All of our tools

00:00 This episode is a fun crossover from our Python news and tips podcast, Python Bytes.

00:05 We've had some big changes over there.

00:07 Brian Okken has moved on and Calvin Hendryx-Parker has joined the show as a new co-host.

00:13 To kick off this era, we decided to do a longer and more personal episode called All Our Tools.

00:19 The idea is both of us talk about some of our most useful day-to-day developer and business owner tools

00:25 that we think you all will find useful.

00:28 It was so well-received that I'm bringing to you as a crossover episode.

00:32 Enjoy, and we hope you find something new and awesome to help you with your software

00:37 and data science day-to-day.

00:39 And if you enjoyed this episode, please consider also subscribing over at Python Bytes.

00:43 We cover tips and tricks like these each week over there.

00:47 This is Talk Python To Me, episode 553, recorded as Python Bytes, episode 484 on June 16,

00:55 2026.

01:13 Welcome to Talk Python To Me, the number one Python podcast for developers and data scientists.

01:18 This is your host, Michael Kennedy. I'm a PSF fellow who's been coding for over 25 years.

01:24 Let's connect on social media.

01:25 You'll find me and Talk Python on Mastodon, BlueSky, and X.

01:29 The social links are all in your show notes.

01:31 You can find over 10 years of past episodes at talkpython.fm.

01:35 And if you want to be part of the show, you can join our recording live streams.

01:39 That's right.

01:39 We live stream the raw uncut version of each episode on YouTube.

01:43 Just visit talkpython.fm/youtube to see the schedule of upcoming events.

01:48 Be sure to subscribe there and press the bell so you'll get notified anytime we're recording.

01:52 This episode is sponsored by Sentry's Seer.

01:55 If you're tired of debugging in the dark, give Seer a try.

01:58 There are plenty of AI tools that help you write code,

02:00 but Sentry's Seer is built to help you fix it when it breaks.

02:04 Visit talkpython.fm/sentry and use the code talkpython26,

02:08 all one word, no spaces, for $100 in Sentry credits.

02:12 And it's brought to you by our Agentic AI Programming for Python course.

02:17 Learn to work with AI that actually understands your code base and build real features.

02:26 I want to just set the stage before I throw it to you, Calvin.

02:29 This episode is something that we came up with is like, you know, Calvin's joined the show.

02:35 It's kind of a slightly different take, you know, and let's just kick it off rather than

02:41 going around and find a bunch of other tools people are talking about.

02:43 Like, let's just do a check-in on the tools that we're using because we love them so much.

02:49 Like, that's why we're part of the show.

02:51 I know.

02:51 No. And it's just never ending.

02:53 I had trouble sleeping last night because I was trying to pick my favorite tools to put into the list.

02:58 Like it was running through my brain as I was falling asleep.

03:01 I honestly, I had a hard time too.

03:03 And it's like, well, which one of your kids do you really like?

03:06 It's so true though.

03:09 Well, we're going to find out.

03:11 This one also comes a little bit, you threw this out there,

03:14 but this also comes a little bit by special request from me because I'm like,

03:17 I've been hearing about this and I know that you do this, Calvin.

03:20 So tell me about Pi and the other things.

03:22 Yeah.

03:22 Okay.

03:23 So the first one I picked was a combination of two tools.

03:27 The first one being Pi, which is the orchestrating agentic harness.

03:32 And the other thing I compare with it is superpowers, which I'll talk about here in a minute.

03:37 But I'm definitely a terminal first person.

03:40 I always have been, even since college.

03:42 When I first got introduced to the terminal and typed vi tutor for the very first time, I was hooked.

03:48 So obviously that gives you a slant for my bias towards these things.

03:52 And I've tried IDEs now and then.

03:54 And I mean, they've gotten better.

03:56 They've gotten really good for doing agentic AI stuff.

03:59 I still will go fiddle with them and play.

04:01 And I've been using anti-gravity and codecs and like Claude's co-work, which gives you

04:07 kind of that UI wrapper around them.

04:09 But there's still just something about the terminal that really gets me excited.

04:13 And I don't know if it's just the rawness of it or the control or how I feel about it.

04:18 Maybe it's the minimalism of it, which if you know me,

04:21 I'm not necessarily a super minimalist person, but I'm opinionated.

04:24 And I think that the terminal gives me that power

04:26 to be very opinionated.

04:27 And so Pi is a terminal first open source coding agent.

04:32 So you can customize it and build it and make it yours.

04:35 I think that's what really drew me to Pi was I've been playing with codecs

04:40 and I've been playing with Claude Code and not to poo-poo some other people's favorite tools.

04:45 This one is just stripped back.

04:47 I think this is actually one of the reasons why I liked Ader chat when I first started using

04:51 some of these agentic tooling, was Ader again was kind of less opinionated

04:55 and bring your own pieces to it to make it the thing you exactly want to be.

05:00 Some nice things about Pi is that it's very session management

05:03 as a first class citizen here.

05:05 You can rewind and start from any place in that tree of your session as you're going through

05:10 and building the code.

05:11 So how many times have you gone down a rabbit hole

05:14 only to be like, ooh, that was not where I wanted

05:16 agent to land or the model to land on certain things you can rewind up just branch off of that

05:22 point and keep going the extensions model is incredibly uh composable it's almost aggressively

05:29 composable although i use it in a very minimal way it doesn't come with things like subagents out of

05:33 the box it doesn't come with issue trackers out of the box it doesn't come with much out of the box

05:37 which is why from a context window standpoint you build it and make it what you want it to be so i

05:43 use it with superpowers so combining superpowers which is a spec driven development uh tool so

05:49 you've probably heard i've talked about spec kit before on some of my linkedin videos i've actually

05:54 moved over to superpowers because i feel like it's a little more lightweight combined with pi it

05:58 basically detects your hey you're building some software i see you're building like clippy like

06:03 see you're building some software there uh let's uh take a step back and try and figure out exactly

06:09 what you're trying to do. So it's basically a workflow that runs you through brainstorming,

06:13 planning, execution, verification, review, and then committing and merging through that whole process.

06:21 And just it'll ask you questions, it'll propose options, and it seems that these things aren't

06:26 magic, they are just stacks of markdown skills on your file system. But I like the way it

06:31 interoperates at that kind of bare bones level with Pi. So where Pi, I feel like Pi, in the

06:36 agentic coding tool space. Pi is the one that hands you the keys to the car versus the guardrails

06:43 that are put in place by a lot of the other tools. So it may not be for everyone. It's definitely a

06:47 little more power user thing. And Michael, you've not tried it out yet at all? I have not tried

06:51 either of these. Oh, superpowers. Again, lightweight, fun. I tend to be, I try to be token efficient,

06:59 which again, the branching piece allows me to back up and not, I think people get stuck in a trap

07:04 sometimes they fall down a hole and just burn more tokens trying to get out uh and right right no

07:10 you're doing the wrong way keep no no i wanted you to do this right of just just just back up

07:15 rewinding time like that didn't happen time travel is amazing yeah so i'm a believer of skills and

07:21 certainly things like superpowers which my understanding is kind of like harness plus a

07:27 bunch of skills that you can use and so on and i think the reason i haven't tried i'm trying to

07:32 like why haven't i tried this yet well let me give you two reasons and i'll i'm interested to hear

07:36 what you think one there's just so many of these things that are like i have the magic that makes

07:43 ai work if you just use my seven markdown files it's going to be amazing you know what i mean and

07:48 there's like a million of them and i'm sure 95 of them are junk but this one i think is really like

07:55 go to the top of your page here how many stars does this have oh it's uh just a couple right yeah

08:00 Yeah, just a few, 229,000 stars on it.

