Episode 89 – I Don’t Have That Many Tendons In My Brain
Scott: Friends with Brews.
Peter: Friends with Brews.
Scott: I’d like to move us right along to a Peter Nikolaidis.
Scott: Peter Nikolaidis, how are you?
Peter: I’m not only A, I’m the Peter Nikolaidis.
Scott: Now you’re quiet again.
Peter: I blame Riverside.
Peter: Hashtag blame Riverside.
Scott: Hashtag Riverside.
Scott: Death to Riverside.
Peter: Yeah, well, that’s maybe a little extreme.
Scott: Was it Riverside, a town in California where the Zodiac killer did some of his stuff, was in Riverside?
Scott: I’m pretty sure.
Peter: I’m sure there’s a place called Riverside, but I don’t remember the Zodiac killings.
Peter: It wasn’t, I don’t think I was around for that.
Scott: I certainly lived in Riverside.
Scott: We lived in California during that time.
Scott: I was a baby, you know, I was tiny.
Scott: I was like one or a baby or something.
Scott: But yeah, we lived down there at that time.
Peter: Well, awesome.
Scott: So anyway, Peter, I don’t know if you know this, but I’m going to have a hard time typing the letter I for a while now.
Peter: I do know this.
Scott: Oh, how did you know this?
Peter: Because you had an encounter with gravity, a gravity induced, I should say, a gravity induced encounter with a solid object today.
Peter: Is that not the case?
Scott: Yeah, but here’s the funny part about it though.
Scott: I don’t think I actually hurt my finger when I tripped.
Scott: I think I hurt it when I was trying madly to wipe up the coffee.
Scott: And I think when I was shoving the towel forward across the carpet, I think I jammed my finger then.
Scott: Here’s the weird thing, people.
Scott: Today, as I was coming into the computer office to podcast, and I was carrying a coffee in one hand and a water in the other, and I was watching those carefully so as not to spill, I tripped over something that’s not normally in our hall because my daughter’s room was being cleaned out at the time.
Scott: And I was an idiot, and I tripped.
Scott: I mean, I knew stuff was there.
Scott: It’s not like I’m a c-
Scott: but yet I still moronically tripped over it.
Scott: So I trip and fall, I spill, and then I’m pretty sure I didn’t actually put my hand-
Scott: now that I think about it, I don’t think I even put my hands down in the fall.
Scott: Then I was mad.
Scott: I was just raging mad.
Scott: And I went and got some towels and I started furiously scrubbing the carpet because I wanted to get the coffee out of it.
Scott: And I think I bent my finger down at that point.
Scott: I think when I shoved it forward, my finger caught.
Scott: And basically what happened is in my middle finger on my right hand, the distal knuckle tendon apparently ripped.
Scott: So it’s not dislocated.
Scott: It’s not even broken.
Scott: There’s just a ripped tendon.
Scott: So it doesn’t hurt super bad at all.
Scott: I was spared a lot of pain, but the end of the finger droops down.
Scott: Now I can hold it up straight without any pain.
Scott: So now it’s in a splint, of course.
Scott: And I got to go have surgery on it and get the tendon repaired, which is fine.
Scott: And I have to say, Peter, listen, here’s the thing.
Scott: Today was one of those days in the Willsey household where everybody woke up grumpy and things went downhill from there because my daughter’s room had to have an emergency cleaning.
Scott: My wife wasn’t happy about it because she already knew that she had to get ready for work in a little while and go work in the hospital until midnight.
Scott: So that was already something that she had on her day.
Scott: And I didn’t know what was going on because I was working.
Scott: I was with you talking about stuff that, you know, asking you questions about different things and trying to apply it to some computer work I was doing.
Scott: And then, you know, we decided to podcast.
Scott: And so I was annoyed with all the, you know, I was unjustly super annoyed with everyone around me for making a lot of noise.
Scott: And I didn’t know what the commotion was about.
Scott: I didn’t know why they brought this spring cleaning on themselves, which turns out they didn’t.
Scott: And it was actually a wake up call.
Scott: I actually, when I did this to my finger and I realized I’m going to have to go to the ER, I immediately calmed down.
Scott: I immediately was like, you know what, I am being a dick.
Scott: I am overreacting to everything today.
Scott: And it is not fair to the people around me.
Scott: And it’s not pleasant.
Scott: And I need to apologize to people.
Scott: And after that, I’ve been in a great mood all day, even though I now have a splint on my hand and my tendon is ripped.
Scott: I am a far better person now than I was this morning.
Scott: I will say that for sure.
Peter: You did seem to take it really well.
Peter: And I’ve definitely been watching you get really worked up at times over things that I wish that you wouldn’t.
Peter: And you took this really well.
Peter: So yeah, I think this does seem to have been a wake up call.
Peter: So good.
Peter: Good morning.
Peter: I’m glad you’re awake.
Peter: Good morning.
Peter: That was him.
Scott: All right.
Scott: So let’s talk about drinks, Peter, before we get into how your day is gone.
Peter: Yeah.
Peter: Well, we are.
Peter: We’re getting a little bit of a late start on this one for reasons that, yeah, you know, emergency room trips have that tendency.
Peter: I’m frankly surprised.
Peter: I didn’t think we were going to record today.
Scott: I didn’t either.
Peter: You know, I was pretty sure when I heard you like broken or whatever your finger that like I was like, OK, that’s not happening.
Peter: But here we are.
Peter: So even though it’s after 8 o’clock, it’s like 830 Eastern Time, my time.
Peter: And I generally don’t drink this time of night, like anything, not just alcohol or whatnot.
Peter: But to commemorate this fact that we are recording, I did make myself a single shot of decaf espresso made from the Wegmans decaf espresso roast.
Peter: So that’s what I’ve got.
Peter: Yeah.
Peter: So it is just a shot.
Peter: I’ve got this in this little shot style, you know.
Peter: And I did just make a fresh batch of simple syrup today.
