Relieving pressure

At Friends of the Earth, our Experiments team is small. That means we’re constantly looking for ways to reduce pain points, unstick ourselves when we get bogged down, and expand our capacity. AI has become a surprisingly useful ally in this.

One concept that inspires our approach is the theory of constraints, popularised in Eliyahu Goldratt’s process-improvement thriller, The Goal. The idea is simple: identify bottlenecks, and focus on easing those to improve the system as a whole. That’s how we’ve been using AI: not to replace work, but to relieve pressure points and accelerate where we usually get stuck.

Producing discovery reports

Creating exploratory reports on emerging topics or macro-trends used to take weeks. Several team members, multiple workshops, and plenty of coordination went into producing a 10–20 page report.

Now, with AI assistance, we can generate a draft report in just a few hours, something ready for human sense-making, covering:

• Organisational SWOT analysis

• Trend snapshots

• Signals and drivers

• Alternative scenarios

• Systems analysis (including causal loop diagrams)

• Three Horizons exploration
  

We run this using an iterative ChatGPT project folder based process, feeding each section back in as it develops. The AI critiques its own outputs, we critique those critiques, and together we refine both the report and the process. The result: faster first drafts, stronger prompts, and more space for the team to focus on meaning and implications.

Prototyping

Early-stage prototypes, like storyboards, used to eat up mornings. With custom GPTs, we now generate them in minutes. The speed and consistency this brings means we can test more ideas, more often. We’ve even made some of these GPTs available via our blog for others to try.

Compounding

Borrowing from a concept described in Every, we’ve found value in what we call compounding: giving feedback to AI and asking how the prompt or process itself could be improved.

Take our weekly UK innovation funding scan. Early attempts produced long, unfocused lists of irrelevant grants. By iterating on the prompt for the task, the outputs are now sharper, more targeted, and more consistently useful.

Early agentic experiments

We’re also beginning to explore tools with agentic capabilities - AI that can take steps towards completing a task. One example is Comet, a web browser built on Perplexity.

Right now, we use it to take the funding opportunities identified in our weekly scan and automatically populate Trello cards with criteria, deadlines, and links for quick evaluation.

These tools remain immature. They often fail, introduce new risks, and need close human oversight. But they hint at what’s coming.

Capturing ideas on the move

AI isn’t just for the desk. As a keen runner, I often find myself turning over half-formed ideas out on the hills. To capture these, I built a simple Apple Shortcut: dictate a thought, and ChatGPT transforms it into a neatly written Apple Note.

The results are sometimes whimsical, often poetic, and always thought-provoking. They’ve turned what would have been fleeting trail-side notions into notes I can return to and build upon.

Closing reflections

Across all these uses - reports, prototypes, compounding, early agentic tools, even running notes - the theme is the same: AI helps us get unstuck. It doesn’t replace the thinking or the judgement, but it lightens the load at the chokepoints, giving us back time and attention for what matters most.

Author: Christian Graham, the Experiment’s team co-lead at Friends of the Earth. Find out more about their work.