#455: Gilded Python and Beyond

00:00 Hello and welcome to Python Bytes, where we deliver Python news and headlines directly to your earbuds.

00:05 This is episode 455, recorded October 27th, 2025.

00:10 And I'm Brian Okken.

00:11 I'm Michael Kennedy.

00:13 And this episode is sponsored by the wonderful people at Python Bytes.

00:17 Actually, both and everybody here.

00:20 So please check out the work we do for a little bit of money.

00:26 We've got Talk Python training with tons of wonderful courses.

00:29 And his new book, of course, Michael's new book.

00:32 And then also, if you want to check out pytest, there's the complete pytest course.

00:38 Or you can take the pytest course at Talk Python Training.

00:41 And as always, thank you to our Patreon supporters.

00:45 You guys rock.

00:47 We don't shout you out all the time, but we always appreciate you.

00:50 If you'd like to submit a topic idea or some feedback for us or just say hi,

00:57 Check out the show notes at pythonbytes.fm.

01:00 There's links to our socials.

01:04 We've got the show, and both Michael and I are on Bluesky and Fosstodon,

01:10 but that's really Mastodon, so wherever you're at there.

01:14 If you're listening to this and you're like, what, sounds like they're recording live somewhere.

01:18 We can see it.

01:19 Yes, we are recording and streaming, but you can watch all the back episodes or participate in the chat if you want to.

01:28 Check out pythonbytes.fm/live with all the details there and a link to when we're going to record next.

01:36 Before we get started, one last thing.

01:39 We do send out a newsletter that's getting better and better as we go along,

01:44 thanks to both Michael and I working on it a little bit.

01:47 So there's a newsletter we send out with all the links.

01:51 So you don't have to take notes.

01:53 We'll send you out an email.

01:55 So sign up for the newsletter and we'll send you those links and extra details too.

02:00 It's a great resource.

02:02 Yeah, I think people really appreciate it, Brian.

02:04 You know, if you look at when you send out the email and then book back in the campaign

02:09 and how many people actually open to the thing that we send out or click on it,

02:13 those are really close to 100%, which is ridiculous.

02:16 So that's a really good sign that other people might also like it.

02:20 Yeah, I think that people are using it in place of taking notes.

02:23 And also, if you're in a car or something, you don't want to try to remember.

02:26 You don't have to.

02:27 So awesome.

02:29 All right, Michael, what do you got for us right away?

02:33 Well, let me see what I can do here.

02:37 Let's see what the Cyclops says.

02:39 A Cyclops.

02:40 It's like a little programming kitty, but it's a Cyclops.

02:43 It's a cute.

02:44 Like Opt, O-P-T.

02:46 operations cycle.

02:48 I get it.

02:49 So we've heard of Click.

02:51 You and I were both just actually singing the praises of ArgPars,

02:55 which is pretty interesting.

02:57 I am such a fan of ArgPars these days because it's just built in and it's

03:02 simple and it's good enough.

03:04 If you were building something where the CLI

03:08 API or interface or dev user experience was super

03:12 important, you might choose something other than ArgPars.

03:15 So what are your options?

03:16 You could choose Click, you could choose Typer.

03:19 Typer is actually built on Click, so in a sense it's like Starlette and FastAPI.

03:23 You choose one, you get the other in a sense.

03:25 So there's a blend there, and there's others.

03:27 I know there's a bunch of different CLI building options.

03:31 But the Cyclops is one that sort of has the second mover advantage over Typer.

03:37 And their sort of tagline, I don't remember where I saw it, but

03:40 somewhere in here, it's not actually right here on this page.

03:44 but no shade at typer, but it says what you thought typer was going to be.

03:48 And my second boomer advantage is like, all right, we've seen different interface things built based around type information.

03:55 And now that we've seen the pluses and the minuses, could we do a slightly different version that has more pluses and fewer minuses?

04:02 And so there's a whole page of the typer comparison here.

04:05 And I'm not a super expert on typer.

04:09 My CLI, who is not that great.

04:12 So there's 13 different things, 13 different components or features or aspects of the API

04:20 that they chose to address to say, Typer's great.

04:23 We're basically inspired by Typer.

04:25 That's what we're doing.

04:26 However, there's a couple of these 13 areas I think we should make better, right?

04:30 So you could actually go through and look at all of them.

04:33 Probably the biggest one is that Typer was made with Python 3.7, 3.8,

04:39 where Python's types were at an earlier stage in their life.

04:44 You had to say things like, this is an argument of int,

04:49 or it is an int when it's actually an argument of int or something like that.

04:52 And this Cyclops thing was built once annotated was introduced,

04:58 which allows you to have sort of a dual role.

05:00 Like, it really is one of these, but its type could also be like its native value, right?

05:05 Allowing the removal of proxy defaults.