Peter: So this does have one teaspoon of simple syrup.
Peter: So it’s like almost a Cubano style.
Scott: Tip that towards me a little bit more if you can without spilling, because I feel like this is one of those movie actor things where you’re holding a cup that’s empty.
Peter: No, there really is a beer.
Peter: And I don’t think I can switch to desk view in Riverside.
Peter: Otherwise, I would show you that it’s sitting on my heating, like my coffee warmer at 176 degrees.
Peter: And if this is empty, I’ve done that before once when I had an empty cup on my coffee warmer and it turned to like plastic or something.
Scott: It turned back to the clay that it was originally made of.
Peter: Exactly.
Scott: Yeah, the funny thing is sometimes, have you ever noticed that sometimes in movies and TV shows, when they’re drinking out of an empty thing, you can tell, especially if it’s like a stupid Starbucks paper cup or something, the weight is totally different.
Scott: Even if they try, there’s something about the way.
Scott: Yeah, you can’t do it.
Scott: Apparently, you physically can’t do it.
Scott: So anyway, it’s funny.
Scott: All right.
Scott: So how’s your day going?
Peter: I mean, so far it was pretty good today.
Peter: I pretty much had a day off.
Peter: It was a rest day.
Peter: Work was very quiet.
Peter: I had like three meetings, three calls today.
Scott: Oh, you didn’t go running today.
Peter: No, today was a rest day.
Peter: So I only walked about six and a half miles today.
Scott: Oh, well, that’s a rest.
Peter: Yeah, I wanted to go kayaking, but it would have been perfect, beautiful kayaking weather.
Peter: I just…
Peter: So I live literal, not a figurative.
Peter: I live a literal stone’s throw from the Mystic River in Medford, Massachusetts.
Peter: And I’ve been spoiled because I have a kayak that I can just inflate right on my back deck and carry it and set it right in the river.
Peter: But ever since they did the reconstruction in the bike path, there’s now all these gates and barriers between me and the water.
Peter: So like all the other schlubs, I have to now put my kayak in the car and take it to a dock somewhere.
Peter: And frankly, I was just too lazy to do that.
Peter: So instead, I just walked six and a half miles.
Scott: Okay.
Scott: So Peter, I am having, I’m having, I have to find my notes for this.
Scott: I’m having a new one.
Scott: This is one that I found at New Seasons.
Scott: It’s Silver Falls Coffee Company, which Silver Falls should be near us.
Scott: So let’s see if this is indeed the Silver Falls near us, because there’s Silver Falls, Oregon.
Scott: I don’t know, they don’t have anything about us.
Scott: Ah ha, we are family owned, and we don’t say, like, about which Silver Falls it is.
Scott: But I’m guessing it’s Silver Falls, Oregon, just because why not?
Scott: Yeah, so anyway, Silver Falls Coffee, it’s their Cascade Blend, oh, of course, Cascade Blend, jeez.
Scott: Cascade Blend Decaf Swiss Water Process.
Scott: And it’s a darker bean than I normally like, but it’s not bad.
Scott: It’s pretty good, a decaf.
Scott: Let me taste it for you.
Scott: Yeah, yeah, it’s not bad.
Scott: I give it a go.
Scott: It’s the thing about this is you get a giant bag for the same $16 you get the normal size bags from the other people with.
Scott: So you do get more beans for your $16.
Peter: Okay.
Scott: And since it’s not horrible, and I get more beans for my buck, I’ll take it.
Scott: So thumbs up to more beans for your buck from Silver Falls.
Peter: And up yours too.
Scott: Yeah, I agree.
Peter: Yeah, there you go.
Peter: Okay, cool.
Peter: Sorry, I got distracted.
Peter: I wanted to…
Peter: So we both bought Mac minis on Prime Day last month.
Scott: We did.
Peter: We did.
Peter: Or Mac mini M1s.
Peter: So the original M1 processor, which is still great little utilities.
Peter: I use one as…
Scott: It’s still a processor in their lineup.
Peter: It’s still a processor.
Peter: It’s still decent.
Peter: Although the last time we recorded an episode, I was on one and we were using the same technology we are now using the Riverside platform.
Peter: And it was definitely choppy on my end, the video and sometimes your audio was choppy.
Peter: Now, you didn’t notice that there was no recording problems as far as I know, because I’m sure you would have told me loudly and clearly if there was.
Peter: But the user experience was a little choppy.
Scott: John Siracusa would say, I didn’t yell at you, it was a gentle correction.
Scott: Okay, yeah, sure.
Peter: Anyway, so I have remote control screen sharing enabled on both of these boxes, and one of them is in Vermont.
Peter: And it was on and I was connected to it, and I was playing around with ICE, which is the alternative to bartender or hidden bar, which I’ve been using.
Peter: And I also had a really slow connection.
Peter: And, you know, VNC, I’m kind of really kind of surprised that that’s what Apple standardized on for, because as far as I’m concerned, it’s kind of the worst of the remote desktop solutions.
Scott: Yeah, but you also have to remember, they’re running a Unix platform.
Peter: Yeah.
Scott: And tell me what screen sharing solutions people use on Unix platforms.
Scott: VNC.
Peter: I know.
Scott: 100%.
Peter: But this is Apple, dude.
Peter: They usually do things.
Peter: Oh, God.
Peter: They used to do things better, you know?
Peter: Like they didn’t adopt the Linux file system, right?
Peter: There’s a lot of stuff that they, you know, overlaid their own stuff.
Peter: And remote, you know, like their whole remote administration, the commercial product, that wasn’t based on VNC, was it?
Scott: I think it is.
Scott: It’s also $80.
Scott: Let’s see here.
Peter: So I’m thinking, maybe it’s the same.
Peter: I don’t know.
Peter: I’m out of date.
Peter: But the one that you used to purchase, like for, you know, back when Apple was under the delusions of becoming an enterprise.
Scott: Yeah, Apple Remote Desktop.
