AI Agents Just Got Employee Badges: Developer Launches First Work Platform Where Agents Are Full Users

In a structural departure from AI assistant add-ons, new agent architecture gives AI peers the same roles, permissions, and audit trails as human staff.

BigBlueBam, the open-source work operating system entering public beta this month, today detailed the architectural decision that its creator considers the most important differentiator in the product: AI agents are not features. They are users.

BigBlueBam gives the AI a seat in the org chart. In the database, agents are users. They have roles. They appear in audit logs.”

— Eddie Offermann

In BigBlueBam’s database, agents are represented by standard “users” records with a single flag (`is_agent = true`) distinguishing them from humans. They hold the same role grants. They appear in the same user directory. They generate the same audit records. And they act through a Model Context Protocol (MCP) server that exposes over 340 tools at the time of writing, every one of which is gated by the same permission system that governs human access.

The design is a direct rejection of the “AI copilot” pattern that has dominated enterprise SaaS product announcements for the past three years.

“Most AI work products give you a chatbot in a sidebar,” said Eddie Offermann, the single developer behind BigBlueBam. “That is a cosmetic change. BigBlueBam gives the AI a seat in the org chart. In the database, agents are users. They have roles. They appear in audit logs. When one approves a purchase order, the approval looks identical to a human’s, because architecturally, it is.”

Also Read: AiThority Interview with Glenn Jocher, Founder & CEO, Ultralytics

Why the Architecture Matters

Treating agents as users has several downstream consequences that conventional AI-in-SaaS implementations cannot replicate:

– **Unified accountability.** Every agent action is attributable to a named principal (the agent), scoped by the same permission system humans use, and visible in the same audit log. There is no separate “AI activity” surface to reconcile.
– **Role-appropriate authority.** An agent granted “Project Editor” in Bam could edit tickets but not approve invoices. An agent granted “Finance Approver” in Bill could. The same granularity that applies to human delegation applies to agent delegation.
– **Behavioral configuration separated from authority.** Agents have additional settings (confidence thresholds, rate limits, auto-publish rules, human review triggers) that humans do not need. Those settings live in a separate table. Permissions stay uniform.
– **Natural governance path.** When regulators or auditors ask what the AI did, the answer is a query against the same tables used to answer the question of what any employee did.

The MCP Layer

BigBlueBam ships with a Model Context Protocol server that exposes over 340 tools covering every product in the suite, from creating a project in Bam to sending a Banter message to reading a Beacon knowledge article to generating an invoice in Bill. Agents call those tools through the MCP surface. So does Bolt, the suite’s automation engine, which compiles visual workflow rules into chained MCP calls.

That design produces a single, audited execution substrate for both AI agents and human-authored automation.

“Once MCP is the execution layer, the distinction between ‘the AI did it’ and ‘the workflow did it’ and ‘a human clicked a button’ collapses into a single record of what happened, who caused it, and whether they were authorized,” Offermann said. “That is what compliance looks like when it has been designed in, not bolted on.”

The Governance Question

The architecture surfaces questions that most enterprise AI deployments have not yet had to answer. If an agent approves a fraudulent invoice, who is accountable? BigBlueBam’s answer is not a policy argument. It is a schema.

“Corporate governance was written for a world where every action traces to a named human,” Offermann said. “The honest way to extend it is to give AI agents the same kind of traceable identity. Not a service account. Not an API key. A real user record, with a role, with a manager, with a performance history. You can govern that. You cannot govern a chatbot.”

Also Read: ​​The Infrastructure War Behind the AI Boom

[To share your insights with us, please write to psen@itechseries.com]