Incident management · Production

Five sessions and a hackathon: engineering team shifts from low AI adoption to daily agent use after structured enablement series

The problem

An infrastructure engineering team had low and shallow AI adoption despite having available tools and leadership encouragement — most engineers used AI only for autocomplete or not at all, because a poor-quality early experience had led the team to write off AI as a serious tool, and no one had given them a framework for where it fit in their workflow.

First attempt

An earlier wave of AI adoption failed because model output quality did not meet the bar the team held their own work to, causing them to classify AI as a 'fun toy, not a serious tool' and stop experimenting entirely.

Workflow diagram · grounded in source

Low-adoption trigger

trigger

“adoption was low and shallow. A handful of folks were using AI daily. The rest were using it for autocomplete and stopping there or not touching it at all.”

Five weekly 30-min sessions

output

“The series I developed moves deliberately from concept to capability, with each session compounding the last.”

Decompose-Assign-Recombine frame

ai_action

“Decompose → Assign → Recombine. Decompose: Take any workflow you own — writing a Trouble Shooting Guide (TSG), triaging a customer issue incident, rolling up a weekly KPI. Break it into the smallest steps you can. Assign: For each step, …”

Fear conversation

human_review

“Halfway through, I just asked: "What's the thing about AI you don't say out loud?"”

Two-hour hackathon capstone

trigger

“The capstone consisted of a two-hour hackathon. Same room, same people, with one rule: Build an agent that solves a real problem on your own plate. Today.”

Production-grade agents shipped

output

“Two hours later, the room had shipped a fleet of production-grade agents. Here are some genericized examples: An incident triage assistant that pulled context from logs and previous similar incidents and proposed a starting hypothesis. A…”

Behavior change and program expansion

feedback_loop

“What began as a series for one team is now expanding across the broader org. We're planning a multi-team hackathon at a much larger scale — same shape, more builders.”

Reported outcome

Daily AI users grew from two or three engineers to essentially everyone on the team; the two-hour hackathon capstone produced a fleet of production-grade agents, and the program is now expanding to a multi-team hackathon at larger scale.

Reported metrics

daily AI users before seriestwo or three engineers

daily AI users after seriesessentially everyone

KPI rollup analyst time10 minutes instead of two hours

Reported stack

MCP Servers

Source

https://medium.com/data-science-at-microsoft/five-sessions-and-a-hackathon-how-we-turned-skeptics-into-agent-builders-4320f3eb1af1

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Daily AI users grew from two or three engineers to essentially everyone on the team; the two-hour hackathon capstone produced a fleet of production-grade agents, and the program is now expanding to a multi-team hackat…

What tools did this team use?

MCP Servers.

What results were reported?

daily AI users before series: two or three engineers; daily AI users after series: essentially everyone; KPI rollup analyst time: 10 minutes instead of two hours (source-reported, not independently verified).

What failed first in this deployment?

How is this incident management AI workflow structured?

Low-adoption trigger → Five weekly 30-min sessions → Decompose-Assign-Recombine frame → Fear conversation → Two-hour hackathon capstone → Production-grade agents shipped → Behavior change and program expansion.