05:08 Why does that matter?

05:09 Because if you build a CLI with Cyclops, and you use types,

05:14 you can still use it as regular functions as if they just had an integer

05:18 where it said it took an integer and so on.

05:20 So things like that.

05:21 It's got about, I think, a thousand GitHub stars or so.

05:24 It's not crazy, crazy popular.

05:26 But if you're really loving that concept of having types for your CLI,

05:31 give this thing a look.

05:33 Yeah, actually, I got a shout out to him for doing things like comparing to Typer

05:39 because, like you said, a lot of us, I use Cyper as well.

05:43 And I've used Typer, I've used Click and ArgParse.

05:47 And so you're probably not going to grab too many of the ArgParse people

05:53 because they're just using it because it's built in, maybe.

05:57 You're like the unit test people, Brian.

05:59 What can you do?

06:01 You talk pytest for 10 years.

06:02 I yell louder.

06:04 That's what I do.

06:06 But I think it's a fair thing because I'd be thinking, well, if I'm not going to use ArgParse, I'm probably going to reach for Typer or Click, lately Typer.

06:17 But having this comparison here and then even a migration from Typer, I think that's cool to show that.

06:27 And there's always room for making CLI tools better.

06:31 We love CLI tools.

06:32 Yeah.

06:33 Yeah, there definitely are.

06:34 But yeah, so cool.

06:35 I ran across this and I built something with it recently and it was nice.

06:40 Oh, did you?

06:41 Checking out.

06:41 Yeah.

06:42 A little, you know, I'm starting to add.

06:44 I've been kind of, I've drunk the Kool-Aid now that maybe having an actual CLI,

06:50 even from my own tools, might be a little bit better than just,

06:53 here's a script I run and maybe I'll pass it something

06:56 and just look at sys.argv and see if it's in there.

06:59 Oh, man.

07:00 Let's do a little bit better.

07:01 And now that I've started to do that, I've been playing with these different ones,

07:04 which is, I think, where I kind of got inspired to go down this path and talk about it today.

07:08 Yeah, awesome.

07:09 Cool.

07:11 I want to talk about the present, the future, a little bit of both.

07:17 So there is this interesting article, of course, in Python 3.14.

07:22 Well, we had it in 3.13, is that the GIL was optional with what they call the free threading version, right?

07:32 Free threaded Python.

07:33 Not the gill-less, not the galectomy.

07:37 Free-threaded, I think, is where we landed from the PEP.

07:40 Okay, but is it kind of the same thing?

07:42 Yeah.

07:43 Okay.

07:46 Oh, right.

07:47 The galectomy was a different project.

07:48 Yeah, it was.

07:49 Gosh.

07:50 Anyway.

07:51 Larry Hastings was doing that.

07:53 But I believe, I mean, the free-threaded stuff, there's a lot of things that had to happen.

07:57 And actually, my next topic is also on free-threaded Python.

08:00 And it's not just let's take out the gill, but they had to rework memory management.

08:04 the structure of object.

08:05 There's like a lot of moving parts in order to make this possible.

08:09 Okay, cool.

08:11 Okay, so with the free-threaded Python, so now we have in 3.13,

08:16 we had two versions of Python released.

08:18 So if you go to download Python, you get a choice.

08:21 And with UV2, you can add a T to it, 3.14 T, and you get the free-threaded version.

08:29 But like, and also definitely encourage everybody test and publish that you've tested.

08:36 In CI, you go ahead and test it because the CI tools support both.

08:42 Why do we care about this?

08:44 Well, because some people are migrating to it because as with 3.14,

08:49 it's no longer just an experimental thing.

08:52 So here, I'm going to cover an article by Giovanni Barili.

08:58 Sorry, Giovanni.

08:59 I can pronounce that one.

09:00 I'm pretty sure this is the same Giovanni that makes Granian.

09:04 Oh, okay.

09:06 I think so.

09:07 The future of Python web services looks gill-free.

09:10 So it starts out with two major changes when comparing free-threaded variant of 3.14 versus 3.13.

09:19 First, free-threaded support now reaches phase two, meaning it is no longer considered experimental.

09:25 So everybody's on board with this is what we're going to at least try to do for a while.

09:30 not just try to do but we're the python core team is supporting this and we're going to do it

09:36 moving forward what does that mean it means that people can depend on a free threaded version and

09:42 change their code accordingly if necessary which is great especially for a lot of async stuff

09:49 like web services so uh secondly the implementation is now completed meaning that the workarounds

09:56 introduced in 3.13 make code sound without the gil are now gone, and free-threaded implementation

10:05 now uses the adaptive interpreter as the gil-enabled variant. What does that mean? It means that the

10:13 additional optimizations... Oh, okay. Also, those facts plus additional optimizations make the

10:19 performance penalty that we had in 313 way better or way less annoying so in 313 the free thread if

10:27 you're doing non-async code just straight code you had to face a like a 35 time penalty which sucks

10:35 um now it's like a five to ten percent difference and i think it's going to shrink even more

10:40 but um so if you're if you don't need the free threaded don't use it unless or or use it whatever

10:47 But if it's time critical, don't right now.