Scott: I’m pretty sure it’s a VNC based client.
Peter: Is it?
Peter: Okay.
Peter: Well, I don’t know.
Peter: I’m using the real VNC and it’s a crappy experience, even though I have it on auto or lowest video quality.
Peter: You know, like, just spare me the bandwidth.
Peter: Just get me a connection.
Peter: So you can see it like pixelated rendering the screen very, very slowly.
Peter: And so I clicked and I thought I was clicking to close something.
Peter: And what that something was was tail scale.
Peter: So I quit tail scale.
Peter: So now I don’t have remote access to it.
Peter: But here’s the thing.
Peter: I’ve got remote access to another Linux box on the same network in my place in Vermont.
Peter: But I can’t SSH to the Mac.
Peter: So I don’t know if it went to sleep after tail scale turned off or what.
Peter: I thought I had it set to stay on altogether, but I can’t reach it anymore.
Peter: So not going to use that one as an exit node anytime soon.
Scott: Peter killed his exit node.
Peter: Derp.
Scott: So I don’t think it would…
Scott: If speed is an issue, this probably wouldn’t help.
Scott: But in terms of VNC clients, Screens 5 is a good VNC client.
Scott: You’re not going to get an RDS type experience.
Scott: Let’s face it, RDS is pretty good.
Peter: Yeah, that’s something that Microsoft…
Peter: Well, I was going to say Microsoft got right, because they stole it from Citrix.
Scott: Right, so that experience is really good.
Scott: When you’re used to administering Windows to Fulte, and you’re RDSing in, and then you have to use VNC, yes, it is 100% painful.
Scott: No doubt about it.
Scott: But if you have to use something that’s based on VNC, the Screens 5 client is really good.
Scott: I use that for my Mac Mini, and it works great through TailScale.
Scott: It even supports TailScale.
Scott: It’ll show you a TailScale view.
Scott: So when you have TailScale running, it’ll show you which ones are your TailScale clients that you can access and stuff like that.
Peter: Cool.
Scott: It does cost money.
Scott: I don’t remember if it’s a one-time fee or a subscription.
Scott: I think it’s a sub, but just try it and see if you like it, and if you decide you’ll use it enough, it’s probably worth it.
Scott: It’s worth it for me.
Scott: Let’s put it that way.
Peter: All right.
Peter: If I can reconnect and get the thing running again, so without driving to Vermont.
Peter: And since I’m not going up this week and since I’ll be heading to Europe pretty soon after that, I don’t know when that’s going to happen exactly.
Scott: So you don’t know when that’s going to happen?
Peter: When I’m going to get back to Vermont.
Scott: Oh, right.
Scott: What are the odds of you?
Scott: Oh, did I?
Scott: Yeah, I did.
Scott: What are the odds of you going to Germany and never coming back?
Scott: Zero?
Peter: Never coming back?
Peter: Slim.
Scott: The odds are the odds 10, Peter?
Peter: I wouldn’t say the odds are zero.
Peter: I mean, but I am planning.
Peter: I mean, I did buy a return trip ticket.
Scott: Okay, you’re going to hear a can opening.
Scott: It’s not any kind of alcohol.
Scott: I’m having a polar seltzer water.
Scott: It’s just a carbonated water.
Scott: It’s hot in this room, so I’m going to have this too, because I’ve been drinking hot coffee in case you didn’t notice.
Scott: There we go.
Scott: All right, Peter, what are we here to talk about today?
Scott: You had topics, you had things, you said things, you did things.
Peter: You’ve been doing a lot of AI-assisted coding lately, and you’ve been doing tons of stuff from the command line, and I’ve been hearing about all these people doing stuff with like warp for coding and clod code, and you sent me a video, a review of the new version of Gemini for, you know, Gemini CLI.
Peter: My AI command line experience has pretty much been limited to, in the early days when GPT-3 came out, I used it to code some Python scripts for me to do some cool things that I promptly forgot about and never did again.
Peter: Or never used again.
Peter: And then I found Fabric from Daniel Measler, and I used that for a number of cases.
Peter: And pretty much I’ve weaned myself off of that for everything except like one use case at this point.
Peter: So what I wanted was for you to catch me up.
Peter: Like, I was all in on ChatGPT, leveraging the web interface early on in 2023, right?
Peter: I was heavy in that, using it constantly, finding new applications for it.
Peter: I didn’t really dig into the API-driven stuff, and very quickly, vendors started integrating AI into their platforms and stuff.
Peter: So I wasn’t an early adapter in that sense.
Peter: But since then, I haven’t really been using it like in novel ways.
Peter: So I’m a little bit behind.
Peter: So what I suggested we do today was, you catch me up, turn me into a command line AI aficionado, and catch me up to speed.
Peter: That’s what my suggestion was.
Scott: So I will say that my experience is limited to mostly Cloud Code and the warp AI agent in the command line, but using mostly anthropic models with that, because although they support GPT-5 now, when that first came out, it was broken.
Scott: And I haven’t tried it since.
Scott: So I still need to investigate other things like OpenAI’s codecs, find out how good that is.
Scott: That’s our command line thing.
Scott: I need to try GPT-5 inside Warps agent again.
Scott: And I want to try the Gemini command line.
Scott: Now, the Gemini command line is interesting, because all you need is a Gmail.
Scott: According to that video link I sent you, all you need is a Gmail address.
Scott: And you can get a whole lot of tokens for free.
Scott: Totally free.
Scott: But it’s also slower, according to him.
Scott: It’s like half as slow.
Peter: But he also, in his use cases, he had it chugging away at things for like 45 minutes at a time or something.
Peter: So I’m envisioning these are things that I’m not going to be sitting here, babysitting it unless I’m bored or just curious, like, hey, what’s it up to kind of thing?
Peter: I don’t expect like, oh my God, I have a deadline.
Peter: We got to get this outdoor by 4 p.m.
Peter: Right?
