0

Your cart

TOTAL excl.
TOTAL incl.
Pay
Go back

Developer AI at scale: the real work begins after the prompt

REXConseil / CoachingDevSecOpsSécurité : Données / Accès / EmailDéveloppementCulture et Méthodologie

2026-12-10 | 04:15 PM - 04:45 PM

Information

Alone within its IDE, the AI assistant is unfamiliar with the company's practices and information system: it remains generic. We have provided it with our practices, securely opened the information system without creating tokens for each developer, and made its usage observable, all of which is available to more than 200 teams.

In our IDEs, AI assistants have become everyday teammates.
Our role, as a DevX team, is to enhance the tools developers already use, connecting them to the company's knowledge and services . The challenge is no longer writing the right prompt, but providing the necessary tools. secure and monitor these uses across a network, without managing new credentials: we reuse the developers' connection to the company.

It all starts with knowledge: transmitting our practices to LLMs via skills , an open format combining expertise and reproducible workflows.
An initial Context7-based implementation using MCP kept us agnostic but didn't deliver a complete skill with references, templates, and executable scripts. We lacked a standard for delivering a complete skill. This standard emerged from the AI ecosystem itself: plugins and marketplaces. We adopted them to distribute these components locally, version-controlled in Git, and shared between teams.
Thanks to this model, we have implemented an open contribution system : each team creates and shares its own plugins.
Our _Core_ plugin already distributes around fifty AI assets (agents, skills and MCP servers), not counting what is shared by the internal community!

We support this adoption through acculturation work with developers: plenary sessions, practical exercises and regular communication (emails, discussion forums)...

But a skill that includes a script is only useful if it can interact with the information system with controlled access. We developed bpi-cli , a cross-platform command-line tool (Go binary) that installs the complete developer environment and enables authentication with the company's IdP via OAuth 2.0/OIDC (PKCE flow), with tokens remaining in the secure keychain on the workstation. It then becomes the trusted gateway through which AI assets call internal services.

We secure calls to APIs and MCP servers without any dedicated tokens or secrets.

Hooks finally make the actual uses of skills , agents and tools observable, to drive adoption and quality.

This feedback shares concrete patterns for connecting developers' AI tools to the IS and scaling it up, with a clear objective: more speed and quality on a daily basis.

Speakers