From Leading Teams to Leading AI Agents

Simon Kihlberg Wallström • January 2, 2026 • 4 min read •

ai-development engineering-leadership founder-life building-in-public

I recently left my position as Head of Architecture and Engineering Manager at Telavox to co-found OAIZ. So now I’m an individual contributor again.

But I’m not coding the way I used to. I’m leading a team of AI agents, designing systems, and shaping product. Turns out all those years in management prepared me for exactly this.

The Unexpected Head Start

If you’d told me my management experience would make me a better solo builder, I would’ve laughed. Managing people and writing code are different skills.

Except now they’re not. Being an IC in the AI era is less about syntax and more about:

Keeping the bird’s eye view while staying hands-on
Breaking down problems into clear, executable tasks
Setting up environments where others can succeed
Actually delegating

Let me walk through how this plays out.

Start with the Bird’s Eye View

Every day begins the same way: zooming out.

Before prompting an agent, I think about where this fits. Product roadmap. Technical architecture. What we’re building toward in 3-6 months.

Then I sketch. Pen and paper. Excalidraw. Whatever gets the structure visible before anything starts executing.

This is pure architecture brain. The same thinking that used to go into system design documents now goes into orchestrating AI workflows. The abstraction level changed. The skill didn’t.

Then Break It Down

Once I have the picture, I break it into pieces.

I spent years onboarding junior developers. The main job? Making sure they didn’t shoot themselves in the foot. Or the head. Or nuke the codebase on day one.

This meant creating guardrails, writing docs, and providing clear examples of what not to do. Packaging work so someone else could execute it successfully.

That’s exactly what AI agents need.

I wrote about the ROI of investing in this context separately. My cofounder Emil goes deeper on how we structure agents and context in practice. The skill of packaging work for others? Transfers directly.

The difference is my agents actually read the entire README.

Then Let Go

Confession: I used to get dinged on performance reviews for not delegating. “Simon, you’re a great operator, but you need to let go more.”

It came up a lot. I like getting stuff done.

Now I’m forced to practice what I preached. Every task I hand off is headspace for the next problem. If the agents mess up, that’s on my instructions. No performance reviews to schedule.

The skill I spent years working on in traditional management? AI development demands it constantly.

Guardrails That Enable Speed

This only works because we have structure. Not bureaucracy. Guardrails.

Small, reversible bets. Tim Ferriss calls it reversible decisions. Jocko Willink calls it default aggressive. Same idea. When there are unknowns, you have two choices: freeze and overanalyze, or take a small step, learn, and adjust. I lean hard toward action. A couple hours on a bet that doesn’t pan out? Fine. Analysis paralysis? Way more expensive.

Checkpoints, not handoffs. OAIZ is four co-founders. No big team. Each of us works independently. We stay aligned with daily async check-ins and weekly syncs. Enough structure to catch misalignment early. Loose enough to ship.

Constant rebalancing. Long-term vision vs. keeping current customers happy vs. new opportunities. We revisit this weekly. Priorities shift. That’s fine. The checkpoint catches it.

This was harder at a big company. More layers. More stakeholders. More reasons to wait. With the right guardrails, we move.

The Advice

If you’re making a similar leap - from traditional leadership into AI-first building:

Invest in your tools. Spend real time on your environment, your prompts, your context. It compounds.

Ship constantly. Get things into hands. Your hands. Users’ hands. Real feedback beats planning.

Build for yourself. Be user zero. Feel the friction. Know when it’s good.

Embrace delegation. If you struggled with it before, this is your chance. Your agents won’t judge you.

Pure IC developers aren’t doomed. Management experience isn’t required. But if you’ve spent years learning to break down problems, create structure for others, and think in systems - you’ve got a head start.

The skills transfer. Surprisingly well.