08:04 So I feel like maybe, maybe this one actually might have a little momentum.

08:07 In polish, right?

08:08 Like, not just one person's opinion and they did something and it works,

08:12 but they could have done almost anything and it would have worked better, you know.

08:15 I highlighted right here, like this part's really nice.

08:18 Just out of the box, test-driven development, red-green,

08:21 and it shows you in the UI, like it's like tests are red,

08:24 you know, verifying, testing went green.

08:26 And then the fact is it thinks more like, I think, as a developer.

08:30 the whole Yagni and dry. A lot of the other coding agents, I believe are just make it go no matter

08:36 what, like put it in a loop and see how much stuff you can crank out. And I feel like this is a little

08:41 more just efficient in the way I would think about it myself. And obviously I'm reviewing the code as

08:45 it's going through. I'm actually building something with Pi right now in the background.

08:49 Nice. Okay. I love it. But I use it to pull off like quick little tools. Like if I've thought

08:56 about a tool, I'm just immediately go over to brainstorm and at least bookmark it into a small

09:00 little project so that I can start building a quick little tool. Right now I'm building a tool

09:04 that synchronizes my messages from my Mac or my phone to our CRM. Because I was like, I don't feel

09:10 like copy pasting anymore. That's awesome. Yeah, that is such a cool, that's a whole nother topic.

09:14 This hyper personal software stuff is super interesting. Like I'm not going to publish it,

09:19 but it's going to make my life better. Yeah. Yeah. So you can get Pi. There's a one liner here.

09:23 They've obviously got many different ways you can install it.

09:27 But I would check it out.

09:29 You need to be careful.

09:31 It is definitely the keys to the car.

09:32 You have full control.

09:33 You can build exactly what you want.

09:35 Nice.

09:35 So Pi is an agent harness.

09:37 A little bit like Claude code is to Claude itself, right?

09:41 Yeah.

09:41 And so I can run Claude in Pi?

09:44 You can run Claude code.

09:46 No.

09:47 I can use Claude Opus as one of my agents while I'm doing Pi, for example.

09:52 Yeah, yeah, yeah.

09:52 So Pi, you bring your own models.

09:54 You want to bring Gemini.

09:56 You want to bring OpenAI models, Anthropic models.

09:59 You want to use Ollama with your local models.

10:02 Feel free to pull out Gemma 4 and go to town.

10:06 I recommend LFM 2.5, the new liquid AI models.

10:11 If it's tool using, Pi will use them.

10:15 Awesome.

10:15 If Apple would just ship a new Mac Studio Max, I'll tell you what,

10:20 I would be running some larger models.

10:21 Well, the announcements from Apple last week, pretty exciting, right?

10:26 That's a whole other news item, but I think there's some exciting things coming at the Edge

10:30 when the Edge being your workstation.

10:33 Yeah, 100%, I totally agree.

10:35 It's just beginning, it's just beginning.

10:37 For so long, it's like, well, your computer's basically

10:40 just a view to a browser, which is a view to a server.

10:43 And it's kind of rolling back a little bit.

10:45 I'll talk more about that later in my follow on tools.

10:48 Okay, awesome, awesome.

10:50 Well, let's warp ahead a little because I want to talk about warp.

10:54 And I think there's a couple people out there who are just curious,

10:57 like trying to detect a theme or like, what do we find important as people have been doing this for a long time?

11:02 We both have AI topics and we both have terminal topics.

11:07 So I think that's noteworthy.

11:09 So my right now, for the last couple of years, I've been using warp from warp.dev.

11:15 And it's a really nice or fresh take on the terminal.

11:19 It's not just, oh, this one has GPU acceleration or whatever.

11:24 And I know this is a little bit different than your take,

11:26 which I'm also very excited for, Calvin.

11:27 But if you go look at it, it's like every piece of software these days.

11:31 Oh, you can ship software with an agent.

11:33 Like, I thought this was a terminal.

11:35 What in the world is going on here?

11:36 You know what I mean?

11:38 But it has some kind of agentic stuff.

11:41 I barely have used it before.

11:43 I mean, it's like, oh, this Linux command didn't work.

11:44 Like, why didn't this Linux command work?

11:46 Like, you forgot the dash F.

11:47 Oh, yeah, okay.

11:48 But I think there's better agentic tools than this warp for the terminal.

11:53 But you can just turn those off.

11:55 It's a really nice terminal with a lot of different angles on how you work.

11:59 It's got some team collaboration features.

12:02 It has really nice predictive capabilities.

12:06 So if you go to type something, if you type two letters,

12:09 it might suggest if you just right arrow, not tab complete,

12:13 because it's what you've recently done or something like that.

12:15 But just, hey, we think you might be doing this whole command here that we've seen you do sort of AI-like, sort of like an AI line completion, you know, like GitHub Copilot initially was, but for the terminal.

12:27 And that's super helpful.

12:29 Anyway, I just, I think it's a really nice.

12:30 I'm not sure if I'm ready for that yet.

12:32 I guess my level of control might not be ready to give up to write erroring over or even using the arrow keys at all.

12:39 I know.

12:41 Well, you know what's nice?

12:42 It has.

12:43 So this is part one.

12:45 We both have a couple of multi-part things like your Pi and superpowers.

12:49 So my part two is OhMyZShell, which is an old standard, but I love it so much.

12:54 So what I'm doing is I'm running warp, but then I turn off their terminal magic and just run OhMyZShell and do mostly OhMyZShell things.

13:04 So if people open up a terminal and it looks like the default thing that came out of their computer, it's probably really not a good thing.

13:13 And I've been a ZSH user for ages.

13:16 I can't tell you how excited I've been that this kind of has become more popular.

13:19 Yeah, yeah, it's definitely, I'm 100%.

13:22 I'm very, very excited about it as well.

13:24 And it's been just solid for years.

13:26 And if you open up your tournament on just, maybe it was white and you turned it black

13:30 because that was bad, but it's just the basic thing.

13:33 You know, for Linux or for Mac, and especially if using Command Prompt,

13:37 there's so much room for growth.

13:39 There's so much room for something better.

13:41 So you don't have to necessarily pick what we're picking here.

13:43 Just look around because the defaults are usually not good.

13:46 And so, for example, like why is OhMyZShell, even inside a warp with little like other stuff that it could do,

13:52 why is it cool?

13:52 So I could go to like a Git repository, CD in there, and it'll show me what branch I'm on.

13:57 Are there changes in that repo, right?

14:01 And it'll put that in your prompt.

14:03 It'll also show you if there's a virtual environment.

14:05 Is it activated?

14:06 What version of Python is it?

14:08 All those things are cool from OhMyZShell.

