Closing, Trust, Again

We end where we started. Trust isn't a slide at the end of the deck. It's the whole product.

Back to the beginning

We started with one question:

Would I trust this system to make this decision about me, my family, or my customer?

Everything in between, the principles, the evals, the governance, the empathy, was just engineering toward a confident yes.

The takeaway

You are going to ship AI. A lot of it. Faster than you ever shipped anything else. That is genuinely exciting, and it is genuinely a lot of responsibility.

The good news: the playbook isn't mysterious. It looks like the playbook for every other piece of high-stakes software you've ever shipped:

Know who it's for.
Know what it does, and what it doesn't.
Test it like you mean it.
Make it reviewable.
Own it when it's wrong.
Tell the truth in the UI.

Do those six things, again and again, and trust shows up.

Thank you

Thank you for sitting through this. Thank you for the work you're going to do next week that nobody will see, the eval you'll add, the disclosure you'll write, the "no" you'll say. That work is how this whole thing stays good.

I'm @Codess-Aus. Come find me. Let's keep building.

, Scaling Guacamole, SlashNew Conf 2026.