Most workflow engines run fixed paths. This one builds them as it goes. It takes an interaction — a user action, a Slack message, a call from another assistant — and decides in real time whether to reuse a known flow, assemble a new one, or do nothing. Recommended actions (update the org chart, refresh a doc, send a message) aren’t bolted on at the end; they feed back into the context that drives the next decision.
Overview
The system reads interactions from any source and picks a response by:
- Reusing an existing flow when one fits
- Building a new path on the fly when none does
- Staying quiet when context says no action is needed
Every recommended action becomes an input to the next decision, so the context gets richer as the system runs.
Architecture
Ingestion & Gateway. Normalizes interactions from every channel into one format, and pulls out any embedded directives for later flow construction.
Orchestration Hub. Holds the context and makes the call — reuse, build, or skip — folding recommended actions into that choice.
Flow Planner & Generator. Fetches a predefined flow or generates one from current context, using policy rules and models to pick the path.
Execution Engine. Runs tasks asynchronously so dynamic flows don’t block the main interaction. Handles dependencies, error recovery, and reports status back to the hub.
External Integrations. Connects to Slack, docs platforms, and other services to carry out actions.
State & Context Manager. Keeps state consistent across every interaction and generated flow.
Observability. Event logs, audit trails, live dashboards, and alerts on bottlenecks and errors.
Control Plane. Updates flows and rules — with versioning — without downtime.
Data & Execution Flow
How Context Deepens
The system gets smarter by mapping how actions connect to context, learning from execution feedback, and letting new patterns reshape how it generates flows.
Implementation Notes
- Scale: microservices, async processing.
- Adaptability: a control plane that updates flows and rules live.
- Security: authentication and authorization on every external integration.
The Point
Static workflows become context-driven ones. Instead of running the same path every time, the system adapts each flow to the moment — and every interaction it runs leaves the next one better informed.
Next
- Prototype flow generation against real interactions.
- Wire it into existing context systems for live feedback.
- Add analytics to keep refining the paths.