Tests are more important than ever as AI accelerates code changes, and the most effective strategy is to convert as much as possible into deterministic, automated checks—hand-written unit tests for complex logic and especially end-to-end tests—so outcomes do not depend on the AI's discretion and token spend stays lower than letting agents fumble.
In response to TLC's question about how to test AI-generated code, Wes emphasizes hand-written unit tests for complex logic (e.g., URL parsing) and Scott stresses end-to-end tests as the way to make verification deterministic, noting deterministic tests use fewer tokens than letting agents open a browser to poke around. ✦ AI generated
Wes · Syntax · 2026-08-05 · original ↗
starts at this moment · 14:28
“I hear a lot of people talking about how LLMs have helped accelerate the speed of development, but testing is more important than ever. What I haven't heard is that people are what are people actually using to write tests?”
your tests are more important than ever as as people are just prompting and adding features to absolutely everything. You need to feel extremely confident. There's the amount of pull requests that are coming in, the amount of code changes that are coming coming in have skyrocketed. ... Unit tests first for for complex logic is something that I will do quite a bit. ... the end-to-end integration tests are super important for these types of things because you want to make sure that you can still use it. ... Tools like Playright are so good for this because you can one see it happen. ... I really lean hard on end to-end tests, especially in this and moving those possible things out of the AI's discretion, right? Oh, fire up a Chrome MCP or fire up an agent browser and make sure this thing works. Uh, how about instead write the playright test, click through it, and then you can run those tests. ... you you see them doing that all the time and and that's that's way more expensive than just running a deterministic test.
verbatim transcript · starts at 14:28
14:09analogy like you had mentioned, but the the cardboard nature of some of the the homes that are built in that era, McMansion kind of era of homes >> is very analogous to, you know, the way that AI stuff is is [music] turned out here. So just add it on. >> And if you want to see all of the errors in your application, you'll want to check out Sentry at centry.io/sintax.
14:36You don't want a production application out there that well, you have no visibility into in case something is blowing up and you might not even know it. So head on over to centry.io/sax. Again, we've been using this tool for a long time and it totally [music] rules. All right, next question from TLC says, "I hear a lot of people talking about how LLMs have helped accelerate the speed of
14:58development, but testing is more important than ever. What I haven't heard is that people are what are people actually using to write tests? Are people writing them by hand? Are we using cloud to generate tests on the code we just wrote? What about more specific endto-end and edge cases? It seems like it would be a huge token spend to explore and write pre playright tests for a complex site using an agent.
15:22So yes, a few things. First of all, your your tests are more important than ever as as people are just prompting and adding features to absolutely everything. You need to feel extremely confident. There's the amount of pull requests that are coming in, the amount of code changes that are coming coming in have skyrocketed. And you need to have confidence that adding these things is not going to break uh other parts of
15:49your website, right? Like I get all of these pull requests on some of my sites that that don't have tests and I'm just like I I don't know if you actually spent time doing this or if you just drive by prompt, pull request something and it's going to break something else. So the tests are more important than ever. I totally agree. Um, so a few thoughts here. Unit tests first for for
16:11complex logic is something that I will do quite a bit. So one example on my personal website, I needed to be able to take in a URL for any social network and parse out the ID of that URL. Fairly simple function where I don't know there's like 10 different kinds of social networks and there's probably 20 different types of links. You know, there's YouTube.com, there's you u2.be,
16:36there's twitter.com, there's x.com, there's there's threads, there's blue sky. They all have six or seven different types of URLs that you can. So I basically what I did was I took every possible example of every URL that I want and I wrote unit tests for it. So taking this expect the platform to be Twitter and the ID to be ABC123. And then having that I was so much more
17:01confident being like all right AI please write a function that will cover every single one and you can kind of loop through it and make sure that it hits every single one of those use cases. So unit test extremely important in that regard and then the endto-end integration tests are super important for these types of things because you want to make sure that you can still use