Architecture · Production

Why you need an architect — not just an AI that writes code

AI closes the first 80% well. Production is decided by the last 20%. You are paying for the judgment to catch the architecturally plausible but operationally lethal decision before it ships.

The honest 80/20

AI does this well

  • Prototype, demo, happy path
  • Boilerplate, CRUD, controllers
  • Translating a clear spec into code
  • Test scaffolding, glue code

Where it breaks

  • Failure modes, concurrency, locks
  • Data integrity at scale
  • Behavior under load — query plans, cost
  • Security beyond OWASP top-10
  • Decisions visible only at 100,000 users

What you are actually buying

AI confidently produces code that looks architecturally sound — and turns lethal six months in. Judgment is required to evaluate output against a model of the production system. AI amplifies skill: expert + AI reaches a maintainable system fast; amateur + AI reaches a demo that disintegrates.

Who is accountable in production

When the system drops at 02:14, someone has to have a name and a signature on the architecture. AI cannot sign anything. The vendor of the model cannot either. Hire someone who has been making those calls in production for twenty years.

Evidence base

20+ years
Enterprise .NET
National medical platform
18 years, 100K+ users, 99.99% uptime
M.Sc. TAU
Applied Math & CS · MCPD
14 countries
IL, US, UK, EU, RU

Free download

The AI Pre-Production Review Checklist

22 failure modes AI generators ship into production code, in 7 categories — with detection signals and the architect's mitigations.

Email me the checklist →

Before you ship, get the design reviewed

30 minutes, no slides. Focused architecture review — not a sales pitch.

Book the review