We argue badly, and nothing accumulates. How could we do better?

Millions of people argue every day about the things that matter most — climate change, what to do about AI, how we might build a better world. Some of it is sharp, even insightful. And almost none of it accumulates.

Read a long thread on Twitter, or a busy forum, on any of these. There is heat, and occasionally light. But at the end you cannot point anywhere and say: we have now settled this sub-point, and here is what it implies for the larger question. The good arguments and the bad ones scroll past together, and next week the same debate starts again from the beginning. We never build the cathedral; we just make a bigger mess.

I have wanted a tool to change this for more than ten years. Long enough that I almost built it myself, three or four years ago — and then didn't. This is the story of why, and why I think its time has finally come.

What rigour makes possible

It does not have to be this way. I know, because I have seen the opposite.

I grew up watching my father argue legal cases, and what struck me was what becomes possible when real structure and real resources are brought to bear on a single question. In a courtroom you can actually tie things down: take a claim apart, demand the evidence for each piece, go back to the source and check it.

The historian Richard Evans showed this on a grand scale. For years David Irving had got away with distorting the historical record on the Holocaust. Then, in the libel trial Irving himself brought, Evans spent two years documenting precisely how — falsified footnotes, quiet mistranslations — until the court could follow the evidence and pin the lie down. He wrote it up in Telling Lies About Hitler.

That is what rigour and resources buy you: the ability to make a claim stand or fall on the evidence rather than on who argues longest. The tragedy is that we have almost none of that machinery for the questions we care about most in public.

What I wanted was a graph

Two things had frustrated me for as long as I can remember.

The first is vagueness. Almost nobody breaks down what they are actually claiming. We gesture at positions instead of stating them, and a claim that is never made precise can never be examined. I would read a two-hundred-page book and, at the end, be unable to tell you its central thesis — let alone the sub-claims it rested on.

The second is the lack of accumulation — the thing I started with. Evidence arrives, but it is never clear which claim it even bears on. Nothing compounds.

What I wanted, against both, was a graph: an overview showing the claims and how they relate — a map you can stand back from to see the whole shape of a disagreement, then zoom into the one node where the real work needs to happen.

I didn't have a name for it. I just knew the shape of the thing I was missing. Then, in my early thirties, I found Minto pyramids and issue trees, which pointed at exactly what I had been reaching for. They were useful at once — and slow and laborious to build by hand, so laborious that in practice I mostly didn't. That is when I first thought: there should be an app for this. I got as far as seriously considering building one.

I didn't. Partly timing, partly that the tooling to make it feel effortless wasn't there yet.

Logic trees, and why now

Recently I was shown logic trees — the Logical Thinking Process, out of the Theory of Constraints. I hadn't encountered them before. They felt like issue trees with more power: not just the decomposition of a claim, but a whole pipeline from goal to problem to solution to the concrete next step. The recognition was immediate.

This time, something is different. Two things.

The first is AI. The laborious part — the part that always defeated me, and everyone else — is exactly what AI is now good at. Decomposing a claim, drafting a first tree, reading a document and proposing where its arguments attach: this used to take hours of disciplined effort. It no longer has to.

The second is an idea I keep returning to: the tree as a stable scaffold that discourse accumulates against. Documents come and go; forum threads scroll away. But if there is a persistent structure that new material maps onto — this quote supports that node, this study complicates that branch — then discussion stops being disposable and starts to compound. This is the cathedral, built at last, brick by brick.

The hard part isn't technical

I should be honest about what stands in the way, because it isn't the technology.

Structured thinking is confronting. It takes discipline, it forces you to decide what actually matters, and most of all it exposes misalignment: the moment a group gets precise about its goal is often the moment it discovers it never agreed on one. Staying vague is comfortable. Getting clear is not. I have felt all of this myself, including in organisations I have run — and it is a big enough problem that it deserves its own post.

But I am convinced the prize is worth it. If we can make the disciplined part far cheaper — minutes of AI-assisted work rather than a lost afternoon — then many more people will pay the smaller price that remains.

For a long time I waited for someone else to build this. I have stopped waiting. With David Joseph I have started, and the first rough pieces already exist: you can explore an early demo of the claim-tree and annotation idea. If you want to test it, send me the messiest argument you know, and we will see what the tree makes of it.