10:51 But I think it's going to get better.

10:54 And especially if you are using async and stuff, it's way faster to use the free-threaded, of course.

11:00 So this article does a shout-out to Miguel Grimberg's article about performance.

11:06 I think we covered it last week.

11:07 We did. You did.

11:09 No, maybe I did. We did. I can't remember who started it.

11:13 So this article has a whole bunch more benchmarks.

11:17 And I'm not going to walk through all of them, but talks about doing a service with JSON responses and both measuring with WSGI and ASCII implementations.

11:31 And yes, talking about using Gradient.

11:35 So that makes sense.

11:37 So I'm going to jump down.

11:38 So there's a bunch of metrics, but basically the end result is if you're using ASCII or ASGI, the free-threaded one is obviously the way to go.

11:49 And you don't get really much of a hit because of free-threaded implementation at all.

11:56 You get speedups and memory usage is reasonable.

12:00 So what I want people to do, if you're going to check this out, is to jump down to the final thoughts because there's some great stuff here.

12:08 Essentially, so what do I want to pick out?

12:13 On asynchronous protocols like ASGI, despite the fact that concurrency model doesn't change that much,

12:20 it's a shift from one event loop per process to one event loop per thread.

12:25 I think that's a big change, actually.

12:27 Just the fact that we no longer need to scale memory allocations

12:30 just to use more CPU is a massive improvement.

12:33 That's cool.

12:34 Didn't know that.

12:35 That's neat.

12:36 For everybody out there coding a web application in Python,

12:40 simplifying the concurrency paradigms and the deployment process of such applications

12:44 is a good thing, obviously.

12:47 And the conclusion being, for me, the future of Python web services

12:52 looks like the Guilfri.

12:54 So if you're doing web services, try out the free-threaded version.

12:58 At least one developer there is completely happy with what we have now.

13:02 Yeah, super nice.

13:03 A bit of real-time follow-up.

13:05 Yes, indeed. It is the same Giovanni.

13:08 All right.

13:08 Who creates a granion.

13:10 So with any of these web production app servers where you run your Python code when it's running,

13:16 whether it's Django, Flask, whatever, there's a process that runs your code.

13:19 Very often, the way that we've got them to work around the gill,

13:23 like in the web, the GIL is a much smaller problem.

13:26 It's still a problem, but it's less.

13:28 Because what we do often is we create a web garden.

13:33 So when you set up gran, you can say, and start four copies of yourself.

13:37 So that, yes, they all have the gil, but we'll round robin requests between the fours.

13:43 If you've got four cores, four CPUs, you basically can say, well, each one of these

13:47 is dedicated to a CPU and it can kind of like match, right?

13:51 Yeah, so we're running on good stuff.

13:53 But the thing is, when you have this free threaded option,

13:57 you can actually have true concurrency in your one worker.

14:00 So instead of scaling out four copies, you could have just one and just say,

14:04 let that one take 10 concurrent requests or whatever it needs to take.

14:08 - Yeah.

14:08 - Right, so that's how the free thread gets better.

14:10 And like, well, okay, why?

14:11 What's like six, one, half dozen, the other?

14:14 No, the memory becomes a problem when you create these little web gardens

14:17 because if normally your server would use half a gig and you create four, well now it's two gigs

14:23 and maybe that bumps you up in another tier and so on, right?

14:26 - Yeah, and if you need that memory for data, you don't have it, so.

14:31 - Yeah, so I think that's kind of one of the angles your writing concurrent code, but the foundations of your code running can do so more efficiently

14:39 because they're no longer needing to work around processes, limitations like, well, we got to

14:44 fan out these processes because of the kill. Okay, back to the regular scheduled programming,

14:49 free threaded Python, let's keep going. So here's an interesting article from a snake wearing a very

14:57 fast, it's a very fast snake with jet engines that is wearing a garbage collector. So what I

15:04 want to talk about is unlocking the performance in Python's free threaded future GC optimization.

15:09 So garbage collection is not something that we frequently talk about a ton in Python. You know,

15:15 it's just, we don't think a lot about GC. We don't think a lot about memory. I don't know why, like,

15:20 as you know, Brian, in C, C++, people are like always about it, right? Like, oh my gosh,

15:25 what size of point and what size of integer are we going to use for this, right? And like,

15:29 it's just way more sort of in the forefront, but it's nice every now and then to peel back the

15:34 and get a look at what's going on here.

15:36 So Neil Schemenauer wrote this article over here from QuantSight.

