Customer story · Telecommunications

A network diagnostic agent that saves 15 engineers 4 hours a month each

In four weeks we built a pilot agent that diagnoses network errors and drafts the fix, while engineers keep every change in their hands.
Let’s talk
Sector
Telecommunications
Engineers
15
Duration
4 weeks, 2026
Time saved
4 h per engineer per month
Status
Pilot

The situation

  • Cisco and Juniper equipment, run by a team of 15 engineers
  • Before fixing an error, an engineer rebuilds the context: devices on the path, IP plan, flow matrix, configurations
  • All of it exists, spread across diagrams, spreadsheets and an in-house configuration repository
  • Pulling it together by hand is the slow part of every diagnosis

The architecture

The agent reads the documentation and the configurations, and writes the Jira story. The engineer approves and applies the fix by hand: the agent never touches the equipment.

What we built

1. Grounded in the network’s own documentation

  • Network diagrams
  • IP plans
  • The flow matrix

2. Real configurations, read-only

  • A custom MCP server over the in-house configuration repository
  • It can read; it cannot change anything

3. A diagnosis where engineers already work

  • Engineers ask inside Windsurf
  • Root cause and a proposed fix
  • No orchestration framework: the Windsurf agent plus our MCP servers, which is what made a four-week pilot possible

4. The Jira story, written for you

Through Atlassian’s official MCP server, with:

  • Summary
  • Root cause
  • Affected devices
  • Proposed fix
  • Priority

5. Two approvals, then the engineer acts

  • The engineer approves the story
  • Then approves the change in Jira
  • Then applies the fix by hand: the agent never touches the equipment

6. Left for version 2

The pilot proved the value first. Access today is a username and password with MFA. Next:

  • Logging of every agent action
  • A direct connection to the equipment
  • Packaging inside Microsoft 365 Copilot

What changed

60 h
saved per month across the team
CA$9,000
a month, at CA$150 an hour
CA$108,000
a year, from a four-week build
  • Each of the 15 engineers saved about 4 hours a month during the pilot
  • Four weeks to build
  • Every change still approved and applied by an engineer
What we learned

It saves time. The next gain is reading the equipment directly, not an in-house copy that can drift.

If you are in the same situation

If your engineers spend hours gathering context before they can fix anything, this is what we bring:

  • An agent grounded in your own documentation: diagrams, IP plans, flow matrices, runbooks
  • Read-only access to your configurations through an MCP server, from your source of truth or directly from the equipment
  • Diagnosis and proposed fix in the tool your engineers already use: Windsurf, Microsoft 365 Copilot or Teams
  • Tickets written by the agent in Jira or your tracker, through the vendor’s MCP server where one exists
  • Human approval before any change, with engineers keeping their hands on the equipment
  • A four-week pilot with the time saved measured, before any larger commitment

A four-week pilot. Measure the hours it saves, then decide what version 2 should harden.

Your engineers spend hours gathering context?

Tell us how your team diagnoses problems today. We will tell you what an agent could take off its plate, and what it would cost to run.

Let’s talkAI Agents

The stack

Agent

  • Windsurf
  • Devin

Integrations

  • Model Context Protocol (MCP)
  • Custom configuration MCP server
  • Atlassian MCP server
  • Jira

Knowledge

  • RAG
  • Network diagrams
  • IP plans
  • Flow matrix

Network

  • Cisco
  • Juniper

Version 2

  • Microsoft 365 Copilot