14:10 But then you could do things like get branch tab, and it will list all of the branches locally

14:16 and all the branches on your server that you could check out

14:19 or get checkout, whatever.

14:21 It'll understand way deeper into your projects and not just, say, autocomplete files.

14:27 So it's an opinionated layer on top of CSH.

14:30 I've got my own set of.files that I've carried with me

14:33 for the last 20 years, and so I've never gone full

14:36 oh my ZSH because I don't know how they would work together.

14:40 yeah that's definitely a challenge most of my stuff has grown up around oh my z shell oh my

14:46 zsh and so it's i've i've managed to incorporate but you have you see you're having the same problem

14:51 that i have with superpowers i have like so much structure around my ai stuff i'm like i don't know

14:56 how to make these exist coexist as well as they should you know yeah yeah but and i have an extra

15:01 i'll talk about that makes my zsh do most of that stuff just with one one little trick just let me

15:07 awesome let me just throw what was sort of a round out this part it's not even on my list

15:11 but i just now i feel like it's i should have it so pls i love pls i'm a big fan there we go

15:18 a prettier one and so a prettier ls and it's just so the uh website for it is not uh at least the

15:26 repo is not amazing but yeah like just ls show me what files have changed in git show me you know

15:32 what type of file they are with an icon as you would find in like finder or explorer or whatever

15:36 I think that one belongs in this, I'm going to now call it a trifecta of those three things.

15:41 But I thought about why these are important to us.

15:45 Like as developers, like these workstations, like our MacBook Pro or your Linux machine or Windows machine, whatever it is, it's like that is your life to making, you know, providing for your family or whatever you do.

15:59 I feel like tricking it out and making it your own, if it just provides a next, like all these things just basically layer on top of one another into building this kind of perfect experience for you as a developer.

16:10 And these little creature comforts like PLS or like I mentioned Starship later for my Powerline piece.

16:16 But it just makes the environment feel friendly and easy to use and just more seamless.

16:23 So I love bringing these things together.

16:25 And I don't want to shame anyone who doesn't do this, but why aren't you?

16:30 Like, it's here.

16:32 People have built these beautiful things.

16:34 Yeah, definitely not about shame, but I think it's just like, hey, if you're not doing this,

16:38 there's a great opportunity to make a whole lot of cool tools that start going for you.

16:41 Yeah, 100%.

16:42 Yep, definitely.

16:45 This portion of Talk Python To Me is brought to you by Sentry and Sear AI.

16:49 There are plenty of AI tools that help you write code.

16:52 but Sentry Seer is built to help you fix it when it breaks.

16:55 The difference is context.

16:57 Seer isn't just guessing based on syntax.

17:00 It's analyzing your actual Sentry data, your stack traces, logs, and failure patterns.

17:05 Because it has the full context, it can A, spot buggy code in review

17:09 and help prevent issues before they happen, and B, identify the root cause of production errors.

17:15 It can even draft a fix and hand the work off to an agent-like cursor to open a PR for you.

17:20 Seer turns Sentry into a complete loop.

17:23 You have your traces, errors, logs, and replays to see the problem,

17:26 and now AI to help solve it.

17:28 Join millions of devs at companies like Claude, Disney+, and even Talk Python,

17:33 who use Sentry to move faster.

17:34 Check them out at talkpython.fm/sentry and use code Talk Python26, all one word, for $100 in Sentry credits.

17:44 Thank you to Sentry for supporting Talk Python.

17:47 Like a cat? What is the deal with this?

17:49 Like a kitty?

17:50 Kitty. Oh, Kitty's a, so I'm a longtime Kitty terminal user. I've seen Ghosty and Cmux and all

17:57 these other new fancy, well, Warp kind of came in there as one people had said, well, why don't you

18:01 use Warp instead? I have really enjoyed Kitty as a terminal, mostly because I do move back and forth

18:08 between machines. Like I'm on a Mac right now, but I have a Linux framework desktop sitting over here

18:15 beside me, which is really powerful. And so I'm regularly moving back and forth between multiple

18:21 environments, multiple OSs. And some of the things I like to keep the same between and consistent

18:25 between them is like my terminal. So Kitty is a GPU accelerated terminal. You may ask why the heck

18:31 would you want a GPU accelerated terminal? It scrolls like nobody's business. And when you get

18:36 TMux running in there, and actually one of the nice things about Kitty, you don't even need TMux.

18:41 It actually comes with its own window muxing system built into it, which I've actually started

18:45 on my Mac. I may switch back over to Tmux because of some of my other opinions about how I want to

18:51 use that Linux machine for more things than just being available when I'm on the Linux machine.

18:57 But you get that ability to do the powerful layouts and tiling and scroll back buffers.

19:04 It's just a nice thing is it runs the same on Mac and Windows and Linux. So I install the same one

19:11 everywhere I land and I'm basically at home because I'm very terminal first about a lot of

19:16 this. So it gives me that performance and feel. Performance is a big feature. I think that tools

19:24 like Ruff and uv have proven to us that the fastest speed as a developer is staying out of

19:31 our way as we're thinking through problems and I don't want the terminal to be one of those laggy

19:36 bit. So for my Mac and for my Linux machine, I'm using Kitty Terminal. When I'm traveling on the

19:41 road, I use the Blink browser. I think I've got a Blink tab open here someplace. Did I open that up?

19:47 Nope, that's my Tmux tab. Where did I put all this stuff? I've got too many tabs open.

19:53 While you're looking, tell people what this Muxing is.

19:56 Yeah. So when you are in a, I like single windows too. So browsers that give me multiple profiles

20:03 across a single window, that's basically what I get with my terminal.

20:06 I can actually have one terminal window open, not a bazillion terminal windows

20:11 cluttering up my desktop.

20:12 And inside that one terminal window, I've basically got tabs like you would have in a browser.

20:15 And then those tabs can now be further split down

20:18 into tiles that I wanna be working on.

20:20 So if I'm in one terminal window and I've got mypy open on the bottom,

20:26 I've got a shell open on the top, maybe I've got broot open on the left side,

20:30 browsing the file structure.

20:33 I got lazy git auto refreshing in another one so I can see the file changes

20:38 as the agent is editing the files.

20:40 I can monitor actively what the diffs are all in a single terminal experience.

20:45 That's what muxing gives you.

20:47 And so Kitty gives it to you out of the box.

20:49 tmux gives it to you as an add-on to any system

20:52 you would be running on.

20:54 The benefit of tmux is you're gonna be able to leave,

20:58 detach from that session and reattach to it later.

21:00 And that's where this comes in.

21:02 For my iPad and my iOS experience, I use Blink Shell.

21:06 So Blink Shell is a fully power tool, power user version of a terminal on iOS and for the iPad.

21:14 It does give you a shell on the iPad, but more usefully, it's actually a SSH and Mosh client.

21:21 So Mosh is another piece that I use for this.

21:24 Mosh is the mobile shell that gives me the ability

21:27 to connect over unreliable connections, terrible Wi-Fi, putting your computer to sleep, traveling

21:33 to another network or coffee shop, opening the computer up, and it's like nothing happened.

21:38 Like you're still connected over your SSH session.

21:41 It just has adapted and figured out how to keep you connected and keep the bits flowing

21:46 back and forth between your terminal and where you're at.

