Things I've learned along the way

The speed layer validates. The scale layer serves.

Agentic development makes this more explicit than ever. Most companies are running two cycles at once, one chasing market fit and one serving paying customers. They have different tolerances and different failure modes. A good platform strategy holds both, whether that's two separate systems or one system with enough convention baked in that experiments are easy to spin up and just as easy to throw away.

Cognitive debt is real and it compounds.

Agentic tools make it easy to ship code you don't fully understand. Do your future self a favor and stay disciplined. Know what you're building. Good design, less complexity, fewer blackboxes. It's also worth remembering that frontier model access isn't guaranteed and isn't getting cheaper. Don't build a system that only works if that layer stays available and affordable.

Follow the signal.

Build for the behavior you observe, not the one you hoped for. Get clarity early: vet the direction, align with the right people, commit. Assumptions are expensive and the longer they go unchallenged the harder the pivot.

Reduce before you build.

Ask if it needs to exist. Look for what's already there. Some of the best decisions don't produce any code at all.

Stick with monolithic design.

Don't architect for a version of the company that doesn't exist yet.

Don't re-platform first during mergers.

It always feels like the right call: clean slate, one system, fresh start. The context living in the old system is worth more than you think, and the opportunity cost during a transition is real. Understand the problem space first, consolidate later if at all.