It support · Production

How Qovery built an Agentic DevOps Copilot through 4 technical phases to automate infrastructure tasks

The problem

Qovery wanted to eliminate the grunt work of DevOps by building an assistant that could understand developer intent and autonomously take action on infrastructure, but early hardcoded approaches could not handle real-world unplanned requests.

First attempt

The first-phase basic agent required every new intent to be hardcoded with no flexibility for unplanned requests; the second-phase agentic system improved flexibility but tool chaining was fragile, meaning a single tool failure broke the entire plan.

Workflow diagram · grounded in source
1
Developer submits request
trigger
“"Stop all dev environments after 6pm" → matches the stop-env tool”
2
AI plans tool sequence
ai_action
“the DevOps AI Agent Copilot receives user input, analyzes it, and dynamically plans a sequence of tool invocations to fulfill the request”
3
Validation between tool steps
validation
“tracking intermediate state, running validation between tool steps, and allowing re-planning if an execution fails”
4
Failure recovery and replanning
feedback_loop
“if the agent misuses a tool or the tool returns an unexpected output, the system: - Analyzes the failure - Updates its plan or fixes the step - Retries with a corrected approach”
5
Conversation memory lookup
ai_action
“It allows the Agentic DevOps Copilot to: - Reuse previous answers - Understand references and context - Maintain continuity across a session”
6
Infrastructure action delivered
output
“It helps developers automate deployments, optimize infrastructure, and answer advanced configuration questions”
Reported outcome

The Agentic DevOps Copilot is live in Alpha, helping developers automate deployments, optimize infrastructure, and answer advanced configuration questions, with drastically improved user experience after the addition of conversation memory.

Reported metrics
User experience after memory addeddrastically improved user experience
Multi-step workflow completionssuccessful completions of multi-step workflows that weren't even anticipated during development
Complex task planning latencyup to 10 seconds
Reported stack
Claude Sonnet 3.7QDrant
Source
https://www.qovery.com/blog/how-we-built-an-agentic-devops-copilot-to-automate-infrastructure-tasks-and-beyond/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Agentic DevOps Copilot is live in Alpha, helping developers automate deployments, optimize infrastructure, and answer advanced configuration questions, with drastically improved user experience after the addition…

What tools did this team use?

Claude Sonnet 3.7, QDrant.

What results were reported?

User experience after memory added: drastically improved user experience; Multi-step workflow completions: successful completions of multi-step workflows that weren't even anticipated during development; Complex task planning latency: up to 10 seconds (source-reported, not independently verified).

What failed first in this deployment?

The first-phase basic agent required every new intent to be hardcoded with no flexibility for unplanned requests; the second-phase agentic system improved flexibility but tool chaining was fragile, meaning a single to…

How is this it support AI workflow structured?

Developer submits request → AI plans tool sequence → Validation between tool steps → Failure recovery and replanning → Conversation memory lookup → Infrastructure action delivered.

Related it support cases