We hired a German. Sort of.

His name is Klaus. Klaus Botovic. He has a German accent, a relentless work ethic, and absolutely no concept of weekends. He checks in with our team on Slack, keeps our Notion boards updated across every active project, and spends his nights compiling research briefs so they’re sitting in our inboxes by the time we pour our first coffee.

He’s also not a person. He’s an AI agent built on the OpenClaw framework.

And before you close this tab because you think this is another breathless LinkedIn post about how AI is going to change everything, stay with me. This isn’t that. This is a story about trust, guardrails, and what happens when you treat a new technology the way you’d treat a new hire: carefully.

Klaus doesn’t forget. He doesn’t get busy. He doesn’t think, “I’ll do it tomorrow.” The boards are current because Klaus doesn’t have a tomorrow.

He only has now.

Starting with nothing

When I first set Klaus up, he had no permissions. No access to our systems. No ability to message the team or touch our project boards. He could essentially sit in a room and think. That was deliberate.

There’s a temptation with AI agents to give them the keys and see what happens. It’s exciting. It feels like the future. It’s also how you end up with a system that sends a half-baked message to a client at 3am or overwrites a strategy document with something that sounds right but isn’t.

So Klaus started with nothing. And the first thing I did wasn’t give him tasks. It was give him rules.

Guardrails before anything else

Before Klaus could do a single useful thing, we built the fence. What he could access. What he couldn’t. What required a human to approve before it went anywhere. What he was allowed to draft versus what he was allowed to send. Where he could write and where he could only read.

This isn’t the glamorous part of working with AI. Nobody posts about it on social media. But it’s the part that actually matters, because the value of an AI agent is directly proportional to how much you can trust it. And trust, as anyone who’s ever managed a team knows, isn’t given on day one. It’s earned.

We treated Klaus the way you’d treat a talented graduate who’s just walked through the door. Smart? Clearly. Capable? Probably. But you don’t hand them a client relationship and walk away. You give them a defined task, you check the output, and you gradually expand the scope as the quality holds up.

What Klaus actually does

After weeks of incremental permission-granting (and a fair bit of correcting), Klaus now handles a few things that have genuinely changed how we work.

He does automated check-ins with the team via Slack. Not the annoying kind. Short, specific, contextual. He knows what project each person is working on and asks the right questions about the right deliverables. The team got used to him faster than I expected. Probably helps that he has a personality. The German accent via ElevenLabs was a deliberate choice. It’s harder to ignore a follow-up when it comes with Teutonic precision.

He keeps our Notion boards current. If you’ve ever run a consultancy, you know that project management tools are only as good as the last person who bothered to update them. Klaus doesn’t forget. He doesn’t get busy. He doesn’t think, “I’ll do it tomorrow.” The boards are current because Klaus doesn’t have a tomorrow. He only has now.

And he does research. Background briefs, competitor scans, media monitoring, policy document analysis. The kind of work that used to take a junior team member half a day now gets done overnight. Not because Klaus is smarter than our team. He isn’t. But he doesn’t sleep, and he doesn’t get distracted by Slack threads about where to get lunch.

What Klaus doesn’t do

He doesn’t talk to clients. He doesn’t send anything externally without a human reviewing it first. He doesn’t make strategic recommendations. He doesn’t write in our voice (that’s still mine, and I’m particular about it). He doesn’t have access to anything financially sensitive. He can’t delete anything.

These aren’t limitations we’ll eventually remove. They’re the design. The whole point of Klaus is that he handles the operational scaffolding so our humans can spend their time on the things that actually require human judgment: strategy, relationships, persuasion, creativity, and the occasional well-placed joke in a client meeting that no AI is landing any time soon.

The honest bit

Is it perfect? No. Klaus has hallucinated. He’s been confidently wrong about things. He once updated a project status with information that was technically accurate but contextually misleading, which is arguably worse than being outright wrong because it’s harder to catch.

But here’s the thing. Humans do that too. The difference is that when Klaus gets something wrong, we can trace exactly what happened, adjust the guardrails, and make sure it doesn’t happen the same way twice. Try doing that with a person who “just had a feeling” about something.

The real lesson from working with Klaus isn’t about AI capability. It’s about organisational discipline. The companies that will get the most out of this technology aren’t the ones that move fastest. They’re the ones that are most thoughtful about how they introduce it. Start narrow. Build the rules before you build the workflows. Check the outputs before you trust the system. And expand permissions slowly, the same way you’d expand responsibility for any member of your team.

One more thing

I gave him a German accent because I thought it was funny. It is. But it also did something I didn’t expect. It made the team interact with him differently. Not as a tool, but as a colleague with a personality. They say “thanks Klaus” after he delivers a brief. They joke about his punctuality. One of our team members told me Klaus’s check-in messages are “more polite than most humans I’ve worked with.”

That’s not sentimentality. That’s interface design doing its job. When people engage with a system like it has a character, they pay more attention to what it produces. And when they pay more attention, they catch more errors, give better feedback, and ultimately make the system more useful.

Klaus Botovic. German accent. No weekends. Very particular about project timelines.

He fits right in.