Engineering Leader | Entrepreneur | Educator. Building reliable systems at scale. Pragmatic insights on AI, blockchain & tech strategy.
Toronto, Ontario, Canada
Joined February 2009
- Tweets24.3K
- Following324
- Followers711
- Likes3K
Approval belongs to the revision and evidence inspected. If either changes, check the decision again. Otherwise the gate may approve a different change from the one the reviewer saw.
michaeldabydeen.com/articles…
Passing examples support the cases they cover. They do not prove that a change handles every condition the contract allows. I published a free, runnable review exercise for that distinction.
michaeldabydeen.com/resource…
A fast model can classify a proposed action. Application code should decide whether it is allowed. I wrote about that boundary using a browser fixture and Jev's typed-decision model.
coderlegion.com/28754/system…
An agent-authored PR should carry evidence: proposed and tested revisions, fixture versions, assumptions, and a recovery owner. A YAML policy is only a proposal until checks enforce it.
dev.to/_firelinks/a-review-c…
An API can be documented and still be a poor interface for an agent. I look for explicit side effects, idempotency, bounded retrieval, typed failures, and a safe way to test or reverse a mutation.
michaeldabydeen.com/articles…
Most AI strategy decks are 100% hype and 0% execution. As an IT leader, if you miss these 3 pillars, you’re building a liability. 🧵
5/ The board wants ROI, not magic. Focus on operational rigor over flashy demos. That is how you actually scale in the enterprise.
After attending an AI innovation event today, I've realized that production readiness for agents is about ownership, evaluation, review, traceability, and rollback