15:42 They do a bunch of data science primarily.

15:44 But here, check this out.

15:45 This is news to me, actually, even though I'm still interested in these things.

15:49 It says, first, the most important thing to know is that free-threaded Python uses a different garbage collector

15:54 than the default Python, the gil-enabled.

15:57 The gilded, I still want to call it the gilded Python.

15:59 The gilded Python.

16:01 So we have the default GC, which people probably understand pretty well.

16:05 When you create a number, like a thousand, when you create a list,

16:10 when you create an object from a class, all of those things have a data structure at the top of them

16:16 that allows them to manage and track the memory.

16:19 And people think of the GIL as being a threading thing.

16:22 The GIL is really a protection mechanism for memory management in Python, right?

16:27 It's basically protection against a race condition in the reference counting.

16:32 So basically interacting with this structure that has seven things reference me

16:37 and I reference it, however, you know what I mean?

16:39 Like that sort of deal.

16:41 - Yeah.

16:41 - And when the thing goes out of scope, it decrements that and that's where the GIL comes in

16:45 to like make that decrement safe and so on.

16:47 So that's the regular one, but we don't even have that structure anymore

16:52 in the free threaded one.

16:54 So what?

16:56 So instead, when you allocate stuff, it goes into a special memory managed heap called memalloc,

17:04 or managed by memalloc, right?

17:06 And this allows the GC to loop over all these objects

17:09 and figure out which ones are junk and which ones aren't.

17:13 A little bit like a mark and sweep garbage collector.

17:17 So there's a couple of interesting things.

17:20 One is that most of the mark and sweep garbage collectors

17:23 and that type of thing, they have generations, like a generational one.

17:28 So it has Gen 0s where most objects come, then it'll check that frequently and so on.

17:33 That's also how the garbage collector that looks for cycles

17:36 in the regular Python goes, right?

17:39 But it doesn't work here because of unmanaged C extensions

17:43 and a bunch of stuff like that.

17:45 So what it does is it can mark all the objects reachable

17:50 from what it calls known roots.

17:52 So instead of trying to scan the entire memory space,

17:54 it says, well, what are globals?

17:55 What are locals?

17:56 what are like a you know it sort of follow those and it marks those as active and that can they can

18:02 automatically be excluded from the search which is kind of like a generational optimization so

18:08 pretty interesting but here's the takeaway we were talking about making things faster by using free

18:13 threaded python with this one it's the free threaded gc collection is between two to 12 times faster

18:19 than the 3.13 version wow that's pretty yeah that's pretty wild right yeah yeah so depending on how

18:25 your algorithms work, do you have a super pointer,

18:27 pointer, heavy allocation, heavy type of algorithm,

18:30 then, you know, we'll probably see a bigger benefit

18:32 than if you allocate a few things and jam on those.

18:35 So anyway, if you want to peel back the covers

18:37 and see the GC, I didn't even know they were different,

18:39 but apparently they're different.

18:41 And presumably, maybe they could even have

18:43 like multi-threaded GC stuff going on.

18:46 That might be cool.

18:46 Or background thread GC.

18:48 I know those are a lot of different things that happen in some of these systems.

18:51 These garbage-coach systems.

18:52 Yeah.

18:53 And finally, A little bit of real time, not quite real time, but real world follow up, I'll say.

18:58 So for some reason, I can't remember what I was doing.

19:01 Something I had to do, I had to go back, go through all the examples of my async programming course at Talk Python.

19:07 So I went back in there and I said, all right, well, let me just make sure everything's running, update all the dependencies.

19:14 I can't remember what it was.

19:15 But anyway, something had to be updated.

19:16 So I made sure all the dependencies got updated and pinned them to newer ones.

19:20 And I'm like, well, there's a lot of changes here.

19:21 Let me make sure they're all running.

19:23 So go around and I ran every example from the course, probably 25 little apps.

19:28 And many of them are like, well, here's a web app, or here's just a way to do

19:33 multiprocessing, have a look at it.

19:33 But a lot of them were like, let's actually do this with threads.

19:37 And in this example with doing IO, you see it does make a difference.

19:40 In this example doing computational stuff, no difference or slower because there's

19:44 just overhead and it's still running one at a time, right?

19:47 Well, I decided to type uv-python 3.14t run those examples and run them free threaded.

19:56 The ones that used to have no benefit or a net negative are now like seven times faster.

20:03 It would go like synchronous one, 10.2 seconds.

20:08 Async or threaded one rather, 10.25 seconds.

20:13 And now just put the T on the end, three seconds.

