Swarochish Chekuri https://swarochish.com Writing, research and notes on reducing the distance between intent and execution. en-GB Fri, 14 Aug 2026 09:00:00 +0000 A General Theory of Human-System Interaction https://swarochish.com/research/human-system-interaction https://swarochish.com/research/human-system-interaction Thu, 14 Aug 2026 09:00:00 +0000 Felt ease treated as a regulation problem rather than a design problem. Five channels, complexity as allocation, and friction classified by what it produces. Trust Across Boundaries https://swarochish.com/research/trust-across-boundaries https://swarochish.com/research/trust-across-boundaries Fri, 01 Aug 2026 09:00:00 +0000 An eight-layer model of trust with strict dependency ordering, spanning human, agent-human and agent-agent interaction. Broken at the Seams https://swarochish.com/research/broken-at-the-seams https://swarochish.com/research/broken-at-the-seams Fri, 01 Aug 2026 09:00:00 +0000 Multi-agent systems fail where control, context or commitment crosses a boundary. Ten design principles for that layer. The distance between intent and execution https://swarochish.com/writing/distance-between-intent-and-execution https://swarochish.com/writing/distance-between-intent-and-execution Fri, 01 Aug 2026 09:00:00 +0000 Why the gap between wanting an outcome and having it survives every improvement in speed. State, not persona https://swarochish.com/notes#n09 https://swarochish.com/notes#n09 Fri, 01 Aug 2026 09:00:00 +0000 Personas are a planning tool that quietly becomes a runtime model, and as a runtime model they are wrong in a particular way. They assign a person a fixed identity when the thing that actually varies is their situation. A warning will not fix a slip https://swarochish.com/notes#n08 https://swarochish.com/notes#n08 Fri, 01 Aug 2026 09:00:00 +0000 Three kinds of failure, three different remedies, and applying the wrong one reliably fails. A slip is a correct intention executed wrongly, and it wants undo, forcing functions and better defaults. A mistake is a correct execution of a wrong intention, and it wants the mental model repaired rather Friction should be classified by what it produces https://swarochish.com/notes#n07 https://swarochish.com/notes#n07 Fri, 01 Aug 2026 09:00:00 +0000 Friction is treated as a quantity to be reduced, which makes it impossible to reason about the friction that is doing work. Some of it produces nothing and should go. Some produces attention, deliberation, consent, skill or error containment, and removing it makes the system feel better and behave w Who absorbs what was removed https://swarochish.com/notes#n06 https://swarochish.com/notes#n06 Fri, 01 Aug 2026 09:00:00 +0000 Simplification programmes are usually measured on the party that got simpler. That is the wrong side of the ledger. Complexity has a floor set by the task, and below that floor the only thing a system can do is move work from one party to another. A closed question is not a finished one https://swarochish.com/notes#n05 https://swarochish.com/notes#n05 Fri, 01 Aug 2026 09:00:00 +0000 The mechanism that closed the natural language generation problem exists and is widely available. The next version of Sherlock my co-founder and I wanted is now possible. The answer arrived from somewhere else https://swarochish.com/notes#n04 https://swarochish.com/notes#n04 Fri, 01 Aug 2026 09:00:00 +0000 Natural language generation was the weak link at Penser. Everything up to the finding worked. Saying the finding back to a person in ordinary language did not. Most enterprise architecture is an artefact of a vanishing constraint https://swarochish.com/notes#n03 https://swarochish.com/notes#n03 Fri, 01 Aug 2026 09:00:00 +0000 The shape of business software was set when computation was expensive, integration was painful, and every capability had to arrive as a separate application with its own screens, its own data and its own login. Once and ten thousand times are different engineering problems https://swarochish.com/notes#n02 https://swarochish.com/notes#n02 Fri, 01 Aug 2026 09:00:00 +0000 Doing something once tolerates a person watching, correcting, and quietly handling the case nobody anticipated. Doing it ten thousand times does not. Autonomy is earned, not granted https://swarochish.com/notes#n01 https://swarochish.com/notes#n01 Fri, 01 Aug 2026 09:00:00 +0000 The permission a system holds should track the consequence of what it does. A summarisation can go wrong quietly and cheaply. A payment cannot. The more consequential the action, the more permissions, observability, determinism and reversibility the system needs before it acts unsupervised.