21:49 So I can start on my Mac right here, have a session running on my framework desktop over

21:55 on my other desk over here and then go to a coffee shop and jump on my ipad and reattach into that

22:00 same tmux session and now i'm i'm using the same agents all the pieces that were working over there

22:05 we're all working uh just together for one another so that's the combination of like blink shell

22:11 kitty shell mosh and if i go into uh mosh let's see here the mosh mobile shell oh that one's over

22:18 in the other window of course there we go the mobile shell again that allows that whole roaming

22:24 internet connectivity if you're using ssh you should probably be using mosh you'll have a much

22:30 better experience it's much less fragile when it comes to keeping those connections going and then

22:35 tmux for attaching and reattaching and having multiple sessions so you could have if you've got

22:39 five clients you could have five tmux sessions running and you can just choose to attach to one

22:43 of them and jump back in exactly where you're at yeah it's it's like you're constructing a little

22:48 little ide source but a custom ide setup but out of terminal bits which it sounds like exactly we

22:54 want to live. That's cool. And I want to be able to travel with just my iPad. I don't like carrying

22:58 a large laptop around if I can avoid it. Yeah. This is a perennial desire that's always a bit

23:03 tricky to accomplish. And one thing about Blink that I didn't mention, it does come with a VS Code

23:09 embedded in it. So if you actually just type code from the Blink shell, it will fire up a VS Code

23:15 session there and allow you to connect on the iPad and allow you to connect to code spaces.

23:19 So you can actually use remote compute from GitHub over Blink if you go through all the hoops.

23:25 There are some hoops, but it does work.

23:27 I've done it.

23:28 That's awesome.

23:28 Wow.

23:29 Okay.

23:29 I have a quick, I have one more thing to talk about here on my terminal side is my app actually

23:34 that I built, Command Book, which I use this thing every single day, which is why I build it.

23:40 Because I'm like, why does this not exist?

23:41 Which lets you take things that are just long running.

23:44 If you just want to kick them off and get it out of your terminal.

23:46 So your terminal is like just for terminal things.

23:48 So like I want to have, I don't know, some Hugo watch my markdown files for any changes

23:54 and then auto reload that or like work on talk Python.

23:57 And if anything crashes about it, just restart it and keep it going.

24:00 Right.

24:01 Cause like you can have reload on flask or all the other web frameworks, but if you like

24:06 type a parenthesis and you haven't closed it and it reloads, it was like, can't parse as

24:10 it crashes and it never runs again until you figure that out and go back.

24:12 And so, you want to throw that out there as like kind of take the stuff that you're

24:17 not really doing terminal stuff.

24:18 It was just, that's how you run it.

24:19 Put it somewhere else and just let it be like baked in, which I love.

24:22 Okay.

24:22 So my main next thing to talk about, maybe boring, but just Claude Code.

24:27 What's that?

24:27 What's that?

24:28 Michael, I never heard of it.

24:31 There was like a movie that was like foreshadowing it.

24:35 It was made with like Arnold Schwarzenegger and these like machines made of something with Skynet.

24:43 No, I will tell you what.

24:45 I was very skeptical of the AI stuff for a long time.

24:48 when it was, oh, look, we can do, you can start writing a function. And then if you hit tab,

24:52 it'll write the next five lines. And I'm always like, you know what? No, I'll just get this away

24:57 from me. Because what would happen is like the first two lines, like, oh my God, that's exactly

25:00 what I want. But the third and fourth line are like, that's not what I want. How do I get the

25:04 first two lines and not the third and fourth one? Like, well, I guess I tab it and then delete them.

25:08 I feel like, I don't know, just a low level reviewer of like mistypings and misunderstandings

25:14 all day. I'm like, this is not how I want to think through my day. Like, this is not fun.

25:17 But once it got to the point where it was sort of agentic stuff,

25:20 iterating with tools, working with like super power type things

25:23 or specifications, and you're like, oh my gosh, this is such an amplifier of your skills.

25:29 I know there's a bunch of other stuff.

25:31 I've run local models as well, but when it comes down to stuff that really matters,

25:35 it's like, well, I was using Fable for a while.

25:37 We'll come back to that.

25:38 And you can use local models with Claude Code.

25:41 You can point it at Ollama, and you can use it with superpowers.

25:44 You can have that experience.

25:45 Okay, interesting.

25:47 But yeah, usually for me, it's just fire up Opus and really carefully work with Claude Code.

25:54 There's not a whole lot more to say, but if you're out there and you're a skeptic

25:58 because you tried it a year ago, not Claude Code necessarily,

26:00 but you tried this AI coding stuff a year ago and it wasn't that great

26:03 and you feel like it didn't work, give the tools a look again.

26:06 The top tier tools are really incredible.

26:09 And so it's super predictable that I would pick this,

26:11 I feel, but at the same time, how could this not be one of my tools?

26:14 I use this so much.

26:16 There's a lot to dive in here too, though.

26:17 I think model choice.

26:18 You mentioned just fire it up and point opus at it.

26:20 I feel like there's great ways to use the skills and sub-agents

26:24 to specify for this skill, you really only need haiku.

26:27 Like go fast, find a lot of things.

26:29 There's certain things that are definitely more adept to those other skills.

26:32 You don't need to burn a ton of tokens.

26:34 I have an awesome use of haiku you will not see coming.

26:36 Oh, yeah?

26:37 All right, I can't wait.

26:39 I can't wait.

26:40 One of the things that's really interesting that I've noticed

26:42 Claude Code started doing, and you can encourage it to do so if it doesn't,

26:45 is adversarial pushback.

26:49 So you talked about how you're absolutely right.

26:52 It's not a code review.

26:54 That's been one of the problems with these things, right?

26:57 You give it a goal, and it doesn't care if it generates a ton of code.

27:00 It's just going to jam through until it makes that goal work.

27:03 I need this UI.

27:04 Like, okay, well, why is it 1,000 lines when it could be 100?

27:07 And so on and so on.

27:08 But what I've noticed a lot of times lately is like, now Claude is starting to say,

27:12 okay, I launched 11 subagents to do this work, And then when it was done, I've launched four adversarial sub-agents in these different categories to push back.

27:20 And, oh, I found a bug or a misunderstanding.

27:22 We're going to fix that.

27:23 And it kind of starts to address some of this just over-eagerness to just blindly chase the goal.

27:30 And I think that's really good.

27:32 Yeah, the superpowers verification and review steps bring some of that to the table.

27:36 And you can install custom review agents that you can run in Claude Code to do exactly that.

27:41 Have it use another model from another provider

27:44 so it doesn't share some of the same opinions that were baked into the Anthropic models, for example.

27:49 Oh, yeah. Awesome.

27:51 That could be a whole show.

27:52 Maybe we do need to do the full agentic coding show.

27:56 Maybe, but take us to your next topic.

27:59 Okay. I like typing.

28:02 I feel like I've spent a lot of my life trying to type,

28:06 but I've really recently discovered I should just dictate instead.

28:10 and there's some really great tools out there for doing dictation to your computer.

28:13 Not only dictation, but transcription.

28:15 This one right here is MacWhisper.

28:18 I'm a fan.

28:18 It uses local models to actually do all the transcription.

