Enterprise AI Engineering
Building the engineering organization, platforms, and technical foundations that bring AI into production at enterprise scale.
- Where
- Kainos
- When
- 2026–Present
- Scope
- AI Engineering · Enterprise Architecture
- Written
- 4 of 8 sections
Context
Kainos is a digital services company and a long-standing Workday partner. In July 2026 I moved from leading Workday integration delivery to leading AI engineering, with a mandate I would summarise in one line: build AI infrastructure that ships to production.
Problem
Delivery teams repeat the same work across engagements: read a specification, translate it into configuration and code, validate it, deploy it. That repetitive eighty percent crowds out the interesting twenty percent.
General-purpose AI assistants do not know enough about an enterprise platform to take that work off anyone's hands. The gap is not the model. It is platform knowledge, guardrails, and a path to production.
Constraints
The security, data-handling, and governance constraints that shaped the design: what an agent may touch, where human sign-off sits, how client data is kept out of prompts. Write from the governance review, free of client specifics.
Role and responsibilities
Senior Manager, AI Engineering. Hands-on in the architecture and the code, and accountable for the team building it.
Architect, lead builder, and owner of an agent-first product acceleration platform for Workday, now in production. Product owner for the ServiceNow and Jira agentic accelerators.
Technical approach
Development accelerators that turn specifications into working code, built on agentic tooling: Claude Code, Codex, and Cursor.
An agent-first product acceleration platform for Workday, where the UI, an API, a CLI, and an MCP server all drive the same operations. Whoever or whatever is at the keyboard, a person or an agent, gets the same capability through the interface that suits it.
Workday development acceleration across Extend, Integrations, and Prism Analytics.
An internal enablement platform that puts agents on top of where a delivery team already looks for answers.
Underneath them, a layer of MCP servers and agent skills that carry the platform knowledge the agents do not ship with. Knowledge becomes an interface that can be versioned, tested, and reused across engagements, rather than a longer prompt.
Key decisions and tradeoffs
Two or three real tradeoffs, each with the option not taken: what lives in an MCP server versus a skill, where the agent stops and a human signs off, build versus buy at the model layer.
Outcomes
Adoption and cycle-time figures once they exist and are cleared for publication. Measured numbers only, no estimates.
Lessons learned
What changed in the plan once the accelerators met real delivery teams.
Stack
- Claude Code
- Codex
- Cursor
- MCP servers
- Agent skills
- Internal tooling