← Back to the atlas
AWS & ML

What working in AI/ML at AWS changed about my engineering instincts

In draft Draft date

Model behavior is one part of the system — deployment, observability, interfaces, and failure handling determine whether the capability is useful

Working outline — full essay forthcoming

Rough outline

Start with the engineering problem

  • Introduce the kind of AI/ML integration being discussed without internal project details
  • Distinguish model quality from the reliability of the surrounding service

Make boundaries explicit

  • Trace the path from an input through preprocessing, inference, and downstream action
  • Identify ownership of validation, retries, latency budgets, and error reporting

Explain the change in judgment

  • Describe one engineering assumption that changed through work at AWS
  • Compare the original approach with the approach Anthony would take now

Apply it to autonomous systems

  • Connect service-level observability to agent traces and tool boundaries
  • Close with a concrete design decision that readers can apply using public examples