Thoughts

The bottleneck in AI products isn't the model. It's trust.

I've watched dozens of AI products fail, and it's rarely because the underlying model wasn't good enough. GPT-4, Claude, Gemini—they're all capable of impressive things. The problem is that users don't trust the output.

Here's what I mean: when an AI gives you an answer, can you tell why it gave that answer? Can you verify it? Can you correct it when it's wrong?

Most AI products treat the model like a black box. You put in a prompt, you get an answer. Magic! But magic isn't what enterprises need. They need explainability. They need audit trails. They need to understand what went wrong when things go wrong.

When we built the agentic AI at Protexxa, we spent more time on explainability than on the model itself. Every recommendation showed its reasoning. Every alert explained what triggered it. That's what got us through SOC 2 audits. That's what made security teams trust the product.

The next wave of AI products won't compete on capability. They'll compete on trust. Build for that.

The best PMs I know appear lazy. They say no to almost everything. They let meetings happen without them. They ship less than you'd expect.

But here's the thing: they ship the right things.

Early in my career, I measured my impact by volume. How many features did I ship? How many meetings did I attend? How many stakeholders did I align? I was busy. I felt productive. I was also building a lot of things that didn't matter.

The shift happened when I realized that saying yes is easy. Saying no is the hard work. Every yes is a resource commitment—engineering time, design attention, QA cycles. Most products fail not because they lack features, but because they have too many.

At Tempo, we had a backlog of 200+ feature requests. Some had been sitting there for five years. Instead of trying to knock them all out, we asked: which 20 would actually move the needle? We shipped those. NPS went from 38 to 52. Retention improved. Customers felt heard—not because we built everything, but because we built the right things.

The lazy PMs aren't actually lazy. They're just protecting their team's attention like it's the scarce resource it is.

I used to post on LinkedIn religiously. Thought leadership, they called it. Share your insights. Build your personal brand. The algorithm will reward you.

Then I noticed something: the posts that performed best weren't my best thinking. They were my most shareable thinking. Contrarian hot takes. Rage bait disguised as wisdom. Content optimized for engagement, not insight.

The platform rewards performance, not depth. You learn to write for the algorithm, not for the reader. And slowly, the way you think starts to mirror the way you post.

So I stopped.

This site is my alternative. No algorithm. No engagement metrics. Just a place to write things that might be useful to people who actually want to read them. The audience self-selects. If you're here, you're probably the kind of person I'd want to talk to anyway.

I don't have tips for going viral. I have thoughts I've been thinking about. Some of them might be wrong. All of them are honest.

That's the trade-off I'm making. Less reach. More signal.