28:22 So it'll download and use Whisper by default.

28:24 And I think if you opt for the Pro version, so there's a free version in the Pro version,

28:29 this is a Mac-only app.

28:30 I do have another option here for those who are not on Mac,

28:33 but you can use Whisper models like the Large V3,

28:37 or you can use Parakeet models, the latest NVIDIA voice models,

28:40 which are incredibly fast and incredibly capable.

28:43 So this one gives you transcription of, for example, you can just feed it a YouTube URL

28:49 and it will go transcribe and give you the transcription from that video.

28:54 I have it transcribed my meetings.

28:56 So one of the things MacWhisperer also does is it can intercept your local audio on your computer.

29:01 So I can record audio coming out of this browser

29:03 or out of my Zoom meeting or out of a Teams meeting

29:05 or you name five other things that you're probably meeting with,

29:08 and you want to grab those transcriptions all in a central spot,

29:11 you can do that with MacWhisperer, and it can do on-device speaker recognition.

29:16 So in addition to the local models for the voice transcription,

29:20 it can actually never send your data to the cloud,

29:23 unless you tell it to, for doing things like figuring out who's speaking.

29:26 And then it gives you a UI for finding all the conversations

29:29 I've ever had with Michael, for example.

29:30 It can do that, which is really handy, because I talked to Michael two weeks ago,

29:35 click on Michael's tag inside the software.

29:38 The GUI is pretty nice and it allows you for exploring

29:40 these transcriptions and dictations.

29:43 And then you can dictate into any field or any prompt.

29:45 So I can just be hovering over a prompt on my text box on my computer

29:50 and hit the push.

29:52 I use it in kind of push to talk mode.

29:53 So I've got my right option key on my keyboard mapped

29:57 so that when I hold that key down, it is listening.

30:00 And so I let up, it just pastes the text right into that field.

30:03 So any app, anywhere, terminal, browser, doesn't matter,

30:07 that all works really, really well.

30:09 The other bit on this one that I wanted to mention

30:13 is you can actually, with MacWhisperer, feed it prompts to post-process your text.

30:18 So if I'm dictating into my browser and maybe on Gmail,

30:22 I can give it an email skill or prompt that I want it to pass my text through

30:27 before it pastes it into Gmail.

30:29 So I can just go on about how I need to write Michael on email and tell him about this, this, and that. And it will format it into, hey, Michael,

30:36 wanted to let you know about this cool tool I found. Cheers, Calvin. Like it puts all the

30:40 wrappings around it and just pastes it in there. So that's Mac Whisperer. For those of you who are

30:45 not on a Mac and do appreciate full open source, which I do, there is Handy. This is basically the

30:50 dictation version of Mac Whisperer that is open source. It does do the Whisperer models and the

30:56 Parakeet models as well. And so if you are on Mac, Windows, or Linux, you can do the same thing with

31:02 Handy. So check out Handy. I'm not a user of Handy because I am mostly on the Mac when I'm doing my

31:08 dictation. And Mac Whisperer works on macOS and actually has an iOS version for iPad and phone,

31:14 but I don't find them as good as the built-in one on Apple's iOS devices. But I do find Mac Whisperer

31:20 to be better for dictation than the built-in one on the desktop. So it's a little bit of a different

31:26 on desktop. Yeah, amazing. Do you speak to your computer, Michael? Can you switch back to Mac

31:31 Whisperer? Let me tell you the way, let me recount the ways here. So absolutely, I've been a Mac

31:36 Whisperer user since day one. I was a dragon, naturally speaking, user from way back. And that

31:42 thing was janky, but it kind of worked. I've had two, one short-term-ish and one long-term reason

31:50 to use dictation. I broke my hand so badly, like just in three places all along the left side of

31:56 it so so bad that the cast didn't go to like where your fingers are like sticking out like this yeah

32:02 but where it like completely there was no fingers and it took six weeks to heal so i couldn't type

32:08 at all my left hand or my right hand wouldn't touch the keyboard because it was just a it was

32:13 just like a yeah covered in a cast that was the same reason for this guy to write handy he was in

32:17 a cast and couldn't type that's awesome yeah so i ended up that's i remember what i was using then

32:24 But basically, it was predates MacWhisperer and some of the Whisperer stuff.

32:29 And it was really frustrating, but it was necessary because at least it let me keep going,

32:34 you know, like answering emails and keeping the business going.

32:36 The other one is for a long time, for like 20 years, I've had RSI issues.

32:41 So much so that I've had to have surgery on my wrist to deal with carpal tunnel issues.

32:46 And that was super scary to kind of recover from that.

32:48 I had to just do a bunch of things like type a little bit less, get a ergonomic keyboard of some variety.

32:55 We need a keyboard episode next.

32:57 Oh, yeah, we do.

32:58 Yeah, we do.

32:59 I'm right-handed, but I force myself to be left-handed mousing because your right hand does all the arrow keys, page up, page down, and the mouse.

33:07 I'm like, I got to just turn that down and do all those things.

33:10 But to keep that at bay, and it's totally fine.

33:13 I can type like 10 hours a day, long as I'm careful.

33:16 But to just lessen that, I do a lot of dictation still.

33:20 And I use Mac Whisper, and it is so good.

33:23 You can grab like 100 MP3 files or MP4 files, throw them on there.

33:27 It'll batch generate transcripts or subtitles or whatever you want.

33:30 There's all these different formats, multiple formats for single output.

33:33 It is so good.

33:35 But the dictation is great.

33:36 You can dictate, like you said, to any text field.

33:38 But you can also just dictate to the clipboard if you're working in something that it won't

33:42 let Mac Whisper interact with.

33:44 But you can just say, all right, fine.

33:45 I'll just paste it if it won't go into that control or something,

33:47 which is really sweet as well.

33:49 And the last thing is I have set it up so that for transcription stuff,

33:55 it uses local models.

33:56 It uses the V3 Turbo that you pointed out.

33:59 But for dictation, I like a little bit of correction,

34:03 a little bit of more smarts.

34:05 So like if I say, you know, PyPI and this, and it'll like do it just right and not P-I-E dash or whatever,

34:12 I have set it up to use Claude Haiku.

34:15 Oh, that is a great usage of Haiku.

34:16 Yeah, because it's really fast.

34:18 So you're not waiting as you dictate.

34:20 It comes out like super, super fast.

34:22 But it's just, and it's not sharing other people's information.

34:25 It's just like when I'm talking to it, just flow that through Haiku and then into the text field.

34:28 And that is really something awesome.

34:31 So anyway, yes, 100% yes.

34:33 And another endorsement of this thing, it's not a subscription.

34:36 It's a one-time $30 fee.

34:38 Yeah, so I was hesitant to include some commercial software in here,

34:41 but I feel like this one's worthwhile because he has a free version. It's good to support him. I mean, I'm a huge fan of the local model

34:49 privacy first basis on this. You can pass it to local models as well. So if you've got Ollama

34:55 running on your Mac or wherever you got on your network, you can pass that off with prompts for

34:59 that too. I'm so much so a fan of the local audio processing. My home assistant now runs with local

35:06 audio processing on a Proxmox server down in the basement. So I've got a RTX 3070 card running

35:13 whisper v3 large that processes the voice that I utter in my house. That's awesome. That is so

35:18 awesome. So how about that for a segue to the next thing? Yeah, privacy. So if you've got this

35:24 machine running in your basement or downstairs or wherever, and it works great on your local

35:29 network, whether you're sitting on your couch on your laptop or sitting up here in your framework,

35:33 you go somewhere, work trip, you go to the coffee shop, client meeting. Normally, what do you do?

