Workday Developer Agent, generally available since October 13, 2026, works inside the coding tools developers already use: Workday names Claude Code, VS Code, Cursor and Google Antigravity, connected through MCP. For an in-house developer with one Workday tenant family, that is simple.

For a consultant serving three or four Workday customers, it raises a question that has nothing to do with code quality: when the same editor and AI assistant are used for every client, what keeps Client A's tenant, credentials and context out of the work for Client B?

Where mixing happens

In a typical consultant's setup, the risk points are mundane:

  • One tool connection, several tenants. An MCP connection configured once in an editor points at whichever tenant was set up last. Switching clients means remembering to switch it.
  • AI memory and instruction files. Coding assistants keep project notes, memory files and conversation history. A note written while working on one client's data model can be read in the next session for a different client.
  • Copied snippets. Code, field names and orchestration logic move between projects by copy and paste, sometimes carrying a client's custom object names or business rules with them.
  • Personal vs. company accounts. Workday now offers personal developer accounts with their own Development tenant (in beta). Useful for learning, but client work belongs in the client's company account. See personal vs. company developer accounts.

None of these is a Workday security failure. Workday applies its permissions to each call, and Developer Agent keeps developer data out of model training. The leak, if it happens, is in the consultant's own workspace.

A setup that keeps clients apart

  1. One workspace per client. Separate folders, and ideally separate editor profiles, for each customer. Never open two clients' projects in the same window.
  2. One Workday connection per workspace. Configure the Developer Agent connection, and any other Workday tools, at the workspace level, pointing only at that client's tenants. Avoid global configuration that follows you between projects.
  3. Separate AI memory. Keep assistant instruction and memory files inside each client's workspace, and check that your tool does not also load a global memory file containing client details.
  4. Invented names in reusable material. When you turn a solution into a reusable pattern, rename every client-specific object, field and organisation before it leaves the client's workspace.
  5. Check the tenant before every change. Ask the assistant which tenant it is about to change, and confirm it, before any deployment.
  6. Read the client's AI policy. Many customers restrict which AI providers may see their configuration. Developer Agent's inference runs within the customer's Extend region, but the coding assistant around it has its own data terms.

The longer version of this discipline is in the multi-client consultant's guide to zero data leakage.

How kiweely is designed for this

kiweely was built for consultants who work across several Workday customers. Its structure makes the separation the default rather than a habit you have to keep:

  • Connections. Each client's tenants live in their own connection, with their own sign-ins. A conversation works with the connections you give it, so it does not quietly reach another client's tenant.
  • Projects. Each Workday Extend app or Workday project is its own project, holding its own files.
  • Conversations. Every conversation shows which client's tenants it is working with, and the agent asks for your consent before it changes anything in a tenant.
  • On your computer. kiweely runs as a desktop app with your own ChatGPT or Claude account. Client sign-ins are kept in this computer's secure credential store, not in a shared service.

Developer Agent and kiweely can be used together: developers on a project use Developer Agent in their editor, while functional consultants build, validate and test Extend apps in kiweely against the same client's Development tenant. See multi-client data separation in kiweely for the details.