Case study · A Salesforce consulting partner
An AI delivery platform a partner sells to its own clients
A consulting partner wanted its own AI delivery product to offer clients, not a seat on someone else’s tool. We designed and built it: a self-hosted platform where AI agents plan, build and test Salesforce stories, people approve the plan and the pull request, and GitHub Actions deploys through the client’s org pipeline.
- Salesforce consulting
- Built as IP for a partner with a team of five, live with its first clients since late October 2026
The problem
The partner’s clients wanted AI to speed up Salesforce delivery, but not by pasting org metadata and code into a public chat tool, and not by handing an agent the keys to production.
Off the shelf coding assistants help one developer at a time. They don’t know the team’s backlog, the org pipeline, or who has to approve a change before it moves to the next environment.
The partner wanted something it could own and sell under its own name, installed in each client’s own environment, with the client’s data and credentials staying there.
What we delivered
- A story board where every story moves through AI planning, plan review, AI development, AI QA, code review and deployment, in that order
- Planner, developer, QA and DevOps agents that read the connected Salesforce org, write the plan and the code, and diagnose failed deploys
- People stay in charge: a person approves every plan and every pull request, and environment approvals above the first sandbox always apply
- Deterministic deploys through GitHub Actions, one pull request per environment hop, so agents never deploy anything themselves
- One self-hosted instance per client, running on Docker in the client’s own environment, with credentials encrypted at rest and outbound traffic limited to an allowlist
- Bring your own model: each person can use their own Claude or OpenAI subscription, or the workspace API key, with model routing, budgets and a cost view per story
- Its own MCP server, so the team can read stories and plans from Claude Desktop or any MCP client
Have a project like this?
Thirty minutes with the architect who would do the work. You leave with a recommended next step.
Book a free 30-min scoping call