Why now
Everyday work agents can take on
- 01
An agent takes a goal and does the work
The agent plans its own steps. It looks data up, fills forms in and submits them. It also writes documents, meeting notes and emails.
- 02
Agents can work at any hour
An agent works 24 hours a day and is always on call. It speaks many languages and can write code.
- 03
The difference between companies is now how well they use AI
Companies that are great at AI give everyday work to agents. Staff make the judgment calls and review the agent's work.
Industry scenarios
Industry scenarios
- 01Manufacturing
Rush-order scheduling
- Problem
- High-mix rush orders are scheduled from senior staff's experience, and departments often work from different information.
- What Asgard does
- The agent reads the scheduling rules and exceptions, queries live material and work-order data, and proposes a schedule. Before it changes a work order, it stops for approval.
- Outcome
- Schedules follow live material status, and every department sees the same information.
- 02Supply chain
Inventory and purchasing
- Problem
- Physical stock and ERP data don't match, purchasing decisions go wrong and cash sits in inventory.
- What Asgard does
- The agent compares physical stock with system data and forecasts purchasing needs. When material runs short, it prepares the purchase order and stops for approval before it submits.
- Outcome
- Stock figures match, and purchasing starts from forecast demand instead of waiting for a shortage.
- 03Retail · demonstration project
Store stock transfer
- Problem
- On Sunday night the POS day-end close shows that a best seller at the flagship store will sell out tomorrow. At retailers like this, a transfer order used to come only after the store manager reported on Monday and the operations team met.
- What Asgard does
- The agent uses semantic models to check stock and promotions across stores, the distribution centre and e-commerce, and finds stores that can transfer stock and a restock option. Before the transfer orders, the restock purchase order and the promotion change are written back, the agent stops for the operations manager's approval.
- Outcome
- Sindri gathers the transfer, the restock, the limited promotion and the restock note for the store manager into one decision pack. Only after the operations manager approves them together on Monday morning does the agent write back to the systems.
- 04Multi-brand e-commerce
One back office for many brands
- Problem
- With several brands, products, orders, stock and sync errors end up in different places to check one by one.
- What Asgard does
- Connect SHOPLINE and bring every brand's products, orders, stock and exceptions into one back office, with AI assistants splitting the work.
- Outcome
- AI assistants can act, but every write needs a person's approval before it takes effect.
- 05Public relations
Responding to issues
- Problem
- When an issue breaks, spotting it, reading it and drafting a reply are all manual, so responses lag.
- What Asgard does
- Track hot topics across news, social media and forums, read the trend, then draft statements and social posts in the brand's established voice.
- Outcome
- When an issue appears, the team starts from a draft and reviews it.
- 06Media
From trend to publication
- Problem
- Planning, writing, images and publishing are all manual, so content arrives after the trend has passed.
- What Asgard does
- Detect trends, write, add images and publish in several formats to each platform, as one fixed workflow.
- Outcome
- Production keeps pace with trends, and editors spend their time on review and judgement.
- 07Enterprise knowledge
SOPs and technical documents
- Problem
- Know-how and SOPs are scattered across shared folders, email and paper, and key skills leave with the people who hold them.
- What Asgard does
- Documents, spreadsheets and interviews become the wiki and the procedures, then skills the agents use. Gaps go out as tickets to the department that owns the answer.
- Outcome
- Knowledge updates sync to the agents without a redeployment.
- 08Logistics & admin
Cross-system admin work
- Problem
- Logistics scheduling and data updates across systems still depend on people copying data and filling in forms.
- What Asgard does
- The agent signs in to each system as the staff member, looks data up, fills forms in and updates schedules. For web systems with no API, it drives a browser.
- Outcome
- Staff review what the agent prepared, and approve or reject it.
- 09Your industry
Your company
See customer casesNot your industry? The customer cases page shows Asgard rollouts in other industries.
Rollout
How the rollout works
- 01
Discovery
The FDE team interviews your team and collects how your company works and its documents. - 02
Ask the AI consultant
The FDE team asks the AI consultant to put these documents onto the Asgard Platform. The AI consultant is a coding agent with asgard-cli loaded. - 03
Write the settings
The AI consultant reads the LLM Wiki in asgard-cli. Then it writes your procedures as agent settings. The settings are stored in your Asgard workspace. - 04
Deploy
The AI consultant deploys the settings to Asgard Platform.