20:18 that's pretty cool that is pretty cool man so to me it's really down to what are the framework

20:25 what are the packages and libraries that you're using if it's green across the board for free

20:30 threaded that's pretty interesting it opens up some real possibilities yeah also i i think i'll

20:37 be looking forward to the i know i don't know if we have a deadline or timeline for this but

20:41 when we can go back to having one version of Python and it's being the free-threaded one,

20:49 it'll be, you know, maybe we can switch it to be the default is the free-threaded one.

20:55 And for a couple of versions, there's the other one also.

20:57 You got to say 3.14G.

21:00 Yeah, or something like that.

21:02 And I think that would be good because the default way we think about how to program

21:08 and how to program quickly And the rules of thumb based on the GIL will be gone.

21:15 So we'll teach people different.

21:17 Yeah.

21:18 So anyway.

21:19 Yeah.

21:20 And before we move on real quick, a follow-up to your topic from chart here.

21:24 Also, if you're on a Kubernetes, if you're on Kubernetes,

21:27 it makes setting reasonable CPU and memory requests more difficult

21:31 to have to scale out like the WebGarden type stuff.

21:35 Oh, okay.

21:36 Yeah.

21:36 So if you can just have one process like this is the one,

21:39 I think it's easier.

21:40 I believe that's what he was getting at.

21:41 Yeah.

21:42 And also, I mean, from the C++ world, yes, processes and threads both can work.

21:49 But thinking in threads is easier than thinking about multiprocesses.

21:53 100%.

21:54 Because you have shared memory.

21:54 It just is.

21:55 You don't have to proxy stuff over and figure out how to sync it back up and copy it.

22:00 To queues and all sorts of...

22:02 Anyway.

22:03 Okay.

22:04 I can't resist.

22:05 I got to say, I got to put this out there in the world and see what you think.

22:09 One thing right now, we have the gilded Python and we have the free threaded Python.

22:13 And I can choose to run the free threaded one, as we've been talking about, if I want that.

22:16 But what about this, Brian?

22:17 This is what I want to put out in the world.

22:19 I want, like right now, if I have a project and I've got some data and I need to process

22:24 this in parallel on the gilded one, I need to maybe do multi-processing.

22:28 So that's going to spin up five little baby Python processes all to go jam on that, right?

22:34 That's how you get around the GIL.

22:35 What about this?

22:36 what if you could pass a flag to multi-processing and it would start up free-threaded Python with threads

22:43 instead of a bunch of different processes that can't coordinate on whatever algorithm

22:47 that you just say run multi-processing, but this time do it in free-threaded Python

22:53 as your little multi-process burst and then come back to me with the answer.

22:56 That would be cool.

22:57 - Yeah, yeah.

22:58 - You're not convinced?

22:59 I mean, you could that way, 'cause you weren't right. - I mean, your code

23:01 has to be different though.

23:02 - Just that one function though.

23:04 And you don't have to pay, you know what I mean?

23:05 like this one function that i'm going to call multi-processing on i'm going to write it assuming

23:09 that it can be it will be an advantage to be threaded or it will be an advantage to be concurrent

23:14 but the rest of my code i don't have to rewrite like it'd be a cool way to sort of bring kind of

23:19 like scython lets you bring and say this one function if just this were way faster it would

23:22 make all the difference you could do that but for free threading i would like to see that would be

23:26 interesting yeah yeah or just a chunk to say like yeah this um this subsystem is a through with

23:34 With free threading.

23:37 All right, go ahead.

23:38 Okay, completely different tangent is we're going to go back in time by a couple of one or two episodes.

23:45 We've been, the last few episodes or two or three talking about lazy imports because I thought they were coming in 315.

23:55 They're proposed in 315, but they're not here.

23:58 So, or they're not accepted yet.

24:00 I haven't checked.

24:01 I don't think that anything's changed.

24:04 So last episode, I did talk about lazy imports that you can use today.

24:10 Responding to that, Bob Bilderbos had a discussion on LinkedIn.

24:17 And part of that discussion, we had Will McCoogan hop in and said that he's excited about the PEP

24:26 and that he has a lazy loading mechanism for textuals widgets, which totally makes sense.

24:33 like widgets, you might just need a button, but you don't need all of the other widgets.

24:40 Be cool if you could just load what you needed.

24:42 And he's got that.

24:43 So I checked it out.

24:44 I was checking this out.

24:45 And I think this is sort of a small topic, but I think it's cool.

24:49 So the idea that I want to highlight is I presented last week a method for doing lazy

24:58 loading now or lazy importing now on things that you depend on.

25:02 But what if you're like, you don't have to wait.

25:04 You can be lazy today.

25:06 But what about like, yeah, exactly.

25:09 What about if you're the package?

25:10 If you have a big package that you have a bunch of stuff that people might want to import,

25:15 making it so that you're not the problem, that your package is going to import really fast

25:21 because behind the scenes, you're doing lazy importing for people.

25:25 And that's essentially what Textual does.

25:27 And the implementation is like really easy.

