Pce

In 2026, Anthropic’s operational judgment failed twice in the same way, leading to the leak of the Claude Code source code ↗. The constitution asks Claude to imagine how a “thoughtful senior Anthropic employee would react”, but what happens when the organisation’s structure fails? Anthropic publishes its constitution ↗ along with research about where the constitution works and where it does not. The current version is an ethical treatise addressing Claude discussing safety, ethics, Anthropic’s guidelines, and helpfulness, in that order when they conflict. Anthropic favours cultivating good values and judgment over strict rules. I co-develop the Perseverance Composition Engine ↗ (PCE), an open source multi-agent AI system that solves the problem with a structural approach called Artificial Organisations ↗). PCE assumes the agents cannot be relied on to be honest/harmless/helpful/etc and structures the system so that the inevitable bad behaviour doesn’t surface. ...

I am using my personal Perseverance engine as I help develop the code, and I’m watching carefully to see how useful it is for developing analysis, review and writing. Evidence so far is mixed, but improving fast. I feel in control as I do with any other work tool, which I certainly do not when using a typical error-prone AI text interface. One of the reasons I feel in control is because there are more controls in place, that is the point of the Artificial Organisations concept. But another reason is that this tool is becoming more tuned to me all the time. ...

AI is slow, and Agentic AI ↗ is even slower. I develop an MCP server ↗ that generates PDF documents, and I work with the Agentic Perseverance Composition Engine daily. Tasks that take maybe 5 microseconds on an operating system (eg, does a file called Things-to-Do exist?) can take a million times longer, between 2 and 5 seconds, because each operation requires multiple round trips to a remote LLM (often with timeouts.) It’s a young, unstable stack, comparable in maturity to early MS DOS or the Apple ][. When AI gets hold of your data via an MCP server it can do interesting things, but it is not put together well. ...

AI ethics and safety work mostly focuses on making individual models smarter and better-behaved via guidelines and persuasion, with not much hope ↗ this will succeed. You should especially not feel safe when an AI company reassures you about their guardrails ↗. When you hear guardrails think of telling a dog “Don’t bite the furniture inside the house today”, because you can never know what will actually happen. The concept of Artificial Organisations doesn’t require AIs to be reliable, it ensures that when an AI goes wrong there are hard limits on how much damage it causes. Similarly, we can put the dog outside the house, so no matter how bitey it is the furniture cannot be bitten. I spent a good deal of 2026 seeing if this concept could make AI less dangerous and more useful. I even had it study me as if it were an apprentice. This is mostly the opposite to Anthropic’s idea of a constitution. Science Fiction saw this problem years ago, and it’s a great story. But first, the Perseverance Composition Engine ↗ is an instance of Artificial Organisation theory. We constructed AI personalities which to check each others’ work under strict limits of what each can do. Consul is the front desk personality in my own organisation. Consul can’t do much for itself, but it can coordinate the activities of Critic, which reviews things and Composer, which writes things, and Corroborator which is the fact checker. Each of these lives in their own little world, with access only to the minimum tools and materials they need. When errors occur, one or more of these agents usually notices and the piece of work is run through the organisation again. I explained to Consul I was looking for a human story to explain how these human-style organisations worked, and supplied my initial idea. Consul considered my suggestion, and then sent me the following… ...

There are many problems with the AI billions of people use in 2026, discussed endlessly at all levels of society. From the end of 2025 I became interested in the particular problems of ethics and reliability, and why the approaches taken by all of the large AI companies are not good enough. Predictability, or ‘alignment’ as they call it, is just not something we can expect from this type of AI. A colleague started working on a very different approach from these companies, and from February 2026 I have been contributing to and using prototype versions of the Artificial Organisations ↗ concept. This article explains why I believe Artificial Organisations are a promising new direction. Multi-agent Agentic AI is pretty important, as described here by the UK government ↗, but it is rarely done well. If you want to try for yourself, you can use the core research code ↗, as I do daily. ...