AI is often framed as a productivity tool – something that helps you write faster, brainstorm ideas, or automate small tasks. But what happens when AI is responsible for building an actual product?
In a recent Medium post, Olivier Rivard, VP of Product at Diskover Data, walks through exactly that: how he built a production-ready Model Context Protocol (MCP) connector for Diskover using AI agents from start to finish.
→ Not a demo.
→ Not a proof of concept.
→ A real product that will ship to customers.
This summary highlights the key ideas and why they matter for how modern products get built.
From AI assistant to AI product team.
Olivier’s goal wasn’t to “use AI for coding.” It was to see whether AI could operate as a cross-functional product team spanning product management, engineering execution, and quality review.
Using MCP (an open standard that allows LLMs to securely interact with real systems), AI agents were connected directly to tools Diskover already uses, including Jira and GitHub. This meant the AI wasn’t working in isolation – it could read live project data, create issues, follow workflows, and implement real features.
The result: a fully AI-assisted workflow that went from product requirements to deployed code.
STEP 1: Defining product requirements with AI.
Instead of starting with a blank document, Olivier defined how the AI should behave – effectively turning it into a Senior Product Management Assistant.
With access to Jira via MCP, the AI could:
- Understand Diskover’s existing templates and standards
- Create properly scoped Epics and stories
- Generate acceptance criteria
- Populate the development board automatically
This wasn’t documentation for documentation’s sake. It produced a clean, executable roadmap that engineering could immediately act on.
STEP 2: Scoping the first version.
Rather than trying to do everything, the first version of the MCP connector focused on a small set of high-value Diskover capabilities, including:
- Searching files using Diskover’s metadata filters
- Tagging files and folders for organization and automation
- Retrieving key metrics and usage statistics
- Listing recently indexed datasets
This narrow scope delivered immediate value while keeping complexity low – a classic product principle, now executed with AI.
STEP 3: Creating the full Jira breakdown with Claude.
For each feature, the AI:
- Asked clarifying questions
- Drafted a complete Jira story with acceptance criteria
- Created the issue under the correct Epic
By the end of this step, the early roadmap for the MCP connector lived entirely inside Jira – generated and structured by AI, but aligned with Diskover’s real development process.
STEP 4: Implementing every feature with AI.
AI didn’t stop at planning. It implemented the features too.
Claude Code handled development tasks by pulling Jira issues directly and writing code. ChatGPT Codex was used as a reviewer, flagging bugs, edge cases, and optimization opportunities. Olivier then validated, tested, and guided fixes – acting as the final quality gate.
The human role shifted from “doing the work” to directing and validating the work.
A new model for building software.
One of the clearest takeaways from this experiment is that AI-driven development doesn’t eliminate the need for product or technical leadership – it amplifies it.
Success depended on:
- Clear, unambiguous product requirements
- Strong intuition about user workflows
- The ability to review and validate code written by others (human or AI)
In other words, experience still matters. The tools changed, the responsibility didn’t.
Read the full story.
This summary only scratches the surface. Olivier’s full Medium post goes deeper into:
- The exact AI workflows used
- Lessons learned along the way
- Why this approach is reshaping how Diskover thinks about product development
Read the full post on Medium: How I’m Building the Diskover MCP Connector Entirely With AI
If you’re curious about what product development looks like in an AI-first world, it’s worth your time.