EP 53: Cognitive Surrender: The AI Engineering Mistake That Kills Projects

So Ethan, I would like to start this conversation with the problem that I think is hiding in plain sight inside every AI project.

And I'm pretty sure you will agree with that also.

Most people talk about AI saving time, but if every output comes with a second job, which AI is like providing, so proving that AI output was right, sometimes it adds up more tasks to our case.

So we also sometimes call it as a verification tax, right? Mm-hmm.

I understand you have landed somewhere different from where most people are on this because you generally don't double check anymore.

So let's start there.

Let's start talking about the trust factor of the AI output.

And if I can rephrase the question in a better way, what does trust in a number actually look like day-to-day activities when it comes to AI output or stuff that you have worked on for years? Yeah, I think

there's a spectrum.

And when we talk about trust with whatever the LLM comes back with, there's one side, which is very, very risk averse.

Tell me the top 10 brands that you think would be best for parachutes or let me come up with brand names.

It can't really be wrong.

There's a lot of judgment there.

And when you're saying, well, what were my Q4 revenue numbers for 2026 or 2025 year over year? You want that to be right every single time.

I think what's happened is we get so excited about what AI can do that we just say, hey, the number doesn't really matter.

And then when you give it to practitioners and they get in front of a client and they have to explain a number and they can't, that then becomes a tremendous risk to your business, to trust.

And so I think it's definitely that continuum and trust sits right where that number is, right where you're chatting with your data.

And do you really think that sometimes it is most important for us to have the trust? And I think that's where the data underneath the AI system is more important.

So how do we bridge the gap? Like, in your opinion, like, what was that cost of not having the trust at this moment? If you can explain that a little better, given any examples that you encounter so far, working with an AI practitioner or even business leadership.

Yeah.

When working with account teams and you're trying to bring in marketing data and create presentations, if that number is wrong once, then that fear, uncertainty and doubt has been sowed in that individual who's then going to be creating something again.

And maybe they don't use your system anymore.

It was incredibly important when we started pulling data into cloud and the web interface using MCP.

Well, what does what does the BigQuery MCP connector come back with?

oh, wow, it went to the wrong product, the wrong data set, to the wrong