Before / after

Before

simple · impressive · fragile
DATA
AI
OUTPUT
Production AI system

After

bounded · observable · recoverable
DATA
VALIDATION
RETRIEVAL
AI
EVIDENCE CHECK
ESCALATION / OUTPUT
MonitoringEvaluationsVersioningFreshnessOwnershipFallbacksHuman review

The demo is not the system.

An AI prototype can look impressive while everything is stable. Production begins when:

  • APIs change
  • people edit prompts
  • permissions drift
  • data becomes stale
  • models update
  • users stop double-checking
  • unusual requests arrive

The engineering question is not:

“Does the AI work?”

It's:

“How do we know when it stops working?”

Chanel's framework

01

Frame

What job is the system responsible for?

02

Context

Which information must be present and current?

03

Spec

What behavior counts as correct?

04

Build or decide

What infrastructure deserves to exist?

05

Trust

How will we detect, contain, and recover from failure?

Reliable AI isn't built by eliminating failure. It's built by making failure visible, bounded, and recoverable.

Office hours

This is what “ship like a senior engineer” means.

Office Hours with Chanel — think clearly. build safely. ship like a senior engineer.

If your team has already built AI workflows, agents, or internal tools, the next question isn't always what to build next. Sometimes it's:

  • Who owns this?
  • How do we know it's working?
  • What happens when something changes?
  • Where should humans get pulled back in?
  • What would make this production-ready?

Prototype quality is what happens on the happy path. Engineering quality is what happens on a bad Tuesday.