Compare
Asgard and other enterprise AI tools
- Built in
- Partial, or needs extra build
- Not supported
- Not published
? = term explained
| Capability | AsgardAsgard AI | ChatGPT EnterpriseOpenAI | Claude EnterpriseAnthropic | Microsoft CopilotMicrosoft |
|---|---|---|---|---|
| Isolated environment per taskThe code and files an agent works on stay apart from other work, and are cleared when it ends. | Built in Each task gets its own , removed when done; the agent runs code, handles files, and opens a browser where the person signs in to company systems | Partial, or needs extra build ChatGPT Work runs in isolated VM-backed environments that may be reused for later tasks; in Enterprise the cloud browser cannot sign in to websites | Built in Cowork gives each session its own sandbox, deleted when it ends; the browser runs on the user's own computer | Partial, or needs extra build Researcher with Computer Use opens a sandbox VM per conversation, discarded after use (Frontier early-access program) |
| Vendor-built custom agentsEvery company's processes and data differ, so someone has to build the agents on site to fit them. | Built in An interviews your team, builds agents from your processes and data, gives each agent only the it uses, and hands them over to your IT | Partial, or needs extra build OpenAI FDEs on the Frontier platform build with your team | Partial, or needs extra build Anthropic engineers build with partners; a program trains customer engineers | Partial, or needs extra build Microsoft FDEs co-build through Agent Factory, subject to eligibility |
| Choice of modelsCompanies pick models by task, cost and company policy. | Built in connects Claude, OpenAI, Gemini and other vendors, and your own API keys | Not supported OpenAI models only | Not supported Claude models only | Built in OpenAI and Anthropic models (an administrator enables Anthropic) |
| Layered permissions ()Administrators decide by department and role who can use which agent and see which data. | Built in Management Console grants groups and roles agent by agent and model by model | Built in Custom roles and groups (Enterprise plan) | Built in Custom roles and groups (Enterprise plan) | Built in Uses existing Microsoft 365 permissions |
| Knowledge and database queriesAgents need to read company documents and query live numbers in databases. | Built in Enterprise wiki holds the rules; query databases; cloud drives sync on a schedule | Built in Company knowledge and apps; SQL through apps such as Snowflake | Built in Connectors; SQL through connectors such as BigQuery | Built in Microsoft Graph; live queries through MCP connectors |
| Cloud and on-premisesSome companies cannot put their data in a public cloud and must run systems in their own data centre. | Built in Cloud edition; on-premises delivered as a custom build | Not published Only data residency regions are documented | Partial, or needs extra build Claude Desktop inference can run through your own cloud account | Not supported Documented as a cloud service |
| Several agents and schedulesWork often crosses departments, so several agents take turns and some work runs on a schedule. | Built in Sindri splits one conversation across several agents; Routines run on a schedule | Built in GPT-6 Pro runs several agents in parallel; workspace agents run on a schedule | Built in Cowork coordinates several sub-agents; tasks can run on a schedule | Built in Copilot Studio agents hand off to each other; Cowork runs scheduled prompts |
| Procedures written as A department writes its procedure once, and several agents follow the same steps. | Built in IT or the FDE writes skills in Odin, shared across agents | Built in GPTs, workspace agents and skills | Built in Organization-wide skills | Built in Copilot Studio and custom skills |
| Approval before writing backBefore an agent changes data in a system, a person with permission decides. | Built in Stops and lists what it would write: allow for this conversation, allow once, or deny | Built in Write actions ask every time by default; the builder can change this | Built in Write tools need approval once per task by default | Built in Write tools need approval by default |
| Usage records and auditYou can find out later who used AI, when, and for what. | Built in records the key, model and tokens of every call; admin actions have an audit log | Built in Compliance Logs (Enterprise plan) | Built in Audit logs and Compliance API (Enterprise plan) | Built in Microsoft Purview audit |
The comparison covers each product's enterprise plan. Claude Enterprise includes Claude Cowork. Microsoft Copilot means the version for organizations, previously named Microsoft 365 Copilot. Compiled from each vendor's public information in October 2026. The vendors' latest announcements take precedence. ChatGPT, Claude and Microsoft Copilot are trademarks of their owners. OpenAI, Anthropic and Microsoft have not reviewed this comparison.
Book a two-hour workshop
AI feasibility review of a live workflow