Scott: It is possible, but I will say this, having used Cloud Code to create a Mac app for making, running timers for different work tasks.
Scott: So I can keep track of how much time I spend on a certain work item so I can bill it.
Scott: I will say that there are times when Cloud Code goes off to do something relatively simple and it does take a while and it will come back.
Scott: And unless you say, don’t ask me again during this session, it will ask you at certain times, can I make this change to this file?
Scott: It shows you the diff and it says, can I make this change?
Scott: Can I make this change?
Scott: Which I do like because I look at the diffs.
Scott: I want to know what’s going on.
Scott: And there have been times where I’ve questioned it saying, now that we made this change, isn’t that block right there not used anymore?
Scott: It’s like, you’re absolutely right.
Scott: It’s redundant now.
Scott: We don’t need it.
Scott: Blah, blah, blah.
Scott: And it’s like, why didn’t you figure it out?
Peter: Very Claude.
Scott: Yeah, it is.
Scott: And so, anyway, the point is, there are times where I’m kind of sitting there waiting for things to happen, but it’s also not long enough that I want to leave the computer.
Scott: I switched to other tasks, of course, but what a lot of people do is use multiple agents at the same time.
Scott: So they’ll stick some sub-agents on a different task.
Scott: I haven’t really gotten into that yet, to be honest.
Peter: So, and again, since I sort of had a little bit of a falling out with, you know, the Daniel Measler community, I haven’t been there and I haven’t been on like any of his regular tutorials and stuff.
Peter: But I know he was talking about like a couple months ago, how he’s just being super productive with Claude Code.
Peter: And he’s got to do all kinds of things, not just coding tasks, but it’s doing all kinds.
Peter: And I’ve got agents working on this and that.
Peter: And I’m pretty sure at this point that he’s, at least on the spectrum of manic depressive, because he gets like really manic when, you know, he finds new stuff like this.
Peter: But what I wanted to know is like, what kind of things do you think he’s doing?
Peter: And what kind of things can I be doing?
Peter: And first, and the next thing was, I have that one use case, which I said, you know, my workflow used to be on Sunday nights, I’d record my role-playing game sessions with my friends, record the audio file, and I would just use Audio Hijack Pro, dump out an MP3 or a WAV file or whatever.
Peter: I would feed that to a transcript command that comes with Fabric, you know, with Measler’s Project, TS, and it would connect to the GPT API or the Whisper API and crank out a transcript.
Peter: The transcript, I would then feed into Fabric, which is just a prompt to say, you know, to GPT or whatever model I’m using.
Peter: Summarize this, you know, this is a role-playing game session, tell me what was funny about it, what happened here, what happened there, how much damage was done, how many combat rounds, who did, you know, the best role-playing, etc.
Peter: So the manual part was recording it and dumping a file.
Peter: And then from there, it was pretty much, you know, I had to like one little script that would generate the transcript, and then from there generate the whole summary write-up.
Peter: So it was very quick.
Peter: Somewhere along the way, I changed it because the recording that I was doing, for one thing, I found this Discord plugin called Craig.
Peter: It’s a bot, agent, whatever, I forget the technical term is.
Peter: But it logs into our conversations on Discord, and it knows everybody who’s speaking.
Peter: So just like, you know, when you and I were experimenting with Otter AI for recording, rather than just using one audio stream, which may or may not be able to identify who’s speaking, when you use Craig, it knows who’s speaking, because, you know, everyone is identified on the Discord call as speaking.
Peter: So that’s great.
Peter: But now I get four or five different audio files.
Peter: So I now take those, plug those into Mac Whisper and tell it transcribe podcast.
Peter: So that’s a manual process, right?
Peter: I go to Craig, downloading it with a browser, dragging the extracted files into Mac Whisper, saying transcribe after assigning names to all of the speakers.
Peter: Generally, I start that process.
Peter: I go to bed.
Peter: The next morning I come in, take the transcript and then now since I ran out of API credits with GPT and with Claude at the same time, I take my fabric prompt, paste that into GPT or Claude, and then attach the transcript file.
Peter: So it’s now become a very manual process.
Peter: And I was like, okay, you have definitely at times said that there are other people doing what he’s doing, and it’s not so special anymore.
Peter: So I was like, okay, well, let’s use this as an opportunity, catch me up because I’ve definitely let myself fall behind.
Peter: So where do we go from here?
Scott: Okay.
Scott: Well, first off, I want to absolutely clarify that not only I’m not an expert, there’s a lot of these things that I haven’t done.
Scott: I haven’t done a lot of the types of things that you’re talking about.
Scott: However, I do know that thanks to the MCP protocol, which is something that Anthropic created and everybody else is implementing now.
Scott: You can connect to different apps, and there are various MCP servers that are specific for certain use cases.
Scott: Like you can connect to this database, you can connect to this app, you can…
Peter: Let’s…
Peter: I’m vaguely familiar with what that is, but let’s assume I and the listener know nothing.
Peter: What is MCP even taught?
Peter: What does that even mean?
Scott: It’s for context.
Scott: It’s for model context, so that your model has the context of the information that you want it to have.
Peter: Okay.
Scott: So basically what it is, is like you can give…
Scott: Well, there’s an Obsidian MCP, so let’s just use…
Scott: And there’s a Notion one, so let’s just use Notion or Obsidian.
Peter: I’ve got a lot of stuff in Notion, so that’s good.
Scott: So you have a lot of different…
Peter: That actually is a really great choice, because I’ve been taking all of those transcripts and summaries and dumping them into Notion.
Peter: So if I wanted, say, like a historical analysis of our roleplaying games over time, I could maybe point it at that as opposed to just this session, for example.
Peter: Is that a good use case?
Scott: Yes.
Scott: That’s exactly why the other day I was asking you when we were talking about that use case, your transcripts, I was like, where do you put those things?
Scott: Because I was thinking if it was something like Notion, if it was something like Sitting, if it was something like Notes, if it was something like any of these things that an MCP server has been developed for that you can connect to, you can get at that information.