25:32 So he's using a, basically, when you access anything, he's overriding the get adder function.

25:40 So if you access a widget, it's going to try to just grab it out of a dictionary.

25:46 And of course, right away, it's going to fail.

25:50 And then if it fails, he goes ahead and imports it and then stores it in.

25:56 So the next time somebody tries to grab it, it's going to be there.

26:00 And so there's a little bit of misdirection here.

26:04 But in the end, you get lazy loading on every widget access.

26:11 So pretty cool.

26:13 And so I thought, you know, this is pretty neat.

26:16 I don't want it just to be hidden.

26:18 It's not hidden.

26:19 It's an open source project.

26:20 So I went ahead and wrote all this up in a new post called Polite Lazy Imports for Python Package Maintainers.

26:28 Very nice.

26:28 link to that also but i just thought that the implementation is totally clever and um and maybe

26:35 it's maybe for other package maintainers this is like well yeah that's the way you do it but um it

26:40 was new to me i thought it was neat so there we go also want to shout out i didn't realize there

26:45 was both get adder get adder is used for uh for if it's not found for missing items but also there's

26:53 another one called get attribute. And I did look up, I really appreciate that Trey Hunter had posted

27:00 this every Dunder method article. So pretty cool. I totally have this bookmarked so that I'm like,

27:08 if I want to understand the Dunder method in Python, I go here.

27:12 Yeah, very cool. I know it for classes, but for modules as well. Interesting.

27:18 Yeah, I don't think, and packages. So this is, you go ahead and throw it. Well, I guess,

27:23 It is for modules, but if you put that in a dendronit file,

27:28 then it's for your entire package.

27:30 Yeah, exactly.

27:31 It's pretty cool.

27:32 I love it.

27:33 Nice.

27:34 All right.

27:34 Is that all your extras?

27:36 Oh, that's just...

27:37 Your main thing.

27:38 That's my main thing.

27:39 What do you got for extras?

27:40 I do have one extra.

27:40 The one extra, so one of the things I've been doing is I've been,

27:47 you know, I paused Python people a long time ago, And I totally stopped doing testing code.

27:53 I'm focused on this too and focused on work, but I'm also writing more.

27:57 And one of the things I'm doing is writing a book on test driven development.

28:01 And I don't have an announcement yet, but in the next couple of weeks, I think I might

28:06 have an announcement.

28:07 But if you'd like to know what, I guess what I'm going to do is I'm going to write it.

28:11 I'm going to write the book, but to motivate me and to make people not have to wait.

28:16 I'm doing it as a, like a, you know, like Manning does like the early access books and

28:22 Pragmatic has beta books.

28:24 I'm kind of doing that.

28:25 I'm doing a lean, trying to do a lean startup applied to books.

28:29 And I'm going to release it after I've got, I'm like, you know, do a rough first draft,

28:34 clean it up a little bit and then release it for every chapter.

28:38 And then as I go along, incorporate feedback from people.

28:41 And then once the whole thing is good and polished, I might bring on an editor or something

28:46 or maybe just release it.

28:48 But if you'd like to know more about that, I'll announce it here.

28:52 But also, if you'd like to know the minute it's available,

28:56 join the newsletter over at Python Test, and I'll let you know.

28:59 Yeah, we really do appreciate when you join our newsletter.

29:01 You might think, oh, we can just announce these things

29:03 on social media or on the podcast, but it's not the same people.

29:07 A lot of people miss it because they miss an episode,

29:09 or social media is just a screaming feed of stuff.

29:12 It really makes a difference.

29:13 So join the newsletter.

29:15 Also, I want to talk about there's newsletters and then there's newsletters.

29:20 So this isn't really – I don't do a weekly eight tips on testing.

29:24 That would be cool.

29:25 I just don't have time to do that.

29:28 So it's mostly announcements if there's something to keep track of.

29:32 But people like you and me and others, we're using things like – I'm using Kit, but there's others.

29:39 I've used other email things before.

29:41 I don't keep track of who's subscribed.

29:43 So I'm not going to sell this to anybody or anything.

29:45 It's just an announcement thing.

29:47 And you can unsubscribe anytime.

29:49 And I don't even know.

29:50 Like once you're unsubscribed, you're gone.

29:52 I don't have the access to that.

29:53 So there's people that abuse newsletters and people that I don't think there's a lot of people in the Python tech space that do, though.

30:00 So it's not spammy.

30:02 Anyway, moving on.

30:03 Do you have any extras for us?

30:05 I actually have a couple.

30:07 Let's jump over and hear about it.

30:08 Very exciting news.

30:10 I am really happy with this new course that I created called

30:13 Agentic AI Programming for Python Devs and Data Scientists.

30:17 Cool.

