagined GenAI not as a tool to replace our craft, but as a way to unlock new creative opportunities and give teams new capabilities. For example, across our key areas of interest, we asked ourselves:

• Prototyping: Can we democratise the making process by giving everyone the ability to bring their ideas to life – creating better space for creative challenges and diverse conversations.
• Playtesting: Can we create new ways to roleplay and test ideas out quickly through a range of in-person and simulated playtesting.
• Visual development: Can we create new approaches to the visual exploration process and innovate in how we visually communicate ideas and direction.
• Research & Strategy: Can we use models to help us synthesise complex domains, evaluate signals vs trends and test strategies across emerging markets.
• Meetings and facilitation: Turning the spoken ideas in a meeting into immediate visuals and structures.

From the outset, the team were given full access to frontier models to be explored within our internal R&D and early-stage partner projects.

After 12 months of cross-departmental use, we asked the team to reflect on what worked – and what didn’t. The results were nuanced, encouraging, and at times challenging.

Rather than a single headline, six key insights emerged that helped us understand where GenAI genuinely adds value – and where restraint matters most.

The most frequently cited success was how GenAI lowered the barrier to making. Non-technical team members were able to spin up high-fidelity prototypes – mostly Gemini-based  – without waiting on specialist support.

The impact wasn’t just speed; it was authorship. Ideas could be shown, tested, and discussed earlier, shifting conversations from abstraction to experience.

Across disciplines, GenAI proved especially powerful as a way to start. Team members used it to structure brain dumps, scaffold early thinking, or generate a first pass at code or text that could then be shaped and refined.

This was particularly valuable for colleagues with dyslexia or communication hurdles, where the friction of starting can outweigh the difficulty of editing.

Routine tasks – formatting data, writing helper scripts, cleaning up text, summarising technical information – were consistently flagged as strong wins.

These automations freed time and cognitive energy for higher-value creative decisions, and they felt least controversial precisely because their scope was clear and their outputs easy to evaluate.

Specific implementations stood out – particularly within the art pipeline (via ComfyUI) – because the results didn’t feel generic or disposable.

In these cases, the craft wasn’t in the generation itself, but in how the tools were directed. Prompting required finesse: a clear creative intent, well-chosen constraints, and an existing vision to aim at. GenAI was embedded into established creative systems and workflows, amplifying direction rather than inventing them wholesale.

Used this way, the tools didn’t replace authorship – they extended it. Creative control stayed firmly with the team, and the outputs felt deliberate, recognisable, and owned.

Alongside the gains, a recurring concern was the recognisable “stink” of AI in final artefacts – especially decks and visual outputs that feel verbose, generic, or lacking human terseness.

The risk wasn’t failure – it was dilution. When generation becomes abundant, editorial pressure can drop. “Good enough” becomes easier to accept, particularly in low-stakes contexts, and standards can quietly drift. The differentiator remains taste, clarity, and intentional restraint.

The most uncomfortable reflections weren’t about what GenAI could do, but what it might quietly erode. Several team members noticed that when problems were solved for them rather than by them, the learning didn’t fully stick. Creative and technical growth depends on productive friction. Remove that friction too early, and understanding can become shallow. This risk isn’t immediate failure, but gradual cognitive debt – skills used less often, instincts that dull over time.

There was also unease about environmental cost. Inevitability alone wasn’t seen as justification. If generation becomes effortless, restraint becomes essential – choosing when the value outweighs the cost, and when human effort is the more responsible path.
These weren’t blockers, but signals – reminders that judgement, responsibility, and long-term thinking can’t be delegated.

Balancing these forces means being clear about what ‘good practice’ looks like for us – and putting the right governance in place to support a new kind of quality control.

As our internal GenAI practice evolves, our focus is on setting the conditions for thoughtful, playful, human work to coexist alongside powerful new tools – without surrendering the judgement that defines our craft.

GenAI is going nowhere. Neither are our taste, restraint, and craft.

Stay tuned for more updates in the coming months – not as final conclusions, but as part of our ongoing experiment.