Teresa Torres

<start of post>

I don’t think ‘vibe coding’ is the scary part.

I think the scary part is how quickly we stop noticing complexity when it ships in a friendly wrapper.

Last week I had three separate conversations where someone said some version of: “we can just build it now.”

And, sure. Sometimes you can.

But ‘can build’ and ‘should maintain’ are not the same sentence.

Here’s the thing I keep coming back to: the build vs. buy decision didn’t go away. AI just sped up the moment where you have to make it.

Because prototyping is cheaper.

Because demos are easier.

Because you can get to “looks like it works” in a weekend.

And then you wake up a month later with a tool that is now a dependency.

And nobody knows who owns the edge cases.

can we get something working?

can we keep it working?

can we make it safe and boring?

That last one is where most “quick builds” go to die.

A practical way to sanity-check yourself

Start with what happens when the model is wrong.

Not hypothetically wrong. Wrong in the exact way your users are least able to tolerate.

If the cost of being wrong is high, you don’t just need accuracy.

citations people actually click

clear boundaries for what the system won’t answer

an audit trail someone can explain to a stakeholder who is already unhappy

And yes, this applies outside healthcare.

Finance. Legal. Security. Internal tooling that touches payroll. Anything where “oops” becomes a meeting.

A second sanity-check: who owns the data shape?

exports that preserve meaning, not just rows

portability that doesn’t require re-learning your own history

versioning, because your workflow changes and you will want to go back

That’s when “just buy” gets weird too.

So what do I do?

Honestly, I’m slowing down.

I’m trying fewer tools.

I’m doing more small experiments.

I’m writing down what I’d have to maintain if this became real.

Sometimes the right move is to watch and see.

Not because you’re behind.

Because you’re making room for signal.

If you’re in the middle of one of these decisions right now, I’m recording an episode about it.

What we’ll cover

🔁 Why build vs. buy keeps resurfacing

🧪 How AI changes the cost of getting to a prototype (and why that can be misleading)

🔐 The maintenance cliff: evals, monitoring, regressions, and “who debugs this at 2am?”

💎 Data ownership and portability (what to ask vendors before you sign anything)

🧾 Treating the decision as discovery, not a debate

Spotify: https://example.com/spotify

Apple Podcast: https://example.com/apple

YouTube: https://example.com/youtube

And if you’ve built something you now regret maintaining, tell me what surprised you. Comments are open.

<end of post>