Production engineering.
Now agentic.

Treat agents like senior engineers.

Same governance you'd apply to any senior IC: audit trail by default, human review on the fault line, invariants that codify what you've learned the hard way. The agentic delivery method is this principle, worked out into seven concrete practices and an invariant library that travels with us between codebases.

Each invariant is the answer to a specific class of production failure. Each one is a rule of the form “don't do X, here's the incident that taught us why,” with code-level enforcement when possible. Like this:

# Schema changes require migration files alongside
# the consuming code, in the same commit.

# Failure mode
#   Drizzle's .returning() and .select() enumerate the
#   schema's columns regardless of whether the mutation
#   sets them. A nullable column added in code but not
#   yet pushed to production DB causes 500s on every
#   query that touches the table.
#
# Real-world cost
#   Three production outages in three days at PickNDeal
#   (2026-04-18 → 2026-04-20) before we hardened the
#   deploy script.
#
# Enforcement
#   - schema-touch detection gates db:push in CI
#   - no || true on the push step (set -e wraps it)
#   - .last-deployed-sha only updates on push success
#   - one-line tRPC verification on the affected
#     endpoint after deploy completes

The full method (seven principles, 20+ invariants) is documented at the method page. The invariant library lives across our production codebases; a public template will appear at /open-source once we've extracted the first artefact. Production-specific extensions stay inside client engagements.