After 2.5 years building, we're shutting it down

For two years I've been telling agency founders to build AI-native businesses. Now we're doing it to our own, where you can watch.

We're turning all of it, the frameworks, the processes, the methodology we've built, into a data layer that AI can actually think with. Then we're building agents that run parts of the business. And we're documenting it publicly each week: what we're building, what's working, what's breaking, what we're learning. On LinkedIn, in this newsletter, and on YouTube.

I've been thinking about why I want to do this in the open, and the honest answer is simple. I've told a lot of agencies to do this. If we can't do it ourselves, in our own business, with our own methodology, then what are we actually selling?

So here's the plan, in a nutshell.

Step 1: The data layer

This is where it starts, and this is where the bulk of the work lives.

Before you can build agents, before you can automate anything meaningful, you need your data organised in a way that AI can actually use. A lot of agencies skip this step entirely. They go straight to "let's build an AI agent" without realising the agent has nothing to think with.

For us, this means collecting and structuring everything:

• Quantitative data: client performance metrics, revenue data, delivery KPIs
• Qualitative data: our frameworks and methodology, how we think about agency operations, founder psychology, hiring, offer design
• Sales and delivery playbooks: what we actually do with clients, step by step
• Decision frameworks: how we diagnose problems, how we prioritise, what we look for in an audit

All of it needs to live in one place, properly named, properly structured, queryable. We're leaning towards Vercel for storage plus Obsidian for the infrastructure, and bringing in an engineer to build the data architecture properly. This isn't a "write some docs in Notion" job. It's the foundation everything else runs on.

I'll be straight with you. This is the hardest part. It's not glamorous at all.

Step 2: Agents

Once the data layer is built, the next step is designing agents that actually run parts of the business. Real operational agents that own a function, well beyond chatbots that summarise your emails. We'll turn some of them into micro-SaaS products our clients can use for free.

• Client Success Agent
• Agency Operations Expert Agent (likely Kristina's digital twin)
• Agency Growth Agent (likely my digital twin)
• Content Strategy Agent (already built and in motion)

What does the entire system look like for us? I don't fully know yet. That's part of the point of doing this publicly. We'll figure it out as we build it and share what we learn along the way.

Why I'm telling you this

Two reasons. First, a lot of agency founders want to do this but don't know where to start. Watching someone else go through it, including the messy parts, is more useful than another "5 ways AI will change your agency" post.

Second, I want to be accountable. If I'm going to advise founders to build AI-native businesses, I should be building one myself. In public. Where you can see the progress and call me out if I'm full of it.

Each week from now on, I'll share an update: what we did, what we learned, what surprised us.

The best feedback isn't always from your clients

This week one of our clients forwarded me a message. Not from them, from their client.

Seeing the systems and processes we build together reflected in the experience Tobi's clients receive is one of the more rewarding parts of my work. The best feedback doesn't always come directly from the people you serve. Sometimes it comes from the people they serve. That message made us smile.