Scott: Okay.
Peter: That’s cool.
Scott: Additionally, you should be able to connect to a lot of these to the desktop apps.
Scott: You should be able to connect a lot of them to Claude Mac OS app and the ChatGBT Mac OS app.
Scott: Apparently, there’s an update for the Mac GPT app today.
Scott: Let me see if I can update it.
Scott: Here we go.
Scott: Enable work with apps, manage apps.
Scott: So I can hook it up to BB edit, I can hook it up to VS code.
Scott: This is ChatGBT.
Scott: I can hook it up to Apple Notes, I can hook it up to Notion, I can hook it up to prompt, I can hook it up to script editor, I can hook it up to terminal, I can hook it up to text edit, I can hook it up to warp, and I can hook it up to Xcode.
Scott: That’s aside from the other MPC servers, which are also called connectors.
Scott: So if you go into the ChatGPT settings and click on connectors, there’s a bunch of things that you can connect to different online services like Outlook Mail, Teams, which hey, that actually sounds kind of interesting.
Scott: Maybe in theory, you could have it transcribe a Teams meeting for you possibly, or at least search for stuff for you.
Scott: It can connect to Notion, you can connect to your Outlook calendar, you can connect to your Outlook email, stuff like that, Google, etc.
Scott: So not even just the command line stuff, but even the Mac apps themselves should have the ability for you to connect to a lot of different things and get some workflows going that way.
Scott: Now, here’s one thing that I used to do with Raycast AI, that I haven’t developed an analog for just using these other AIs yet, whether it be through the Mac apps or through the command line, and that is specific workflows.
Scott: Like in Raycast AI, I would have something set up to where if I instigated this workflow, it automatically knew, you are here to do this one particular job, and I only want you to do this one particular job.
Scott: I haven’t found it without just, I don’t know how to create a custom prompt that is separate from how I normally use Chat GPT, and I just click on it and suddenly it’s running under that custom prompt.
Scott: Then take that and extend it to yes, and when I run you, not only do I want you to be acting in this mode, but I also want you to just run this workflow.
Scott: Like there should be a way for you to be able to use these MCP servers to get at the stuff that you want, and also have a script or a workflow that you’ve predefined that you can tell Claude to run or GPT to run or something like that, and it’ll just go off and do it.
Scott: So instead of babysitting your, transcripting your games, you just tell it to do it and it should be able to do the whole workflow for you.
Scott: I haven’t done that kind of thing yet.
Scott: So far, my use has been, I am asking GPT or Claude specific questions and it says, oh, I can look in these specific information things.
Scott: I will check that for you.
Scott: And there’s been times where I’ve said, no, no, don’t store that in notes.
Scott: I don’t want that in notes.
Scott: Put it over here in this other thing.
Scott: And then for the command line stuff, it’s been basically using it to write a Mac app, vibe coding, if you will.
Scott: I mean, I’ve used them as code assistance in different code editors before, where it’s like a glorified autocomplete, where it’s not doing the coding for you, but you can get it to say, oh, I bet you’re going to put this code here based on this, or here’s what I would do, and you can accept or reject.
Scott: But the command line vibe, like Claude code and GPT, what did I call it?
Scott: Codex?
Scott: Those are two different, those are completely different, where they’re just going to write the program for you.
Scott: You tell it what to do.
Scott: You tell it what you want.
Scott: You tell it to come up with a plan.
Scott: You tell it to figure out the to do’s.
Scott: You have some files in there that it looks at to keep track of what the status is and what you want from it.
Scott: And then it makes the code and it just presents it to you and says, are you okay with this?
Scott: And either you are or you’re not.
Scott: That’s how I’ve been using.
Peter: Vibe coding, right?
Scott: That’s vibe coding, and that’s what I’ve been using Claude code for.
Scott: And I’ve mostly been using the warp AI agents for that too, although they’re pretty good for other things.
Scott: They’re good for asking general questions of, if they’re good for manipulating files that you can get to through the terminal, stuff like that.
Scott: So I can throw a problem at them.
Scott: Like I can say, write a bash script that uses OX to do this thing for me because I can’t be bothered and warp AI agent can go off and do it for you.
Scott: So things like that.
Scott: But I haven’t done the type of thing that I want to do, which is figure out how to make a workflow that I can fire off from Claude or from GPT and just have it go do it and it can access some of these resources and then get the end result for me.
Peter: Are you ready to beep me?
Peter: You want to just do it?
Peter: So Scott and I have this internal little inside joke.
Peter: What’s the guy’s name?
Scott: Theo.
Scott: He’s on YouTube.
Peter: He’s an AI YouTuber that you follow.
Peter: And you said that he’s usually pretty good.
Peter: And he, this guy, we should have, we got to put a link to this in the show notes.
Peter: He was, he had, because of his, you know, influencer status, I’m assuming, he-
Scott: VOT3, yeah.
Peter: VOT3, yes.
Peter: He had an early preview of GPT-5.
Peter: And I’ll stop, you know, I won’t, I won’t make you bleep the hell out of this section, but I was watching it and he is just like one F-bomb after another describing how blanking amazing GPT-5 is.
Peter: It just blanking does it.
Peter: It doesn’t, you know, that used to be, you have to tell it coerce it and blah, blah, blah, blah.
Peter: And it’s just blanking does it, blanking does it, does it just blank, you know, it’s like F does it.
Peter: It just F does it, F does it.
Peter: It just F, like constant.
Peter: And, you know, I put it in the comments.
Peter: I said, you know, when you, effing have to use effing almost every other effing sentence, it effing starts to lose its effing meaning.
Peter: And it just reminded me of like season three of the HBO series Titans, which was, you know, DC Comics thing, which I liked up until then.
Peter: But whoever was writing it decided literally every scene, someone’s got to be dropping an F bomb in like the stupidest places.
Peter: And then like this is all I could think of.
