Software Got Cheap. Contracts Didn't.
When writing code costs almost nothing, value moves from applications to the handshakes between them.
Thread
Systems
Software as a modelling problem, and the systems people work inside: why enterprise software is the way it is, agents and the tools they are given, and what cheap code does to authorship, judgment and the price of trusted work.
- 01
- 02
Fifteen Guests and a Wallflower
Design the occasion a room has to survive, and let the people in it score the plan.
- 03
Work You'd Do for Fun
Why business software was dull, how AI changes the economics of game feel, and what work looks like when software answers back.
- 04
The Organisation Decides With Less Than It Knows
Institutional memory leaves when experts do. PROXY turns tacit judgement into queryable agents, advisory councils, and an immutable ledger.
- 05
Low Background Steel
The internet is no longer majority human. A fleet of sunken battleships suggests what happens next.
- 06
Small to Big
Why BIM cannot compose: moving the built environment from locked CAD files to executable spatial contracts.
- 07
Spec to Site
The internet was written by hand until it wasn't. The built world is next, and the scarce thing at the far end of that shift is not drawing.
- 08
The Advertising Machine Is About to Learn How You Choose
Multimodal AI moves advertising from demographic probability to probabilistic models of individual judgment and context-dependent choice.
- 09
When Code Is Cheap, Judgment Compounds
The enterprise asset is the accumulated ability to specify, evaluate, and reproduce what works.
- 10
Where to Put the Meter
When raw intelligence becomes cheap, value migrates from software seats and labor hours to the delegation of trusted work.
- 11
After Spec-to-Prod: The Legibility Problem in GovTech
Beyond adaptable software lies the harder challenge: building a continuously maintained state of reality that governments can actually reason over.
- 12
The Knowledge Shock: Why India Needs Cheap Intelligence
As software costs collapse, India’s next macroeconomic advantage depends on turning metered inference into productive capital.
- 13
From Writing to Authorship: What AI Changes About Thinking
When machines can write almost anything, the scarce asset is knowing what you actually mean.
- 14
The Distillation Economy
How open weights, quantization, and local inference are rewriting the geopolitical supply chain of artificial intelligence.
- 15
The Decision as Software
When the unit of government software stops being the dashboard and starts being the decision.
- 16
Models Are Not Applications
What looks like enterprise AI pessimism is a strategic narrative war over market architecture and value capture.
- 17
The Language Games of Law
LLMs conquered the textual surface of legal practice. The institutional context remains an unsolved system-design problem.
- 18
From Apps to Ontologies
AI makes government software easy to build but hard to buy—shifting the real moat from generated dashboards to ground-truth civic semantics.
- 19
The Abstract Machine
Why we stopped writing solvers and started writing the thing that writes them
- 20
The Company Is a Flowchart
Why the hardest problem in enterprise software is a modelling error, and why the fix has to be free
- 21
The Volume Asymmetry of Physical AI
Why physical intelligence requires capex substitution, edge perception, and owning the unrecorded state of space over time.
- 22
AI Decentralization Ends at Your Dinner Table
The intelligence race pushes AI onto owned hardware, then onto your body. A three-to-four-year chain, told as a story.