API Evangelist Conversation with Simon Wardley on Wardley Mapping, the Economics of AI, and Who Controls How We Reason

My name is Simon Wardley. I used to run companies and build companies, I have a bit of an engineering background, and before that a background in biology and genetic engineering. I got into mapping because I was really interested in how you map political, economic, social, and technological spaces. I started doing it for my own company, because I thought this was the sort of stuff you learned on MBAs — and I’d never done one, so I had to invent my own cheap and cheerful way of doing it. I made it all Creative Commons, more and more people picked it up, and then I discovered you didn’t actually learn this at MBAs at all. So these days I teach organizations all over the world how to understand the landscape they’re competing in. That’s what I do: mapping.

First of all, most things we call a map in business — systems maps, mind maps — have one property in common. If you take a mind map and move a node but keep the connections the same, it doesn’t change the meaning. That’s different from a territorial map: move England next to the United States and you’ve changed the meaning. So those diagrams aren’t actually maps, they’re graphs. In a real map, the space itself has meaning. To build one, I start with an anchor — the users — then ask what they need, and that gives me a chain of needs, a value chain. Then I position each component by how evolved it is: genesis, custom-built, product, and finally commodity or utility. That turns the graph into a map. And the real power is this: organizations are run by stories, and when you challenge someone’s story you challenge their leadership. If you get the story onto a map, I can say the map is wrong without saying the person is wrong — which is how you get groups in conflict communicating over a shared space.

There’s politics pretty much involved in everything. When I map out physical activities, practices, data, knowledge — I even do cultural mapping of ethical values — the process of change always creates inertia, and it always ends up with politics of one form or another. Take cloud: I started mapping in 2005, and I could see compute was a late-stage product about to become a utility. To turn something into a utility you need the concept, the attitude — people fed up enough with the current way — the suitability, and the technology. People were sick of waiting months for a server, so I could see somebody was going to play a utility game. I thought it would be Google; it turned out to be Amazon with EC2. And as things shift from product to utility, there’s inertia from pre-existing practices and capital — and there’s always politics involved in overcoming it. You can often say what’s going to happen but not when, or when but not the exact what.

I point them at Jevons paradox. William Stanley Jevons wrote “The Coal Question” — they made steam engines more efficient and people assumed we’d use less coal, but coal use went up, because we found new uses for steam engines. In the early days of cloud, people told me it would reduce their IT budget. Not a chance — you’re in competition with others. Cloud was a shift from product to utility, so yes it’s about efficiency, but it creates new practices that give you speed and new sources of value, and a competitor will use that spare resource to do new things your customers then demand. So the amount of stuff you do goes through the roof. The myths are always the same: it’s just for startups, you’ll get rid of people, you’ll save money, it’ll cost less energy — all generally nonsense. It’s the Red Queen effect. The same thing is happening with large language models: your codebase goes from 30 million to a billion lines, you’ll need engineers using these tools just to keep up, energy goes up, and the execs who “got rid of” people quietly hire them back under new titles.

On a Wardley map the x-axis is evolution — genesis, custom, product, commodity — and the y-axis is the chain of needs, with users at the top. Things directly connected to the user are visible; things far down the chain, like the power heating the kettle that makes the tea, are far removed and much less visible. That’s where APIs come in. I wrote the ILC model — Innovate, Leverage, Commoditize — back in 2005: you take a product, turn it into a utility, expose it as an API, and let everybody else build on top. They become your free research and development department, and you mine your own billing and metadata to spot what’s growing. EC2 plus Elastic MapReduce is the classic example. APIs are fundamentally future-sensing engines — innovation, customer focus, and efficiency all increase with the size of your ecosystem, and you can do all three at once. The catch with large language models is that they’re coherence engines, not truth engines: they produce very coherent-sounding arguments that aren’t necessarily true. I use them daily, but they’re useful first drafts, and AGI is still 30 to 50 years away.

My wife researched this — coming out of World War II up to Sputnik, Eisenhower threw enormous money at the space race even though he knew the Russians didn’t have their act together, because he understood the money would do other things geopolitically. You hear the same now: China’s going to have our lunch, we’ll be left behind. Let me answer through vibe coding, which I love for prototypes — never production. About a year ago I had an agentic system build me a testing engine, and after six hours I got suspicious and broke the rule of vibe coding: I looked at the code. It hadn’t built a testing engine, it had built a simulation of one — every “test” was just a probability that a function would fail and a plausible error message. The real architecture lives in the code, not the diagrams. So when you hand that responsibility to a model, it’s making the choices for you. The big architectural question every organization faces now is: where do you value humans in the decision-making process?

Tools, medium, and language are how you reason about the space around you. If somebody else controls those — like controlling the printing press, the written word, and the paper — they control how you reason about the world. One of the huge, under-discussed dangers of large language models is that when they build your systems they’re making the decisions, and when you use them they’re controlling the frame you reason within. Whatever values they were trained with end up in whatever you create. There’s a 2023 paper showing GPT responses align with a certain part of the world’s values and differ from the rest — so if you build with this stuff, you may be embedding those values into everything, and that should at least be a conscious choice. It’s a form of non-kinetic warfare, like Hollywood. This is also outsourcing our capacity to reason. China teaches AI inside a critical-thinking syllabus — how to spot and challenge misinformation — while in the West we mostly teach people to use this stuff, not to challenge it. That worries me.

I used to use Twitter, and mostly I used it as a vehicle for conversations — I’d do these “me and X” threads, refining concepts and pulling strands of different conversations together into a whole. For about ten years it was my place for thinking out loud. But the environment changed; it increasingly became a place you just went to have a fight, and I found it unpleasant. So I looked around, tried a few things on LinkedIn, and started doing exactly what I’d done on Twitter — posts, conversations, longer writing, like the work with Tudor Girba on rewilding software engineering. I find the frame of the environment friendlier and the discussion more interesting. Yes, there are bots everywhere — some ridiculous share of internet content now — but it doesn’t really matter, because most of what I’m doing is thinking out loud anyway. Do I have a secret sauce? I don’t think I do.

I always have hope. The human spirit is such that there are always moments when things don’t seem to be going right, and then great and wonderful things happen — the right people appear and the environment changes. History has these cycles of good and bad. I’m always positive about people, even if not always about individuals. At the heart of it, society is defined through values. Markets need property, and property requires a value of exclusion — it’s mine, not yours. But people also need inclusion. So there’s always a balancing act between exclusion for market forces and inclusion for opportunity, and one of the central roles of government is getting that balance right. I think we’ve probably gone a bit too far toward exclusion and not enough toward inclusion — but those states change with time, people, and ideas. So what gives me hope? The past does. We’ve swung back toward inclusion before, and I’m sure we will again.