Peter: You know, it’s just like, dude, when you’re effing using effing every other effing word, and effing is just effing filler, so stop effing around, you know?
Scott: Yeah, he does swear in his videos.
Scott: He normally does not swear that much.
Scott: Now, just for context, he was at Open AI when he filmed that, and so are some other YouTubers and stuff that were known for their AI videos.
Peter: Funneling the Cool Age at the time.
Scott: They were, and they also swear that how GPT-5 was behaving for them is not the same as GPT-5 was behaving when it was released.
Scott: And I tend to believe them.
Scott: I do think they were over-enthusiastic.
Scott: I do think they bought into the hype a little too much.
Scott: But I also totally believe that that pre-production model, that was before it was rolled out, I do believe that it probably behaved differently.
Peter: Yeah, it probably was better, and then they rolled out crap to the masses.
Peter: I mean, it’s not the first time we’ve seen bait and switches happen.
Scott: Yeah, and I’m not even saying that it was intentional.
Scott: I’m not even saying that they intentionally broke a bunch of stuff.
Scott: They probably made some intentional changes, and it made other unintentional changes that they didn’t mean to happen or something like that.
Scott: You know what I’m saying?
Scott: I don’t think exactly…
Scott: For sure, I know that certain things that happened were not exactly as open as I wanted them, based on what they say.
Scott: Either that or they’re big liars.
Scott: And you would never accuse Sam Altman of lying or saying what people wants to hear, I don’t think.
Peter: No!
Peter: No, we’re not doing that.
Peter: I mean, that’s almost as bad as accusing our president of lying, Scott.
Scott: I think Sam Altman is one of those people that always instills utter confidence that he’s not just saying what you want to hear.
Peter: Of course, you can trust everything he says.
Scott: Right.
Scott: So anyway, you find GPT-5 good for…
Scott: I like it for being quite a bit terser than Claude.
Scott: It basically just tells me what I want to know, and it does it succinctly.
Scott: Now, one thing I don’t like about ChatGPT with GPT-5 is that after every piece of information it gives me, when it’s done, it will say, would you like me to show you an example of how to hone in on this one specific thing?
Scott: No, I never want those.
Scott: I never want exactly what it’s asking.
Scott: And first of all, I just want to digest what it told me.
Scott: Secondly, if I need more information, I will figure out which more information I need, and I will ask for it.
Scott: And I have put in the personalization section, which is supposed to be where you’re supposed to tell it how to behave, as far as I understand, never ask me a follow-up question.
Scott: Never ask me if I want an example.
Scott: Never ask me if I want more information.
Scott: Never ask me if you want me to do this one specific thing related to the conversation we just had.
Scott: Never ask me.
Scott: Never, never, never.
Scott: And it asks me anyway.
Scott: It just doesn’t effing do what I tell it.
Scott: It just effing keeps asking me, Peter.
Peter: It just effing doesn’t do it.
Scott: And I have even said to it, don’t ever do that.
Scott: Oh, I won’t, you’re right, I won’t.
Scott: Okay, great.
Scott: But see, AI, these LLMs are kind of stupid.
Scott: Like, the reason you have a clod.txt in your project when you’re vibe coding is so that it can go back and remember, what’s my status?
Scott: What are we doing?
Scott: What is this program even?
Scott: Because literally in the middle of a session, you can be working on something and then you can say, okay, now let’s work on this other thing.
Scott: And it goes, you’re right.
Scott: Let me look at the code and see how it’s structured and see what this program does.
Scott: And I’m like, you wrote this, you idiot.
Scott: You wrote this, you moron.
Scott: And they are, they’re morons.
Scott: That’s the thing.
Scott: They have no, it’s really bad.
Scott: And then I showed you a screenshot of a thing I was doing yesterday.
Scott: I was having it export CSV reports or PDF.
Scott: I don’t remember which one it was working on at the time.
Scott: And it says, there, I’ve implemented your PDF thing and your export thing and it works great.
Scott: And it will not be crashing.
Scott: It will be robust and professional looking.
Scott: I run it and it immediately crashes.
Scott: And I’m like, it’s crashing.
Scott: And I gave it the exact.
Scott: So I was running it in Xcode because, of course, I’m working on a Mac app.
Scott: So it can edit the files, but it can’t really, it’s not an agent mode.
Scott: It’s not clicking on Xcode and compiling and building it for me.
Scott: It can build it in the command line, which is interesting because it can do the build and see if it builds, which is helpful.
Peter: And we did talk about my examples with, yeah, we talked about my experience with agent mode the other day where-
Scott: Yeah, and I also want to talk about that in a minute, too.
Scott: I think it was GPT-5.
Scott: No, was it Claude?
Scott: Anyway, we’ll talk about that in a second.
Scott: But anyway, so one mistake that it makes is if it can build the project, it assumes that everything’s working as it intended.
Scott: But that isn’t the case.
Scott: Building the project just means it’ll build.
Scott: It’ll compile without errors, or at least with just warnings or something.
Scott: And then you run it, and then you try to export the file, and it crashes because it wrote it wrong.
Scott: So you give it the error thing, you give it the stack trace, and it goes, oh, I see what the problem is, and it fixes it.
Scott: But it was absolutely hilarious because in one line of my screenshot, it is crash-free and robust.
Scott: And right in the next line, I wrote, it’s crashing, and I give it the crash message.
Scott: And then below it says, oh, it’s crashing because it’s just hilarious.
Scott: It’s like it just it gaslights you all day long.
Scott: And it there are so many times where if you didn’t know anything about programming, you would be lost because you have to keep track of it.
Scott: Another thing about Cloud Code is it all of a sudden goes, I’m going to build in all these amazing features.
Scott: A time-tracking app, he must want all these things.
Scott: Let’s import and export from these other things.
Scott: Let’s do all these kinds of crazy reports.
Scott: Let’s build out all these other features.
Scott: And I’m like, I don’t want any of that.
Scott: Get rid of all that stuff.
