Working Thesis

The Factory
Is Going Dark

Software is heading toward the dark factory: agents doing the implementation, humans holding the two endpoints.

My job: write specifications an agent cannot misread, and demand receipts it cannot fake.

01

The Spec Is the Artifact

When agents can produce working code from any sufficiently precise description, the bottleneck moves upstream to the description. The specification becomes the primary engineering artifact; the codebase is a derivative — closer to a build output than to source.

02

Fix the Loop, Not the Model

Agent failures are ambiguity failures of the loop, not intelligence failures of the agent. Smartness cannot supply a fact that was never specified. A real run has a goal, a boundary, tools, artifacts, and receipts — miss one and you made a wish, not a delegation.

03

"Done" Requires a Receipt

An agent declaring success is self-attestation by the party with the strongest incentive to call the job done. No diff, no test run, no artifact — no "done." The same review bar applies whether the author was a human or a machine.

04

Deterministic Shells, Non-Deterministic Cores

Non-deterministic output demands a deterministic, accessible interface as its stability layer. The UI is a contract, not a display — that is what makes streaming AI trustworthy enough for 50M+ monthly users on a financial news platform.

05

AI Is a Motorcycle, Not an Equalizer

AI equalizes execution speed — but execution was already cheap. What it amplifies is specification quality, which is a direct function of domain depth. It makes experts more productive faster than it makes novices competent. Keep earning the depth.

Remember This

  • Casual AI use is already table stakes — only the delegation tiers differentiate
  • Reliability is engineered into the loop, not summoned from the model
  • AI makes experts more productive faster than it makes novices competent
  • Memory and context architecture beat model selection
  • English is now a programming language — write it with an engineer’s precision

But Also This

  • Some engineering decisions only emerge from contact with the code
  • A brownfield system is its own specification — respect what running code encodes
  • AI-expanded scope carries errors you may not be qualified to catch
  • Single-trial outcomes cannot grade a decision process
  • Your worth is not your throughput

"I use AI" stopped being the differentiator. Whether you can hand an agent a boundary and a review bar — instead of still typing every line yourself — is the one that's left.

— from my claim vault, on where the leverage moved
0
Day Streak
Days in a row you've completed all 5 non-negotiables

These are my operating principles, argued from nine years of production systems and tested daily against real work.
Influences worth reading in full: Matt Shumer's "Something Big Is Happening" on urgency, and Kleppmann's Designing Data-Intensive Applications on why the loop patterns are older than the agents.