Essays & reflections
A working collection of ideas.
On building intelligent systems, making sense of them, and deciding what’s worth doing.
All essays
32 essays · newest firstThe Symmetry You Choose
The choice of what should remain invariant can matter more than the model that exploits it.
Two Compressions
Finding a pattern in observations is one achievement. Discovering the mechanism that generates it is another.
The Integration Problem
Moving AI into professional practice creates new demands for expertise, evaluation, and accountable integration.
The Regulator's Mirror
Metacognition as a governing layer: how an agent might monitor and regulate its own operation.
Intelligence Beyond the Next Token
The useful structure inside a model can be richer than the output channel through which we ask it to speak.
The Mapmaker's Alphabet
An agent inherits a conceptual vocabulary from human expression. What can it do with maps it did not make?
The Agent, Not the Task
A case for designing persistent, bounded, adaptive agents rather than organising everything around a task and a score.
The Logic of Saying No
A credible refusal makes a commitment observable. In markets built on trust, that can be part of the product.
Beyond the Encoding Gap
Metafeatures, distributed search, and the possibility that AI captures more of discovery than its finished outputs.
The Residue and the Trace
What separates the recorded products of expertise from the experience through which an expert learns to see.
Why AI Makes Organisations Slower Before It Makes Them Faster
When production accelerates and judgment does not, the bottleneck moves to coordination and integration.
The Archaeological Reversal
AI can recover structure from large systems, while the reasons behind that structure remain harder to recover.
The Composition Problem
The theory that lets parts work together lives in their relationships, not in any single specification.
The Encoding Gap
A working answer is not the same as understanding. Knowledge has to be built through the traversal of a problem.
Augmenting Human Intellect in the Age of Agentic AI
Revisiting Engelbart's ideas when our tools act autonomously and preserving human understanding becomes part of the design problem.
Creating Reliable Systems from Unreliable Parts
A long-running engineering lesson: reliability is achieved through the arrangement of imperfect components.
The World Is Full of Isomorphisms
How imitation, professional norms, and external pressures narrow the space of possible ideas.
Archived summaryBounds, Not Goals
Designing agents around viable operating boundaries, feedback, and the friction that makes judgment possible.
Can AI Agents Replace Experts?
An examination of the assumption that enough context can substitute for the understanding an expert develops.
What the Ancients Knew About Agency
Greek and Vedic accounts of purpose, action, and limits offer a way to think about agency in an AI-mediated world.
AI Changed What's Possible
The accumulated experience behind judgment becomes more consequential as the ability to build grows.
SummaryThe Ladder Is Gone
What happens to apprenticeship and agency when AI absorbs the work through which people once learned?
The Protocol Prison
The interfaces that made software usable can become constraints on what intelligent systems are able to do.
The Hollow Middle
As cognitive production becomes cheaper, value moves toward physical constraints, purpose, and accountable judgment.
The Reasoning Gap: A Technical Analysis
A longer treatment of hybrid systems, external verification, and the architecture behind effective AI tools.
Technical versionThe Reasoning Gap
The systems that work combine neural proposals with something that can check them. Reasoning becomes an architectural property.
From For-Loops to Colonies
How agentic coding grows from a persistent loop into coordinated systems, and what that asks of their human designers.
Feedback Is Substrate
Feedback is the material of AI engineering: an ongoing practice of observation, correction, and cultivation.
Human Purpose vs AI Capability
Capable models become useful through verification, direction, and a clear account of what is worth doing.
The Catalyst Effect
AI lowers the cost of getting started. The consequences reach into learning, motivation, and the skills we choose to develop.
Trust as Cognitive Multiplier
Why predictable behaviour and calibrated trust matter as much as capability in human–AI collaboration.
Archived summarySoftware Engineering in the Age of AI
What changes when software engineering becomes the design of a system of people, machines, and feedback?