Own system / AI agents · 2026
Agent platform
A platform on which teams of AI agents take on tasks in sales, order-to-cash and marketing. Every step that leaves the company waits for a person's approval.
Situation
The platform was to hand AI agents connected tasks such as researching leads, preparing quotes and following up invoices. No agent may send a message, book an invoice or delete data without a person's consent.
Approach
Each business unit is a workspace with its own team of agents, its own memory and its own inbox. Each agent has a charter and only the capabilities explicitly assigned to it: skills, tools and connections through the Model Context Protocol. The agent files every outward step as an approval with its exact command, and only a person's yes runs it. Guests work in a locked-down mode with no access to keys or system files.
Result
32 agents in 8 workspaces. Sales (sales-os) takes a lead from research to a won deal, with CRM, sequences, calling, quotes and KPIs. Order-to-cash (company-os) runs from the request through quote, job and invoice to payment and forecast. Marketing and social media have their own teams. Everything is operated through a dashboard, a desktop app or a messenger, and the AI engine is interchangeable.
- Workspaces
- 8
- Agents
- 32
- Library plug-ins
- 22
- Interchangeable AI engines
- 3
What this means for a group
For a group, the platform shows how AI agents can be organised by business unit, how access rights are granted per agent and where the line between proposal and execution runs. That line is also where compliance and co-determination come in.
Year
2026
Role
Concept, architecture, development
Fields
AI agents, Platform, Approval workflows
How this system works inside the Dionics AI operation is described under AI in operation. Your own plans are a topic for a confidential first call.