Agents for in-house systems
- Customer
- A materials manufacturer
- Industry
- Manufacturing
- Slides
- 13
- Industry
- Materials manufacturing
- Where it stands
- All in-house systems turned into operating skills, system agents live
- What was delivered
- Agent configuration and an operating skill library for each system
- Where the knowledge came from
- System APIs with business meaning added, plus SAP and the system databases
Redesign and relearning
- Every new system meant the company redesigned its process around it.
- Every new overseas plant meant local staff relearned every system.
Data lake and semantic models
- Asgard loaded SAP and system data into a data lake, with semantic models above it for Mimir.
- Managers ask Mimir for figures in plain language, and the agent queries the data lake through the semantic models.
Methods and approvals
- Each system already has its own API specification.
- Asgard added business meaning to each system's API specification, explaining what each API is for.
- Asgard packaged these specifications as skills. An agent given a business request knows which API to call.
- Asgard changed none of the company's systems.
- Each system gets its own agent. A coordinating agent takes the request and hands it to the system agents.
- Staff give one instruction in Sindri, and each system's agent operates its own system.
- Example: staff hand off overtime, work reporting, pallets, purchase requests and batch lookups in one message. The system agents handle them at the same time.
- When the agent has no valid login, it asks the staff member to sign in to the system, then carries on.
- For anything irreversible, the agent stops for approval.
The customer is de-identified on the slides.
This deck is in Traditional Chinese. There is no translated version.
Slide 1 of 13
Click into the deck to use the arrow keys. On a phone, swipe left or right.
Get started
Book a two-hour workshop
AI feasibility review of a live workflow