30:18 Yeah, and the idea is basically, so we have these agentic AI tools

30:24 like Claude Code and Cursor and Juni and all of them,

30:28 but how do you actually be successful with them, right?

30:31 I've talked a few times about just insane productivity

30:34 and really good outcomes that I've had, and people are like, how are you doing that?

30:38 So this course is like a real world set of guidelines and examples and best practices

30:46 for working with agentic AI.

30:47 So build, spend an hour building like a really detailed application.

30:52 But we also, it also talks like almost for an hour about like guardrails and roadmaps

30:56 and how do you get it to do exactly what you want.

30:59 You're like, when I say build an app, don't just build the app, but build it with uv,

31:03 write high test tests, format it with rough, with this Tommel.

31:07 You don't have to do any, you don't have to, but you know, like how do you get it to give

31:10 you what you want?

31:11 Not something in the vague general space of what you asked for.

31:15 You know what I mean?

31:16 So sort of a practical agentic AI programming course.

31:20 So I really, I'm getting like crazy amounts of good feedback on this course.

31:24 So people check it out.

31:25 The link is in the, in the show notes.

31:27 I think it's talkpython.fm dash slash agentic dash AI.

31:30 So that one's fun.

31:31 What else?

31:31 I'm also going to be on talking with Hugo Bowen Anderson.

31:37 He's running the Vanishing Gradients podcast.

31:40 He was on Talk Python a while ago, and now I'm going to be on his show,

31:44 Data Science Meets Agentec AI.

31:47 Sort of a follow-up from my Talk Python in production book,

31:50 plus this course I just talked about.

31:52 Those are those kind of ideas.

31:53 We're going to spend some time riffing on that.

31:54 So that's tomorrow night US time.

31:57 So check that out.

31:58 All right. Also, OpenAI introduced an AI browser wrapping Google Chrome called Atlas.

32:05 This is interesting.

32:06 I played with it.

32:08 I'm not convinced yet.

32:09 I mean, it's kind of fine.

32:10 I don't know.

32:11 I don't even know what to think about these things.

32:12 That's not why I included it.

32:13 Maybe it's sort of interesting as a side note.

32:15 But I linked to the Ars Technica article.

32:18 Holy smokes, you guys.

32:19 Check out the comments.

32:21 They are not loving it.

32:23 And it's not OpenAI's fault exactly.

32:25 It's just this idea of AI taking over everything is not.

32:32 At least if there's not everyone hating on it, there's a very vocal group of people who don't love it.

32:37 So, and I just noticed that Kyle Orland wrote this and he actually wrote a whole book on Minesweeper,

32:42 which, you know, props to him for that.

32:43 But check out the- - Oh, that's awesome.

32:45 I gotta check that out.

32:47 - I'm like, geez, it's been a long time to thought about Minesweeper.

32:50 Anyway, this is actually just a really interesting cultural like touch point, I think.

32:54 So people can check that out.

32:55 All right, James Abel, one of the, I think he was behind Pi Bay this year, but yeah.

33:01 In the San Francisco area said, Hey, we somehow we were talking about dunder name.

33:07 If dunder name equals dunder main today, I have a package.

33:10 It's real, real simple.

33:11 But instead of writing this, if you want to like get people started, you can just

33:15 import my thing and say, if is main as a simpler, saner way, it should, that maybe

33:20 should be a built in, don't you think?

33:22 Python.

33:23 Yeah.

33:24 Or what I would rather see is if isn't main, here's what I would like to see, but

33:28 I don't know that Python has a mechanism for it.

33:30 I would like to say at the top, something to the effect of register this function as

33:36 me, and then when it's done parsing the file, then it runs it because even this,

33:42 which is super helpful, it's still, if you accidentally put code below it, it's a

33:46 problem, you know what I mean?

33:47 I would like Python to enforce, like we'd load the whole file and then it runs

33:50 like a cool mechanism for that.

33:51 But still, this is pretty nice.

33:53 And at first I thought I was just like wrapping if name is main, right?

33:57 But in fact, it's like doing some a little bit different here.

34:01 It's using the inspect stack to go back and then look at the value.

34:06 Because obviously, it itself is not main, so it can't do that test.

34:10 So I don't know, kind of cute.

34:12 Yeah.

34:12 If you don't want to have a dependency on this library,

34:16 you could take that line of code and put it somewhere as a utility class in yours.

34:20 But, okay, I'm just going to throw this out there.

34:22 If you want to do something really fast, make sure you time this.

34:26 Because every time I've added inspect, I love the inspect library.

34:30 But whenever I add it, it slows things down.

34:33 Okay.

34:35 Just benchmark.

34:38 Benchmark it.

34:38 All right.

34:39 Benchmark it.

34:40 All right.

34:41 Inline.

34:42 All right.

34:43 If you're an IntelliJ person like PyCharm, wouldn't it be nice, Brian?

