Get new posts
A business, an educator, a community or an innovator with something to offer? Let's build this together.
Get in touchJoin the community for new pieces, the best of the discussion, and what is coming up.
Just transition is one of the most borrowed phrases in climate policy, and one of the hardest to deliver. Now it's turning up in debates about AI and work. I went looking for who's already working on it, and what nobody has worked out yet.
Just transition is one of the most borrowed phrases in climate policy, and one of the hardest to deliver. Now it's turning up in debates about AI and work. I went looking for what it would actually involve, who's already working on it, and what nobody has worked out yet.
This post opens a new thread in our Foresight writing. It maps the ground rather than arguing for a position, and later posts take each question in turn.
The phrase "a just transition" didn't start in a boardroom or a government department. It came out of the North American labour movement in the 1970s, and it's most often credited to the union organiser Tony Mazzocchi, who argued that workers in polluting industries shouldn't pay the price of cleaning them up. Over four decades it moved from union halls into international policy. In 2015 it entered the preamble to the Paris Agreement. The same year, the International Labour Organization adopted guidelines for a just transition, agreed by governments, employers and workers together.
The ILO framework rests on four pillars: social dialogue, social protection, rights at work and employment. Put plainly, the people affected get a say, they aren't left without support, their rights hold, and there's decent work on the other side.
That history comes with a bit of an obligation. Agent Build Club is a small company, not a union. We're borrowing a phrase working people built, to see whether it can help us think about where AI is taking work. Climate campaigners already warn that it's being stretched into a label for any transition that suits whoever is using it. So we should use it in the sense it was made for: the people doing the work at the heart of the decision, not only on the receiving end of it.
It'd be easy to say the just transition is still only an idea. That's not quite right. It's written into treaties, set out in guidelines and backed by commissions. The harder truth is that it's recognised on paper and difficult to deliver.
Scotland is a useful example, because it has tried harder than most. It set up an independent Just Transition Commission to advise government. In 2023 the commission concluded that the current path wouldn't deliver a just transition, and named the Grangemouth refinery as an early test of whether the idea meant anything. As the closure approached, one commissioner told MPs at Westminster that Grangemouth hadn't done well on that test, and that a just transition should be done with people rather than to them.
The lesson isn't that the idea fails. It's that principles, commissions and good intentions don't deliver a transition on their own. Early planning does, and so does the trust of the people whose work is changing. Both take time, and both are hard to build after the announcement.
If it's this hard with a known site, a known workforce and a closure date, AI is harder still. Four differences stand out.
And there's a fifth. In climate, most people agree on the destination, even when they argue about the route. With AI there's no agreed picture of where we're heading, which makes it harder to say what fair would mean on the way.
I went looking for who's already working on this. Here's what I found under the rock. Plenty of people are working on versions of it, even when they don't use the phrase. This is a first map, not a complete one.
Unions. In the UK, the TUC has published a pro-worker AI strategy. It asks for a share of productivity gains to reach workers, which it calls a digital dividend, and for workers to have a say in whether and how AI is used, starting with its model AI Bill. Globally, IndustriALL frames AI explicitly as a just transition question, and is prioritising algorithmic transparency, skills, health and safety, redistribution and organising power.
Bargaining that has already happened. The 2023 Writers Guild of America contract, agreed after a 148-day strike, was the first collective agreement in a creative industry to set rules for AI. AI can't be credited as a writer, writers decide whether to use it, and studios must disclose AI-generated material. It's the clearest example so far of worker voice turned into enforceable rules.
Economists. At MIT, Daron Acemoglu, David Autor and Simon Johnson argue that the direction of AI is a choice, not a given. Their 2026 paper sorts technological change into five types and finds only one, technology that creates new tasks for people, unambiguously good for workers. They set out nine policy directions, from tax reform to protecting worker expertise.
Industry and civil society together. Partnership on AI's Shared Prosperity Guidelines give developers and employers tools to assess an AI system's likely effect on jobs before it's deployed.
Government. In June 2026 the UK government launched the Early Careers Jobs Alliance, backed by £20 million and co-chaired by the general secretary of the Prospect union. Its job is to map how entry-level work is changing and how to keep routes into work open. It has promised an initial report this autumn and a final report next summer.
Researchers using the phrase. In Singapore, Singapore Management University and the National Trades Union Congress are developing a tripartite framework for a just AI transition, with a specific focus on agentic AI. In the United States, Jobs for the Future is funding pilots of early-career pathways designed to hold up as AI spreads.
Across all of these, the proposals fall into five families. Each deserves a post of its own, and each will get one.
None of these is new. What's new is applying them to a change this fast and this widely spread.
There's one angle I'd add, and it's the one Agent Build Club cares about most. Most of these proposals treat the people doing the work as people to protect. They're also the people most likely to find a better way to do it. Anyone who looks at how something is done now and finds a practical way to improve it with AI is an innovator.
That changes the question. Disruption is real, and proper disruption is painful. But we already know a lot about introducing new technology well. Everett Rogers spent a career studying how innovations spread, and his Diffusion of Innovations showed that people take up new ideas faster when they can see the benefit, try them out, and fit them to the way they already work. Innovations that land are user-centred and human-centred. People pull them in because they want them. They don't have them pushed onto them.
Done that way, adoption doesn't have to be slower. It can be faster. You don't do innovation to people. You do it with them, and often they do it themselves.
I've seen this close up, watching innovators in local authorities work through their net zero plans. In December 2023, Newham Council became the first council in the UK to approve a just transition plan, putting its residents at the heart of the borough's route to net zero rather than treating carbon targets on their own.
What I noticed is how change actually travels inside an organisation. It starts with innovators, the people trying new things at the edges. They reach the champions first: the people inside the organisation who can see what could be better, and who bring others with them. Champions are the ones who turn one person's experiment into the way a whole team works. They're also the people we most want to find at Agent Build Club.
So maybe a just transition is less about cushioning change. Maybe it's about helping more people see what could change for the better, for everyone, and making the space for them to get involved as innovators themselves. That's the kind of democratic design I care about now: not only asking people about decisions that affect them, but widening who gets to invent.
Most of the proposals on the map work at national level or inside a single firm. Education usually appears as a supplier: the place that produces skills for employers to use. It's rarely treated as a place where the transition itself gets worked out.
That may be a gap. Universities are where learners, researchers and employers already meet, and where people first form their picture of what a profession is. If the routine work that used to teach judgement is going, someone has to design what replaces it, and it isn't obvious that employers or governments can do that alone. Universities could also be the space where more people get to try being innovators, on real problems, before the change reaches them.
I think this is where we can be most useful. I'm holding that lightly, though: it's a question to test, not a claim we can defend yet.
These are the questions this thread will take on. We don't know the answers yet.
Next in this thread, we'll start with voice, because the Writers Guild gives us the clearest worked example of people shaping how AI enters their work.
In the meantime, two questions. First, who have we missed? If you know an organisation, agreement or piece of research that belongs on this map, tell us.
Second, the question that started Agent Build Club. Where you work or study, who's deciding how AI agents are introduced into the work? And are the people who do that work part of the decision?
Follow our Luma calendar and join our free online events, starting with our talk series on the future of engineering, and engineers, in the age of AI. Follow us on LinkedIn for news. It's a broad church, on purpose. If you're a business, an educator, a community or an innovator with something to offer, get in touch.
And if you're the champion in your organisation, the person others come to when they want to try something new, I'd really like to hear from you.
