Agentic Enterprise Blueprint
Digital Workplace Data & Agentic AI Strategy
How a regulated enterprise moves from scattered agent experiments to agents that are designed, governed and adopted the same way — on ChatGPT and Microsoft Copilot.
01Context
Agents were becoming the new interface for colleague servicing and knowledge discovery. Teams could build them quickly on ChatGPT and Microsoft Copilot — but each one was solving integration, security and governance on its own.
In a regulated financial-services company, that is the slow path: every agent becomes a bespoke review, and nothing learned on one is reused on the next.
The challenge: Create one way to design, connect, govern and adopt agents across the Digital Workplace — fast enough for teams to want it, rigorous enough for architecture, risk and governance to approve it.
02Approach
Set the strategy and roadmap
Defined the Digital Workplace Data & Agentic AI strategy and the enterprise roadmap for knowledge discovery and next-generation servicing.
Standardise how agents connect
Established reusable MCP and A2A patterns, so tools and agent-to-agent handoffs are built once and reused across surfaces.
Simplify the orchestration core
Simplified the Amex Agentic Brain architecture that agent workflows run on.
Make governance part of the design
Authored Architecture Decision Records (ADRs) that secured architecture, risk and governance approval for production agentic workflows.
Decide build vs buy on evidence
Led the TechCare Blueprint assessment for incident categorisation — one of 21 critical journeys — comparing custom AI, ServiceNow Now Assist and Moveworks on architecture, security, risk and data readiness.
03Architecture
04Key decisions
Open protocols over platform-specific integrations
- Why
- Agents had to work across ChatGPT and Microsoft Copilot, not just one of them.
- Trade-off
- A shared MCP / A2A pattern takes more design up front than a single-vendor plugin, but every later agent reuses it.
Governance designed in, not reviewed in
- Why
- In a regulated enterprise, approval is the critical path to production.
- Trade-off
- Writing ADRs alongside the architecture is slower for the first workflow and much faster for every one after it.
Build vs buy decided per journey
- Why
- Some journeys are best served by platform features, others need custom AI.
- Trade-off
- Scoring options on architecture, security, risk and data readiness replaces a blanket policy with evidence for each journey.
05Outcomes
- Agent adoption scaled on ChatGPT and Microsoft Copilot across the Digital Workplace.
- Production agentic workflows approved through architecture, risk and governance review.
- Reusable MCP / A2A patterns that new agents build on instead of starting from scratch.