34:47 Wouldn't it be incredible if while you're working, a little pet could walk around in the bottom?

34:53 Anthony Shaw, I believe, is the one who added this to VS Code.

34:57 Yeah.

34:57 PyCharm people can love animals too.

35:00 So you could install the pet into your PyCharm now.

35:03 Oh, cool.

35:03 Nice.

35:04 I don't know that I'm going to be doing it.

35:05 This is a last minute edition.

35:07 I just got a brand new M5 iPad.

35:10 How insane is that?

35:11 I've had the last one I had, I got five years ago.

35:13 I'm like, it's probably time.

35:14 And I got a 13-inch iPad Pro, which is an insanely large iPad.

35:19 But I started using it as the second monitor for my laptop.

35:23 if I'm traveling or if I'm at a coffee shop or something.

35:26 And that's a super cool experience.

35:27 Just put them side by side.

35:29 If they're on the same network or the cable, they just do real, real low latency, dual monitor stuff.

35:35 And yeah, it's pretty neat.

35:36 So just a little shout out to that.

35:37 That's incredible.

35:38 It's like having a new laptop, a second laptop for the price of a second laptop.

35:43 Yes, it is.

35:44 But here's the thing.

35:46 Yes, I agree.

35:47 But it's also a really nice reading device, which I do a lot of reading and stuff.

35:50 And I was going to get one of the cheaper ones.

35:53 One of the things that drives me, I have a MacBook Air as my main laptop computer and

35:57 the same chip for my main computer.

35:58 I don't have a high end computer and it's plenty good for Docker stuff or programming

36:04 stuff, all these things.

36:05 It's totally fine.

36:06 But here's the thing.

36:07 The iPad Air, while great, it's peak brightness is something like five or 600 nits.

36:13 And if you're working, like I like to work in my back outside area in the summer, like

36:17 that's kind of not relevant right now.

36:18 But in general, you know, sit outside, enjoy the weather, get out of the office.

36:22 And if it's at all bright through a window or somewhere, it's like really a pain.

36:26 This thing is like the best screen you can buy on a Mac, period, whatever.

36:30 And it's a thousand nits.

36:32 So you could push the computer to the side and just put the laptop in front of you and

36:36 type on it.

36:36 It's really nice.

36:37 Anyway, that's the main reason.

36:39 I want a brighter screen without having a MacBook Pro.

36:41 I feel so bad for you having to deal with like working outside in such bright light.

36:47 It's really horrible.

36:48 I know.

36:48 You should really.

36:50 It's hard.

36:50 It's hard doing me.

36:52 All right.

36:54 Carrying on with the jokes.

36:55 No, go ahead.

36:55 Well, before we get to the joke, I wanted to, I guess, highlight, going back to my announcement

37:01 of possibly a book.

37:05 Japanol7 says, a TDD book by Brian.

37:08 Can't wait.

37:08 Also, Talk Python training courses are great.

37:11 Kudos.

37:12 Yeah.

37:12 I'm awesome.

37:13 Yeah.

37:13 I'm looking forward to getting that book out, and I'm looking forward to that AI course

37:17 of yours.

37:18 Yeah, same.

37:20 All right.

37:21 I'm not looking forward to surgery, I'll tell you what.

37:23 And it's getting to be weird, Brian.

37:25 I mean, like doctors using AI and stuff.

37:28 Actually, we probably are going to get better diagnoses as for certain things.

37:32 Oh, dear.

37:33 But here's a surgeon situation.

37:35 There's a person who just, they're in post-op, okay?

37:38 They're laying there like, oh, man, a little woozy from the anesthesia coming out of it.

37:42 And the doctor, which is a robot with a ChatGPT-like, I don't think it's, is it the same?

37:48 know something an ai logo for a robot face and the patient says but why is the scar on the left

37:55 if the appendix is on the right the ai surgeon says you're absolutely right let me try that one

38:00 more time please don't try it one more time it's so bad it's pretty bad it's pretty funny that

38:10 actually drives me nuts when when i'm like this doesn't sound right is this you know

38:15 The user is pointing out I've made a mistake.

38:17 Oh, you're right.

38:18 Yeah.

38:19 Oh, well.

38:20 Yeah.

38:22 So, yeah, just get a second opinion.

38:25 So if opening is going to operate on you, have Anthropic be the backup, I guess, is the moral of the story.

38:31 I don't know.

38:33 I don't think that's the moral of the story.

38:34 You don't think so?

38:36 Okay.

38:36 Maybe somebody trained in medicine.

38:39 Trained on medicine?

38:40 I'm sure they are.

38:43 Trained with heavy medication.

38:45 Yeah, exactly.

38:47 All right, cool.

38:47 Well, fun as always.

38:49 Well, definitely fun as always.

38:50 And we'll see everybody next week.