Scott: Quit adding it.
Scott: It kept adding some of that stuff back in over time.
Scott: So you have to manage these things like crazy.
Scott: I find…
Scott: Either I’m not very good at this, which 100% is possible, because I’ve just started using these things this way.
Scott: So it’s entirely possible that I’m not using, like, anthropics suggested means of setting cloud code up for success or something like that, whatever.
Scott: But I have found that there are times where it just goes off the rail so badly, or it’s just so slightly dumb, that I don’t know how all these guys on YouTube have all this massive success.
Scott: Like, they just say, oh, it’s amazing, it does this thing.
Scott: No, I bet they fought their way through it.
Scott: I guarantee they did.
Scott: They had to have.
Scott: It wasn’t just that easy.
Scott: For sure, you can’t one shot stuff.
Scott: But even just iterating, you have to work really hard to make sure that it’s giving you what you want.
Scott: And I will say, it’s probably better at things that aren’t Mac apps.
Scott: I bet that Swift UI is not one of its super strengths.
Scott: I bet if I was doing a web project, I wouldn’t have the same exact experience as I’m having.
Scott: But even so, I would say, if you’re going to vibe code, you have to be a programmer and you have to stay totally on top of these things.
Scott: So I think what we should do though, Peter, is make a list.
Scott: What I’ll try to do is gather some information and make a list of things that we should install for you and try to build a workflow.
Scott: So you come up with a workflow that you want, and together we’ll try to figure out if there’s a way to do it, and then we’ll see if we can do it and we’ll see if it works.
Scott: And we can talk about that at some point.
Peter: To query my specific notion database of all of these transcripts that I have from my role playing games would be a start.
Peter: And some of it too, another stuff, I have more player written summaries also in a Google Doc.
Peter: So that’s another repository that we could, I’m assuming Jem and I would be easily querying that for me if I wanted to.
Scott: I would hope so.
Peter: You know, I would think.
Peter: So some interface to query those like, has this character ever encountered this character before?
Peter: Has Superman ever fought the Hulk, for example?
Scott: Right.
Scott: So you want to, yeah, okay.
Scott: So it would be nice to have some kind of, even if you’re doing it through typing, whether it be in the command line or whether it be in a chat app, and you’re typing a question to GBD, it would be nice to have some pre-configured work flow where when you open that up, it knows that it’s going to be accessing Notion, it knows that you’re going to be asking it this type of questions, and it knows that you want these types of answers, and then it has the ability to do that.
Peter: Yep.
Peter: That’s one use case.
Peter: Another would be streamlining my whole recording, transcription, summarization process.
Scott: Interesting that you talk about that.
Peter: Yes.
Scott: Because I have not done that with any of these tools.
Scott: What I have used for that recently is I’ve kind of gotten fed up with the integrated AI note takers that will log on to your meeting for you and proclaim that they’re Peter Nikolaidis even though they’re not.
Scott: I love it the way yours says PN, and I think, oh, that’s not Peter.
Scott: It’s a stupid note taker.
Scott: It just so happens to have the same initials as Peter.
Scott: And so anyway, I got fed up with those and I started and I use Mac Whisper, but I haven’t really used anywhere near most of its power because I normally I mostly use it these days for transcribing Syracuse says clips.
Scott: And then I realized because I just use the Apple Podcast transcripts for our podcasts.
Scott: They’re so good.
Scott: They’re so bloody good.
Scott: They’re way better than anything Mac Whisper can do or any whisper.
Scott: But then I realized, wow, Mac Whisper has the ability to see that a Teams call or Teams meeting is in progress and record it.
Scott: You can record it and then you can immediately after it’s done and has the recordings, it’ll start transcribing for you.
Scott: So one thing I found out though, the hard way is that being a Mac app, Mac Whisper has a microphone setting and you have to go into settings.
Scott: Now, what I did do though, however, Peter, is I changed it so that it follows the system default.
Scott: So whatever I set my system input to, that should be the input for Mac Whisper from now on.
Scott: I will double check it like a zealot for a long, long time until I believe it, but that is how it should work.
Scott: Then the other thing too, is I noticed I like to use my studio display mic and speakers for meetings.
Scott: So I can sit there without earphones, without having to worry about anything.
Scott: Teams and other VoIP callers, they do a pretty good job of screening out the audio coming from the speakers bouncing back into the microphone, right?
Scott: So if you have the AI note takers running in Teams like Granola and Otter and all those, you won’t have a problem where it thinks that you’re saying some of the things that other people are saying.
Scott: But I found with MacWhisperer when I recorded you and I talking in a Teams meeting, it was attributing all of your words to you correctly, and then some of your words also to me because it was getting the feedback.
Scott: MacWhisperer does not have the ability to block that incoming reflecting audio from the speakers to the mic.
Peter: And again, that’s where something like an API-driven recording like Craig or otter.ai or something which can actually track the inputs discretely has an advantage, right?
Scott: Well, it’s interesting because MacWhisperer does separate the speakers.
Scott: It doesn’t know their names.
Scott: You have to assign the names.
Peter: Right.
Scott: But it does separate them by Speaker 1, Speaker 2, you know, whatever.
Scott: And yeah, so it’s interesting.
Scott: So in theory, you would think that it would know that the speaker that’s saying those, first of all, you would think that it would be able to say, Oh, why is Speaker 2 repeating everything Speaker 1 said?
Scott: And why does it sound like Speaker 1 when Speaker 2 repeats it?
Peter: Yeah.
Scott: So I might send.
Peter: I’ve noticed also using MacWhisperer, you know, doing the transcript to the transcription, it injects a lot of garbage like you saw, you know, it’s like Speaker 1, you, Speaker 2, you, Speaker 3, you.
Peter: It’s like you’re saying you, you, you, you.
Scott: Now, you had that problem in a different case, but in my case, the only time I had that problem was when I didn’t realize that MacWhisperer had its own microphone setting, and it wasn’t actually recording my mic because it was recording the wrong mic.