35:38 like, well, normally I would run this on AI, but I can't.

35:41 I'm sorry.

35:41 Or you expose it to the internet like an insane person,

35:44 and then it gets hacked, and then something bad goes down.

35:48 Or you could use my new religion, Tailscale.

35:52 Are you a Tailscale person?

35:54 I'm not.

35:55 I have all Ubiquiti equipment in the house, so I've got the Dream Machine SE downstairs,

35:59 and they have their Teleport VPN, which gets me exactly what I need from all my devices.

36:05 So that works for me.

36:07 But if I didn't have that, I would absolutely be using this.

36:10 Yeah, so I'm not a ubiquity person.

36:11 I have a super cool Wi-Fi setup.

36:14 Like, there's so many shows we could do.

36:17 But I don't open any of the ports, nothing like that.

36:20 It's like, let's not do any of those things.

36:22 And I was doing weird SSH tunnels and other stuff to kind of get safe access to this stuff,

36:27 or maybe using ngrok for a certain use case, rarely.

36:32 But tail scale, I believe it's called an overlay network,

36:34 where it's like you have your regular network, You're just doing whatever.

36:38 But then it's kind of like a VPN, but you don't, you're just kind of ambulantly there.

36:42 It doesn't intercept your normal traffic, but it just makes visibility to other subnets possible, right?

36:48 So I have my Mac Mini M4 Pro over there that's kind of maxed out, and it runs LM Studio with my models.

36:56 It runs my database server that I use for different apps that if I'm developing on one of them,

37:02 it needs to have access to its database instead of running that locally or even installing that at all

37:06 I just have it running over there.

37:09 Local network, it would connect to it.

37:10 But anywhere outside of there, now it will just see that server

37:14 no matter where you are in the world.

37:16 You need to jump in and do some screen sharing.

37:18 No problem.

37:18 Just jump back and either screen share on Mac or Windows app or remote desktop,

37:23 whatever the heck they call it these days, back.

37:26 And you have access to it, but you don't never have to expose that to the internet.

37:30 And that's the thing that's super, super cool.

37:32 Yeah, yeah, I love that.

37:34 That's why I use the teleport on the Ubiquiti stuff.

37:36 But if you aren't, most routers have full support for WireGuard at a bare minimum.

37:40 You should be VPNing back to your own house if you trust your own house, your home network.

37:45 Yeah, and if you don't trust your home network, well, you're already on.

37:48 Yeah, you have other problems.

37:50 It's pretty bad.

37:51 I mean, there's the whole internet of things, but maybe put that on a guest network or something.

37:54 Yeah, I mean, those are the right things to do.

37:57 They're just harder to do for most normal mortals.

38:00 But you can do really interesting production stuff too.

38:03 You could put tail scale on your server.

38:05 Once you get it set up, I've not been this brave.

38:07 I'll tell you what I actually do.

38:08 But you could set up Tailscale on your server, log into your server, and then block port 22 SSH.

38:15 Period, for the internet.

38:16 Tailscale has clients, are they the ones that have clients for like Apple TV?

38:20 You can make Apple TV exit nodes into networks.

38:23 I don't know about Apple TV, but they certainly have it for like iPhone and stuff.

38:26 Yeah, yeah, yeah.

38:26 I'm pretty sure, for example, if you're supporting aging parents in place

38:30 and you wanna try and help them support their network,

38:33 you can put this on an Apple TV So they get the benefit of home entertainment and secure remote access to help them out.

38:38 Yeah, exactly.

38:39 You don't have to have them do some crazy thing.

38:42 But yeah, so you can even like use this to connect your servers back to you securely without opening an SSH port on the internet.

38:48 And I have almost the same.

38:49 Like I'd almost, I'm not willing to, I'd almost be willing to tell you my SSH key to the server.

38:57 Because you also have to find a way into my home network in order to access it.

39:02 because it's not accessible anywhere.

39:04 Like port 22 on the server that runs all of our stuff,

39:09 it's only accessible from my home IP address.

39:12 Yep.

39:12 And no matter where I am with tail scale, I just say, use my Mac mini as the exit node.

39:17 And it looks like traffic is coming out of there.

39:18 That one is permitted to go into the server and nothing else.

39:21 So it's just another layer of security.

39:23 And it's such a sweet, and that's a real pain in the butt

39:26 if you're always moving around and your IP is stressed.

39:28 But because your tail scale just lets you like always go back to a stable machine,

39:32 then you're good to go. There's a lot of use cases for this. I interviewed Alex Kritschmar

39:38 who's done a ton of work, done a bunch of self-hosting stuff. He ran the self-hosting

39:44 podcast for a while, which is really awesome. But he's also now working for Tailscale doing

39:48 some of that. So we did a whole section on Tailscale with him and then it was really fun.

39:52 So people would check out that Talk Python episode. This portion of Talk Python To Me is brought to you

39:57 by us. I want to tell you about a course I put together that I'm really proud of,

40:03 Agentic AI Programming for Python Developers. I know a lot of you have tried AI coding tools and

40:09 come away thinking, well, this is more hassle than it's worth. And honestly, all the vibe coding hype

40:15 isn't helping. It's a smokescreen that hides what these tools can actually do. This course is about

40:21 agentic engineering, applying real software engineering practices with AI that understands

40:27 your entire code base, runs your tests, and builds complete features under your direction.

40:33 I've used these techniques to ship real production code across Talk Python, Python Bytes, and completely

40:38 new projects. I migrated an entire CSS framework on a production site with thousands of lines of

40:44 HTML in a few hours, twice. I shipped a new search feature with caching and async in under an hour.

40:51 I built a complete CLI tool for Talk Python from scratch, tested, documented, and published to PyPI in an afternoon.

41:00 Real projects, real production code, both Greenfield and legacy.

41:04 No toy demos, no fluff.

41:06 I'll show you the guardrails, the planning techniques, and the workflows that turn AI into a genuine engineering partner.

41:13 Check it out at talkpython.fm/agentic-engineering.

41:17 That's talkpython.fm/agentic-engineering.

41:20 The link is in your podcast player's show notes.

41:24 Essentially, you mentioned the Zeoxide in the comments there.

41:28 I use Zeoxide all the time.

41:30 If we're kind of jumping into extras here, that would be one worth mentioning.

41:33 Let's jump into extras, yeah, and just take it away.

41:36 Okay, well, I mentioned Zeoxide, but I also put a couple pieces in here that just didn't fit,

41:42 or they were, again, Mac only.

41:44 For viewing Markdown files, I want to, sometimes I want the most beautiful Mac experience I can get.

41:51 When things flow in that Mac world, and it's the old school Mac versus PC,

41:55 and I just want things to feel like my Mac, Telescopo is the Markdown viewer,

42:01 and now they've just added in editing.

