Khalid ShamsPrincipal Solutions Architect

Writing

Notes from the work.

Short essays on architecture, agentic AI, and the decisions that actually cost money. Published here first, then on LinkedIn. RSS.

October 15, 2026

The cheapest migration is the one you skip

In one assessment, a third of the application estate had no owner. You don't migrate that. You retire it.

October 13, 2026

A landing zone is a political document

The subscriptions and policies are the easy part. A landing zone is really an agreement about who owns what.

October 8, 2026

Where I put the human

I architected an EMR holding 250+ patient records, and it changed where I put the human.

October 6, 2026

The agent inherits the user's permissions

The most important design decision in an enterprise AI system isn't the model.

October 1, 2026

Change the altitude, not the substance

An ad hoc note I was inspired to write. A conversation with people who think about building, and the one habit that lets you talk to a CTO and an engineer in the same hour without lying to either.

October 1, 2026

The dashboard that got applause

I've built dashboards that got applause and then quietly became the most expensive thing in the system.

September 29, 2026

Root cause before remediation

A PostgreSQL escalation, a room full of confident theories, and why we refused to resize first.

September 24, 2026

Ask to see the evaluation harness

The one question that separates real AI projects from expensive demos.

September 22, 2026

$4.5M without looking at a price sheet

$4.5M in cost avoidance without touching a price sheet. The savings weren't in negotiating; they were in noticing.

September 17, 2026

1,000 tenants and a problem you can't reproduce

A performance problem nobody could reproduce across 1,000+ tenant databases, and why the fix was written rules, not a bigger tier.

September 15, 2026

The demo is never the hard part

Most agent demos work. Most agent systems don't. The four things nobody demos are the whole difference.