The Aviation Model
Aviation got safer by training pilots to manage automation, not by trusting autopilot more. Let routine fluency fade; keep enough to land the plane — which is what Manual Flight Days are for.
A-01 · Ayman Diab · Tech Lead — Montréal
Humans as architects of intent.
I design how engineering organizations work with AI — the methodology, the security posture, and the CI enforcement that let teams get real leverage from machines without surrendering the judgment that made them good.
A-02 · Selected work
Four systems, one thesis.
Methodology, enforcement, security, and enablement — each designed so an organization can adopt AI at scale and still be able to defend every line it ships.
W-01
The Engineering Workflow Playbook
A company-wide operating philosophy for how humans and AI build together — and the six-stage engineering pilot that tests it, authored and run against real delivery on SquadBox.
- methodology
- AI enablement
- org leadership
W-02
AI-Robustness Gates
Making a codebase’s architectural invariants mechanically enforceable in CI — because human reviewer vigilance does not scale with AI-authored code.
- quality engineering
- CI/CD
- AI safety engineering
W-03
Threat-Modeling AI Workflows
A multi-framework threat model for AI-augmented engineering workflows — so an agent-adoption decision could be made on mapped risk instead of vendor marketing.
- security
- threat modeling
- AI adoption
W-04
Safe Mode: Unblocking an Org-Wide Agentic Rollout
A bilingual, arcade-style security game that replaced the policy page nobody reads — built to satisfy the guardrail requirement gating an organization-wide rollout of agentic AI tooling.
- security enablement
- change management
- product build
A-03 · In production
And one that's running in a mine.
The same thesis, with a 400-tonne haul truck on the other end of it. An agent notices something the thresholds miss, a reliability engineer decides what to do about it, and the machine never gets a vote.
A-04 · Operating principles
Field notes from the methodology.
Proof of Ownership
You ship it, you defend it in the room — and "the AI just did that" means the change is not ready to merge.
The Capability–Comprehension Gap
The distance between what an engineer can produce with AI and what that engineer alone understands. It widens invisibly, because assisted output keeps looking fine — right up until an anomaly demands the understanding that eroded.
The Cognitive Integrity Threshold
The floor an owner has to stay above: can you verify the output, reconstruct the reasoning behind it when it breaks, and recognize where the work should have stopped? Below that floor, oversight is a signature rather than a check.
Pilots run a checklist before every flight — not because they have forgotten how to fly.
Run the Pre-Flight Checklist →Twelve questions, about four minutes, no email.
A-05 · Contact
Building a human-AI engineering practice? Let's compare drawings.
I write and speak about methodology, AI security, and quality engineering for AI-authored code. The fastest way to reach me is email.
enquiries@aymandiab.com