On May 28th, ClimateAction.tech partnered with UsTwo and RevAIsor to run a small workshop on values-driven risk assessment for AI development. The following summary was written by the event facilitator, Karol See.
Why we ran this session
As companies are rushing to build AI solutions, with AI coding tools becoming more accessible, more teams are starting to build faster, often without a clear framework for thinking about risk and societies it can affect. This session was designed to help people at different stages of their product development lifecycle start asking better questions before they build, not after.
What we did
With a good mix of participants, we started with a roundtable to understand where each person’s interests and hesitance towards AI lie. Mainly focused on how you think AI can help, and how you think AI can harm. That last question set the tone for the rest of the session.
From there we looked at the broader governance landscape, from UK AI principles and the EU AI Act, through to company values, how it can match your own personal ethos, and how this shapes how teams make decisions day to day. With this discussion, it became clear that regulation can only take you so far. Frameworks like the UK AI Principles, NIST, and the EU AI Act provide important guardrails around issues such as safety, transparency, accountability, and fairness. But they don’t tell organisations what they should build, only what they should avoid, and that’s where values come in. We discussed how company values, codes of conduct, and broader social goals can help teams navigate decisions that regulation alone can’t answer. Questions like: should we build this feature? Who benefits? Who bears the cost? And does this align with the future we want to help create? Those decisions often sit beyond compliance and squarely within organisational values.
The main activity used an AI Canvas to map out a hypothetical product: a personal learning agent that builds your personal growth through curated news, videos, e-learning platforms, and various content. The group filled it in from a standard product perspective first: opportunity, audience, data sources, success criteria.
Then we asked: who else gets affected?
That question opened up a much richer conversation. The group identified various impacts, and one discussion kept resurfacing throughout the exercise. If an AI career coach is designed to help people make better decisions, what happens when millions of people receive guidance shaped by the same data, assumptions, and incentives?
The group explored how a tool, which is intended to support personal growth, could also influence hiring markets, content creation, and even cultural norms. Rather than simply helping people discover unique paths, it could gradually steer users towards increasingly similar definitions of success. That raised a broader question: when does personal optimisation start creating societal homogenisation? It was a useful reminder that the biggest risks in AI often emerge not from individual interactions, but from the cumulative effects of systems operating at scale.
Having these questions out in the open does not ultimately intend to deter people from building AI tools, but rather, have a more holistic approach to their AI strategy and ensure that they’re looking at the effects through various lenses.
The key takeaway
Good AI design isn’t just about building for your target user. It’s about understanding the wider ecosystem your product sits in, and being honest about who might bear the costs of what you’re building.
And it doesn’t stop at launch. The questions you ask at the design stage will look different once you’re in development, and different again once your product is in people’s hands. Revisiting your assumptions at every stage of the lifecycle is what separates a responsible product from one that just had good intentions at the start. That responsibility doesn’t sit with one person or one role either. Whether you’re writing code, defining features, managing a roadmap, or shaping a brand, the decisions you make have real consequences for real people. The workshop reinforced the simple idea that responsible AI starts with asking better questions. The sooner teams start asking to do that, the more likely they are to create products that benefit people without unintentionally leaving others behind.