42:03 So they've got the whole new studio piece, which is fine.

42:05 I don't actually use it for editing, but the viewer alone, when I first downloaded it,

42:09 it was only a viewer, and I was like, that's super cool.

42:12 Why would you want just an only viewer?

42:13 because sometimes you're presenting these Markdown files

42:16 to other people who don't just view Markdown in its beautiful marked up look and feel.

42:22 Telescopo gives you tons of themes and great fonts

42:25 and just that Mac-like feel.

42:27 It just feels normal to a Mac user.

42:30 And so you could probably turn regular old business users

42:33 on their Mac into Markdown Power users very easily by handing them this app.

42:37 I think that's definitely worth doing.

42:39 The other one I wanted to mention over here was Starship.

42:43 This is the cross shell prompt that I use to make it feel a little more fancy

42:47 when I'm actually in my various prompts.

42:50 If you've never checked it out, it gives you all the fancy coloring

42:54 and tells you when you're on master.

42:56 I will use it with the Groovebox theme.

42:59 I don't think they got any examples of the theme here,

43:01 but you can trick this thing out to no end, get the fancy power line looking thing.

43:05 The reason I landed on the Starship was because power line 10K became unmaintained.

43:12 And so I wanted a better alternative.

43:14 And so I went and found Starship, which is fast and rust-powered

43:18 and all the cool stuff.

43:20 And it has the whole nerd font.

43:21 It has to be installed so you get all the fun icons

43:24 and things like that in your terminal.

43:26 You have to have nerd fonts, by the way.

43:28 It just won't work.

43:29 It'll look rough if you don't.

43:31 But when you're switching between machines, you'll see it'll be like,

43:35 oh, you're on your Ubuntu machine.

43:36 So it gives you an Ubuntu logo.

43:38 You're on your Mac.

43:38 So you get an Apple logo in the prompt.

43:41 Again, those little niceties are real, real nice for having that going for you.

43:45 Yeah, the Starship is cool.

43:46 A lot of the stuff I said about Oh My Seashell also applies here.

43:50 Yeah, yep, yep, totally.

43:51 I don't know if I can show the presets.

43:54 It does say presets on the left there, I'm not sure.

43:56 Yeah, yeah, that's where you can come in here and you're like,

43:59 oh, I want it to look like this.

44:00 You just, one command away.

44:01 The one I chose was the Groovebox.

44:04 Like there's the Tokyo Tonight, which is pretty.

44:06 This is the one I've got running, which I really, really like.

44:09 And it's literally a one-liner.

44:10 You just copy-paste that, and now your prompt will look like this.

44:14 That's awesome.

44:15 Yeah.

44:15 What's the platform story for Starship?

44:19 Like it works on all of them?

44:20 Oh, it should work everywhere.

44:21 Any place you can put it.

44:22 I've got it on my Mac.

44:23 I've got it on my Linux machine.

44:25 So I want my experience to be seamless as I move back and forth.

44:28 And so I definitely dress up my Linux box to look a little more Mac-like.

44:33 You know, there's people who prefer Ubuntu because it feels like a Mac,

44:36 and people who like Ubuntu because it feels like Windows.

44:38 I'm on the Mac side of the UI spectrum.

44:41 Lovely.

44:42 All right.

44:42 Very cool.

44:43 Very cool.

44:44 Just a quick shout out.

44:46 Typeora is my markdown thing, but I will be checking out how this goes.

44:50 I evaluate Typeora.

44:52 I can't remember why I didn't pick it now.

44:54 Love it.

44:54 Yeah.

44:54 Very good.

44:55 It has a really nice just single hotkey to just swap between markdown view and WYSIWYG view.

45:02 But even in markdown view, it's still kind of like, you know,

45:04 the H2s are bigger than the H3s and so on.

45:07 I just, I don't know. It's just, it's pretty seamless. I like it a lot. But that's not the

45:11 main thing. That was more of a follow-up. Two extras to wrap up this long but awesome episode

45:18 is I recently interviewed Michael Ione and Michael Chow and Richard Ione from Posit About Great Docs.

45:27 Do you know Great Docs? Are you familiar with this? Yes. Yeah, we talked, well, you talked about an

45:31 extra last week. Yeah, that's right. Because I was talking about like the skills and the LLMs and all

45:35 that kind of stuff. And I'm like, I really have, I have a bunch of open source libraries, like seven

45:40 or eight that are worth like actually talking about, you know, like the one, if you use Umami,

45:44 I wrote that use, if you use, FastAPI, but you want to use the chameleon template language.

45:50 Like I wrote the FastAPI chameleon one. And if you want to use partials with Jinja for like

45:54 flask or cord or whatever, or even FastAPI, like I wrote that to be better to support HTTPX better

46:00 and so on. Right. But they, they only had their readme on GitHub. They didn't have a proper like

46:04 docs site. And after going through talking to them like this, I should really do this.

46:08 So part of this with Claude just helped me jam out like a bunch of stuff into my,

46:12 into my Hugo blog and connect all those things is I actually went through and I created

46:18 documentation pages, a site, full sites for every single one of those libraries last week.

46:23 And they're super cool. So like you over to Umami, it's got what you would see on the readme,

46:27 you know, all the cool stuff, but it also has like a built-in changelog that can be driven from

46:32 GitHub, it's got the API reference like, hey, what is this website ID?

46:36 Oh, show me the source and it'll take you over.

46:38 And it actually pulls up and highlights that function that was the documentation

46:43 on GitHub off of the documentation, which is sweet.

46:46 If you go back, it also has like LLM skills like, hey, here's the things you can do.

46:51 And here's how you authenticate.

46:52 And you actually pass this.

46:53 And here's the function.

46:54 And here's actually an example code that is a part of that.

46:59 And here's how, you know, the typing for this.

47:01 that will teach your agent about this library, which is one of the things I was really excited about, right?

47:07 So none of these libraries are new, but I created the doc sites for each one of them,

47:11 and they all are off of my personal mkiddy.codes.

47:14 Click on tools, click on open source libraries, and that shows you all the links to all the doc sites.

47:19 So anyway, I did that.

47:20 It's beautiful.

47:22 The frictionless way to view docs.

47:24 Yeah, I'm really excited.

47:25 Great Docs is pretty new.

47:27 It doesn't have a ton of GitHub stars.

47:28 I feel like it should have more.

47:29 But it's built on Quarto, which has been around for a long time.

47:32 It's super popular.

47:33 Okay, last AI statement here.

47:35 And I would love to hear your thoughts.

47:37 This is, I alluded to this earlier, and I said, I was using Fable.

47:41 Ha, ha, ha, ha.

47:42 That was fun.

47:44 Well, remember, on the show, we talked about Mythos before

47:48 and about how Mythos found a bunch of CVEs or what became CVEs in Firefox.

47:53 And it couldn't be released because it was so dangerous.

47:56 It was so powerful.

47:57 We had to give it to Apple, give it to Microsoft, and give it to these different companies so they could protect their code at least a little bit before it comes out, right?

48:05 Well, to my surprise, it came out last week.

48:07 Oh, my gosh.

48:09 Now we're turning this thing loose on stuff.

48:11 And it was in a pretty restricted mode.

