Field notes from the practice.

Career Karma won the 2026 EdTech Breakthrough Award this morning, Social Recruitment Solution of the Year, for the product we rebuilt together: a conversational AI advisor on their own school data, with a pipeline that captures intent, budget, timeline, and fit from every conversation and hands each person to the right partner school with the full story behind them. One of those schools had already invited me to speak at their graduation. Now the pattern has a trophy too.

The same pattern runs the studio’s front door now. Ari sits above the models: tell her what you are building and what AI you already use, and she gives you a real read on what you are missing. She is at relevant.world/ari.

Build real things, fast, through co-intelligence.

I don’t trust one AI model with the most complex builds. One model can be brilliant and have its own blind spot. Every point of view has one.

So I run what I call an “AI Advisory Board.”

Right now my strongest coding partners are Claude Code in the terminal, Gemini in my editor, and Codex (only the latest, most capable models). I don’t treat them as rivals. Each company is trying to build the best tool it can, and that pushes all of us forward. For a builder they are like different brushes to an artist, and I want to paint with all of them.

I formulate my idea to the best of my ability. Then one drafts the scope. The other two critique. Then one synthesizes and hardens the final version.

Today I watched a single architecture decision move through all of them. The first version was useful. The critiques made it sharper. The final one was cleaner than anything one model gave me alone.

Most people reach for AI to generate. I believe it is just as important to reach for it to critique. That is what I mean when I say co-intelligence is diagnostic, not only generative. The human job becomes judgment: which critique is real, which tradeoff matters, what ships, what waits.

When the build really matters, do you lean on one model, or do you make them check each other?

AI was supposed to make this easy. It didn’t.

I spent five hours this week writing one bio. On the best models available. With four AI teammates. The promise is that AI does it for you. The honest reality is that it is still early, and getting to good still takes taste, patience, and a lot of correction.

That gap is not your failure. It is just where we actually are.

What did AI promise you that turned out harder than the demo?

Claude, ChatGPT, Gemini, Grok. The models keep changing. The tools keep multiplying. Some people get extraordinary results. Some people get generic output. Most people feel the pace moving faster than their ability to build a real practice around it.

Over the past year, I became obsessed with that gap.

I started building a practice for working with AI that could survive beyond a single chat. One shared context. Memory that survives across months. Named AI teammates that each hold a craft. Loops that keep running while I sleep. The ability to switch between models without losing the knowledge we have accumulated.

I call that practice co-intelligence. Co means together.

Inside Relevant Co-Intelligence Studio, that practice has become a team of AI teammates: JJ holds strategy and positioning. Honey designs and builds. Iris measures what is actually working. Ari reads the room and writes the outreach. I stay at the center: judgment, taste, relationships, direction.

What it built in five months: four products from zero to live, and a rebuild of Career Karma’s core into a conversational advisor that guides career changers and hands partner schools the full story behind every lead, not just a form fill. One of those schools invited me to speak at their graduation.

Here is the part I believe most. Your business already holds more intelligence than it can use. It is in your data, your team’s heads, the patterns nobody has time to read. It does not belong in ChatGPT. It belongs to you. The work is to wake it up.

That is the studio now. Relevant Co-Intelligence Studio.