Enterprise project · Microsoft

When the cloud playbook didn’t fit, we built the one that did.

Microsoft’s cloud adoption and well-architected frameworks were made for typical cloud workloads. IoT workloads run across the cloud, the edge and devices, and the frameworks didn’t fit them. A senior SNT engineer, working through a vendor as a technical program manager in Microsoft’s global IoT adoption team, led the build of a Well-Architected program for these distributed workloads, with ten cloud solution architects and two field engineers. Microsoft piloted it with a Fortune 500 pharmaceutical company, then launched it.

What the program covered

Watch it live

With Microsoft’s product teams, its monitoring tools were tailored for workloads spread across cloud, edge and devices. Customers kept them running after the workshop, so problems showed up as they happened and were fixed right away, not found weeks later.

Learning and an assessment

The five pillars, reliability, security, performance efficiency, operational excellence and cost optimization, taught for workloads spread across cloud, edge and devices, with an assessment that scores a workload against each one.

A three-day workshop

Run onsite by a cloud solution architect: baseline the customer’s workload with the monitoring tools, find the gaps that hold back reliability, security and cost, and leave with a plan to close them.

The numbers

All figures are actuals.

5
pillars every workload was assessed against
10 + 2
cloud solution architects and field engineers on the build team
3
places a workload runs, cloud, edge and device, assessed and watched as one
3 days
onsite workshop: baseline, gaps, plan
5
workshops run, in the US, Europe and South Korea
1
Fortune 500 pharmaceutical company in the pilot

How we use it with AI today

  1. 01

    Review before you build

    Every engagement starts with a well-architected review for AI: the use case and the systems it touches, scored on the same five pillars, before any agent goes live.

  2. 02

    Reliability means tests

    Agents get tests that check their answers, guardrails on what they may do, and a person who approves what matters.

  3. 03

    Security means data and inputs

    An agent sees only what the person asking may see, and what it reads is checked for instructions hidden in the data.

  4. 04

    Cost is designed in

    Model choice, caching and limits are planned up front and measured every week, because an agent’s cost moves with use.

  5. 05

    Watch, catch, fix

    Every agent call is traced from day one, with alerts when its answers, cost or speed drift, and a person fixes it the same day. The same live watch the program gave distributed workloads.

Why it matters for AI. AI agents break the usual playbook the way IoT did: their answers vary, they reach into many systems, and their cost moves with use. Starting with a well-architected review means a pilot is built to reach production, not to stall after the demo.

See it, then talk to us

This was our engineer’s work for Microsoft through a vendor, not an SNT Agents build. Ask us how a well-architected review for AI would work for your team.