Why We Built Teeme

Running AI agents in production is not the hard part. Keeping them from quietly doing the wrong thing — at 3am, without telling anyone — is.

We built Teeme because we lived the alternative. Inscape Interiors, our design and build company in Luxembourg, runs 8 AI agents: one for client communications, one for finance, one for SEO, one for sales CRM, and more. They generate tasks, call tools, send emails, and file proposals — around the clock, without anyone watching.

For the first six months, it worked. Then it didn't.

What Breaks Without Governance

The failures weren't dramatic. There was no runaway agent, no deleted database, no wrong invoice sent. Instead, it was quieter:

• An agent submitted the same proposal four times because it couldn't tell whether the first one had been approved
• Two agents created competing tasks for the same client email, each marking their version as the canonical one
• A finance agent classified an expense in a category that required human review — and then completed the task anyway because the check wasn't wired to the execution path
• A content agent published a blog draft that was technically approved but hadn't been through the brand tone review

None of these were catastrophic. All of them were expensive — in time to debug, in trust to rebuild, and in edge cases to patch.

The Missing Layer

What we needed wasn't a better orchestration engine. We had one. We needed governance: the ability to see, at a glance, which agent was doing what, how much it cost, whether it needed approval, and what would happen if we rolled it back.

That's a different problem from "make the agents run." It's closer to managing a team — you need org charts, approval workflows, budget limits, and an audit trail.

We built that layer for Inscape. We call it Teeme.

What Governance Actually Looks Like

Governance in Teeme is not a policy document. It's a runtime constraint:

• Every agent has an explicit tier: AUTO (just do it), AUTO+ (do it and notify), PROPOSE (do it with a 15-minute rollback window), CONFIRM (wait for a human), or BLOCK (never)
• Every proposal is auditable — who proposed it, what it did, what the rollback would be
• Every LLM call is costed and attributed
• Every agent has an org chart position — you can see at a glance who reports to whom and what each one owns

You don't need this when you have one agent. You need it the moment you have two.

Why Now

The market is about to catch up. EU AI Act enforcement begins in August 2026. Organizations deploying AI systems will need to document decisions, track outputs, and demonstrate oversight. That's not a compliance checkbox — it's governance, and it's what Teeme does by default.

We built it for ourselves. We're shipping it for everyone.