Case study
Building the workflow layer that makes enterprise AI agents usable
Role: Senior AI & Frontend Engineer
Context
Writer is a generative AI platform for large enterprises. Its customers use AI agents and AI-assisted writing inside workflows where errors have real cost — legal, financial services, healthcare.
The problem
The gap between an impressive AI demo and a feature an enterprise will actually roll out is enormous. Output has to be grounded in the customer’s own approved sources. Access has to respect corporate identity systems. Multi-step agent workflows have to be structured, observable, and interruptible — because 'the AI did something unexpected' is not an acceptable incident report.
Constraints
- Enterprise security reviews as a hard gate: SSO, SCIM provisioning, and IdP integration are table stakes, not features
- AI behavior must be explainable to non-technical stakeholders
- The platform ships continuously — production features can’t wait for a grand redesign
- Frontend surfaces span the main app and a Chrome extension with different constraints
My role
I work across the agent platform, knowledge connectors, enterprise identity, analytics, and platform architecture (users, teams, roles, billing) — hands-on, from frontend systems to the server-side logic they depend on.
Approach
Treat agent workflows as structured products, not chat
Built the product structure around how autonomous multi-step workflows are created, executed, and supervised — so users can see what an agent will do, what it did, and where a human belongs in the loop.
Ground AI in customer data through real integrations
Engineered OAuth-based knowledge connectors so model output draws on the customer’s own systems, with token handling and permission scopes that survive a security audit.
Build identity as a first-class system
Designed SSO, SCIM, and IdP provisioning flows from scratch with end-to-end test coverage — the unglamorous work that decides whether an enterprise deal can technically close.
Keep the platform fast while it grows
Ongoing refactoring and performance work across the app and extension, because AI features are only trusted when the product around them feels solid.
Result
- AI agent capabilities that enterprises adopt in production — with the workflow structure, grounding, and identity controls their security teams require.
- Identity and provisioning systems built to meet high-tier enterprise security requirements.
- A platform architecture that lets new AI features ship on top of stable foundations instead of ad-hoc integrations.
Public evidence
Technologies
TypeScript · React · AI agents & orchestration · RAG / knowledge graphs · OAuth 2.0 · SSO / SCIM / IdP · Chrome extension
A note on confidentiality
This case study describes publicly announced Writer capabilities only. Internal metrics and architecture details are omitted.
Related services
Facing something similar?
Describe your situation and I’ll tell you honestly whether my experience maps to it — and what the smallest useful next step would be.