48:13 It was available to Claude Code subscribers, but there was a little dagger subtext that said, this will be part of your subscription for two weeks, and then it will become a purely paid thing.

48:25 So it was going to go kind of unobtainium for most of us anyway, because I'm sure it was brutally expensive.

48:31 But it was still going to be there, right?

48:33 Well, the U.S. government said, you have to restrict access to mythos and fable to only U.S. citizens because this is too powerful.

48:43 And Anthropik said, we can't do that.

48:45 So we're just cutting it off.

48:45 Like, we're closing it down.

48:47 It's gone.

48:47 Good luck.

48:48 Good luck, everyone.

48:49 Thanks for being a very reasonable United States government.

48:51 I am so conflicted here.

48:53 In isolation, I think that the current administration of the U.S. has been pretty self-serving when it comes to a lot of things.

49:01 Like, we're going to, instead of just talking about the bad jobs report, maybe we're just going to, like, stop posting jobs data for the economy.

49:08 That helps us a lot.

49:10 And they got to fight with Anthropic, right?

49:12 Because they wouldn't let them use Claude for the weapons.

49:15 And the guy, Renzo Zagadario, is like, actually, you can, just not yet.

49:19 It's not ready for weapons.

49:20 Like, you're going to go, this is crazy, my understanding.

49:24 and they said fine well then you're barred from all like government use and so there's been this

49:27 feud so this could be the u.s government being spiteful i totally grant that yeah could also be

49:33 though Anthropic has spent months telling us this is nearly like weapon of mass destruction type of

49:39 llm stuff if it gets out it's going to destroy the world we can't even we got to do project

49:43 glass wing wag do all these things to make sure that at least when it comes out it's somewhat

49:47 protected like we got to set up the iron dome equivalent and then they release it and they're

49:51 like well this is too dangerous you know i mean like they both are kind of if you go around screaming

49:56 this thing is so dangerous it's going to destroy the world because it's so powerful and then like

49:59 well people are upset it's dangerous it's like who's yeah but you kind of marketed that yeah

50:03 mike who's raising money michael who's raising money i know yeah exactly then ipo is i think i

50:08 think this is a who's raising money moment they needed this kind of moment to be seen as this is

50:14 the most important model of your life kind of thing well then next month the next most important

50:19 model my life will be out too so i don't know i i'm skeptical of these models they are powerful

50:26 they can do a lot uh they probably should be in the hands of people who can actually make the

50:30 best use of them to go protect their software to patch all the vulnerabilities that exist because

50:36 the open waste models are only a few months behind these things so yeah yeah that's true like the

50:40 genie's out of the bottle we need to give access to the right people so they they can get their work

50:45 down and continue building safely with these kinds of tools.

50:50 And so the people just have Fable still active.

50:52 I got to use Fable exactly once.

50:55 Oh, really?

50:56 I was on the road last week when it all happened and someone's like, did you try Fable?

51:00 Yeah.

51:00 I'm like, oh, I better go try Fable.

51:02 So I had it right and owed to Taco Bell.

51:05 That's my thing.

51:06 I make these.

51:07 I had Fable do one thing.

51:08 Oh, the gordita is powerful.

51:10 Oh, it was good.

51:12 Oh my gosh.

51:12 I used it quite a bit for three or four days.

51:15 I actually blew through my weekly limit.

51:17 I'm like, oh, man.

51:17 Oh, yeah.

51:18 These things are token monsters for sure.

51:20 Yeah.

51:20 And then it wasn't available in my tokens reset.

51:23 Yeah.

51:24 I think we're going to see a new era of right model for the right job.

51:27 A lot of small language models and a lot of machine learning models that turn into tools

51:32 will probably be faster and cheaper to use than things like Fable and Mythos.

51:35 Yeah.

51:35 100%.

51:36 100%.

51:37 And shout out to Mark Little, who's been on the show before, who sent this over when this

51:41 first came out.

51:42 I said, hey, did you notice that Fable's gone?

51:44 I'm like, what?

51:44 where did it go i know yeah we don't know where it went but it's not there all right so ready for

51:49 the joke i am ready for the joke but let me set the stage before we go into this really quick okay

51:53 i think you should be kind to people and understanding and like kind of chill but every

51:57 now and then something just catches it just it it hits a nerve and you just like nope nope this has

52:03 to be right this and this is sort of this person here is like i titled this no second date

52:10 there's this sort of smug looking person there won't be a second date but at least she knows

52:15 that an array index starts at zero not one it's so bad you could just imagine the type of person

52:22 that's like you know what how'd the date go you know what no we couldn't do it yeah it would just

52:27 not something initially want a date but it's pretty funny i thought this is why i try to teach my

52:31 children there's a difference between being right and being correct yep you don't always want to be

52:37 correct and even you don't have to publicly be right to their face right like just you know get

52:43 along with people right but anyway it's i thought this is pretty funny you do the right thing not

52:48 always the correct thing yes exactly exactly well this person's getting no second date nope not

52:53 in this one like i can hear that conversation go down yes i've had that conversation i'm sure

53:00 and it doesn't have to even be a date it just could be like oh i met some people at the conference and

53:04 Oh, now I got to get out of this conversation as quick as I can.

53:07 Yep, yep, yep, yep.

53:08 All right.

53:08 Well, I don't feel that way about this show, Calvin, but I do believe it's time for us to go.

53:13 Yeah, I agree.

53:15 This has gone on.

53:16 We get passionate about our tools.

53:17 I'm excited that I'm here for it.

53:19 I'm here for it too.

53:20 Thanks for listening to the end, everyone.

53:22 And hopefully you'll all enjoy this unique episode.

53:24 Yeah.

53:25 See ya.

53:25 Talk to y'all later.

53:26 This has been another episode of Talk Python To Me.

53:29 Thank you to our sponsors.

53:30 Be sure to check out what they're offering.

53:31 It really helps support the show.

53:33 This episode is sponsored by Sentry's Seer.

53:36 If you're tired of debugging in the dark, give Seer a try.

53:39 There are plenty of AI tools that help you write code,

53:42 but Sentry's Seer is built to help you fix it when it breaks.

53:45 Visit talkpython.fm/sentry and use the code talkpython26,

53:50 all one word, no spaces, for $100 in Sentry credits.

53:54 And it's brought to you by our Agentic AI Programming for Python course.

53:58 Learn to work with AI that actually understands your code base

54:02 and build real features.

54:04 Visit talkpython.fm/agentic-ai.

54:08 If you or your team needs to learn Python, we have over 270 hours of beginner and advanced courses

54:13 on topics ranging from complete beginners to async code,

54:17 Flask, Django, HTMX, and even LLMs.

54:20 Best of all, there's no subscription in sight.

54:23 Browse the catalog at talkpython.fm.

54:25 And if you're not already subscribed to the show on your favorite podcast player, what are you waiting for?

54:31 Just search for Python in your podcast player.

54:33 we should be right at the top.

54:34 If you enjoyed that geeky rap song, you can download the full track.

54:37 The link is actually in your podcast blur show notes.

54:40 This is your host, Michael Kennedy.

54:41 Thank you so much for listening.

54:43 I really appreciate it.

54:44 I'll see you next time.

54:59 We'll be right back.