08 / The human

Building softwarewas never reallyabout the code.

Technology has changed dramatically across Jason's career, from early B2B web applications to commerce platforms to AI agents.

What has stayed constant is the process: seeing something that doesn't exist yet, understanding how it should work, and building the system that makes it real. Then staying with it until it is reliable.

He has led teams of up to 20, run technology operations for a $100M business, and still writes the code himself.

How I build

  1. Understand the system. Map the real inputs, outputs, dependencies and constraints before changing anything.
  2. Find the failure points. Look for the boundaries where state, integrations, assumptions or people can break the flow.
  3. Build the smallest working version. Prove the core behaviour before expanding the surface area.
  4. Run it against reality. Real usage exposes what diagrams and optimistic assumptions miss.
  5. Observe what breaks. Logs, state and failure evidence turn surprises into engineering input.
  6. Automate what repeats. Move reliable repeated work out of human memory and into explicit systems.
  7. Keep improving. Ship, measure, learn and revise without pretending the first version is the last.

What stays constant

  • Humans approve the risky parts. Automation runs the safe work; anything consequential waits for a person.
  • Safe by default. Features ship switched off or locked until someone deliberately turns them on.
  • Freshness is shown, not assumed. Every number says where it came from and how old it is; stale data is labeled or refused.
  • Unknown is not zero. Missing evidence stays missing instead of quietly becoming a number.
  • Append-only records. Decisions, fixes and imports go into ledgers that can be audited and undone.
  • Releases match the commit. What is live is checked against exactly what was tested.
  • Cost has a ceiling. Quotas, budgets and scale-to-zero so a good day never becomes a bad bill.
  • AI stays bounded and grounded. Models advise inside deterministic checks and answer from real data, not imagination.