Skip to main content

Using AI to Review Architecture Proposals

5 min read
Software ArchSolutions ArchEnterprise Arch

Software Arch

AI catches gaps. You catch 'will my team actually run this?'

Enterprise Arch

Use AI for consistency checks across proposals. You own governance.

Using AI to Review Architecture Proposals

TL;DR

  • AI can spot missing failure modes, inconsistent naming, and common anti-patterns.
  • AI can't judge org fit, political landmines, or "will the team actually maintain this?"
  • Run proposals through AI first. Use it as a checklist. Then apply human judgment.

You've got an RFC, an architecture proposal, or a design doc. Someone (maybe you) spent weeks on it. AI can review it in seconds. The trick: knowing what to trust and what to verify yourself.

What AI Review Actually Catches

Reliably useful:

  • Missing failure modes ("what happens when the database goes down?")
  • Inconsistent terminology and naming across sections
  • Common anti-patterns (tight coupling, no retry logic, missing observability)
  • Obvious gaps in security, compliance, or data flow
  • Structural issues (conflicting assumptions, circular dependencies)

Not reliably:

  • Whether the team has the skills to build and operate it
  • Whether this fits your org's appetite for complexity
  • Whether the proposed timeline is realistic
  • Whether stakeholders will actually sign off
  • Whether you're over-engineering for your scale

Think of AI as a very thorough first-pass reviewer who's read every postmortem and RFC ever written but has never met your team.

How to Run an AI-Assisted Review

  1. Paste the full doc — Don't summarize. AI needs context. Include diagrams in text form if possible.
  2. Ask for specific checks — "What failure modes are missing?" "Where might this break at 10x scale?" "What compliance concerns should we address?"
  3. Don't accept blindly — AI will sometimes flag non-issues or miss the real one. Cross-reference.
  4. Use AI for the second draft — After human review, run the revised doc through again. Catch what you fixed and what you missed.

The Human-Only Parts

You still need to answer:

  • Does this match our tech strategy? (We're standardizing on X, this introduces Y.)
  • Can we staff this? (We need 2 Kafka experts. We have 0.)
  • Does this create a single point of failure we can't accept?
  • Will the business tolerate this timeline/risk?

AI doesn't know your org chart, your budget, or your political realities.

AI Disruption Risk for Software Architects

Moderate Risk

SafeCritical

AI automates routine work. Strategy, judgment, and human touch remain essential. Moderate risk for those who own the outcomes.

Manual process. Repetitive tasks. Limited scale.

Click "With AI" to see the difference →

Quick Check

What remains human when AI automates more of this role?

Do This Next

  1. Take one existing RFC or design doc — Run it through an AI assistant. Ask: "What failure modes, security gaps, or scalability concerns are missing?"
  2. Compare AI's list to your gut — Which items would you have caught? Which are new? Which do you disagree with and why?
  3. Add "AI review" to your proposal template — Require that every architecture proposal gets an AI-assisted pass before human review. Document what AI found and what humans overrode.