Scott: It was recording a mic, a mic that I had muted.
Scott: It was recording my podcast mic.
Peter: On yours, it was recording, but for mine, MacWhisperer was just transcribing audio files that I gave it.
Peter: It wasn’t, MacWhisperer wasn’t even involved in the recording.
Peter: So I’m like, it seemed, I think that sounds like a hallucination.
Peter: It also, I got to show you, it injects a lot of weird characters.
Peter: I don’t even know what they are, like if there’s some kind of Unicode or something.
Scott: First of all, there are different models that get better results.
Scott: The Whisper Kit, large model version, whatever, three or four.
Scott: Let me see MacWhisperer.
Peter: And also, let me, another comment on MacWhisperer.
Peter: Sometimes it puts in in its transcripts, like Flawless Japanese.
Peter: And other times, it has detected like tie.
Peter: Not a word or two, but like complete sentences, which I assure you, no one is speaking on these recordings.
Scott: Peter, have you ever considered that you’re somehow recording your entire neighborhood?
Scott: Congratulations, you are now part of the NSA.
Scott: No, I think now, so try it again.
Scott: If you haven’t tried it for a while, try it again now.
Scott: I think it’s much better that way.
Scott: But also try, try the Whisper Kit.
Scott: Which one do you use?
Peter: Whisper Kit?
Peter: Which one?
Scott: Yeah, the Whisper Kit Large V3.
Scott: It’s like 3 point something gigabytes.
Peter: Okay, I am using right now, I am just using…
Peter: No, that’s what I’m using, Large V3 Whisper Kit.
Peter: 3.1 gigs, that’s what I’ve got.
Scott: Transcription models, yeah.
Scott: So try it again.
Scott: It’s been updated, I’m sure.
Scott: So try and see what happens.
Scott: I did try the Parakeet model, the NVIDIA Parakeet ones, which are supposed to be really good.
Scott: I tried the V3 and it’s like a third of the size or something.
Scott: But I got a lot of errors.
Scott: Like you and I were talking about like specific vendors of services that we were talking about and it wasn’t getting those names.
Scott: But the Whisper Kit V3 one does get those names correct.
Peter: And V3 should be better than V3 Turbo, right?
Peter: Turbos are stripped down versions?
Scott: I think so.
Scott: You could compare the model size.
Scott: You could download it and then just reveal and find her and then just compare the model sizes.
Peter: Turbos got to be smaller because I just downloaded it.
Peter: Like I just clicked it and it’s 70% downloaded.
Peter: So there’s no…
Scott: Oh, that’s why it’s called Turbo.
Peter: Yeah, that’s why I was figuring it’s probably a smaller model, right?
Scott: So it’s a Turbo download.
Peter: So that in theory won’t be as good.
Scott: So in the settings, if you right click on one of those that are downloaded, it should have a show and finder option for you in the Mac Whisper settings.
Scott: Just look at one that you’ve downloaded, right?
Scott: Click on it.
Scott: Yeah, reveal and finder.
Scott: And then it also shows the size to right beside it if you’ve downloaded it.
Peter: So it doesn’t show the size on the Turbo one that I just downloaded.
Peter: The regular large V3 is 3.1 gigs.
Peter: The other one is about one.
Peter: It’s half the size, 1.6 gigs.
Scott: Large V3 Turbo is 1.6?
Peter: Yep.
Scott: OK, I just started downloading it to see, and I bet you’ve already given us the information that we need.
Scott: You are on top of it, Peter.
Peter: Yeah, so, yeah, and again, when I used, I don’t remember, I don’t know what model I used specifically, but when I used the TS command that Measler again, bundled with Fabric, that was just making it a call, you know, a Whisper API call and doing the transcription.
Peter: And that seemed to be a lot more accurate.
Peter: But since I purchased Mac Whisper, I was like, well, why would I be paying for every API call to use transcription?
Peter: Plus, I wasn’t sure, you know, how I would correlate and collate four or five different audio files when I’m doing the transcripts, because I’d have to throw timestamps and then, you know, interweave them all together and Mac Whisper does that.
Peter: So, you know, so there’s a lot of stuff that I need to figure out how I’m going to replace my manual workflow.
Scott: So, yeah, you’re absolutely right.
Scott: The large V3 is just about double the size of the large V3 turbo.
Peter: 3.1 versus 1.6.
Scott: Yeah, so it’s a lot.
Scott: And I believe when it’s doing the transcription, I think it has to have that model loaded into memory.
Scott: So just be aware.
Scott: Now, luckily, it does the full recording of the meeting and then the meeting ends and then you’re done recording before it tries to do the transcription.
Scott: What I don’t know is does it load the model into memory as soon as…
Scott: I guess there’s probably an easy way to figure out, but does that model load into memory the minute MacWhisper opens even if it’s not actually transcribing at that time?
Scott: I would assume so, but…
Scott: Right.
Peter: See what happens when you open it.
Scott: You can.
Scott: You can.
Scott: So, physical memory used.
Scott: So let’s close MacWhisper.
Scott: Memory used is 18.6 right now.
Scott: Now let’s open MacWhisper again.
Peter: Real time troubleshooting here, people.
Scott: Yeah, it only went up a gigabyte.
Scott: Now, if I drag some audio files, do I have some audio files for it to transcribe?
Scott: Let me find some.
Scott: I’ve got tons of audio files on my computer, Peter, because I don’t know if you know this, but I am a podcaster.
Peter: I’ve heard that about you.
Scott: Was it said lovingly?
Peter: No.
Scott: Oh, darn it.
Peter: Because I’m a podcaster.
Peter: Podcaster.
Scott: All right, so let me transcribe this.
Scott: Okay, this is hilarious.
Scott: I started transcribing and my memory just went down.
Scott: Now it’s going up.
Scott: Now it’s going up.
Scott: But not much.
Scott: It basically went up from 18 gigabyte to 19.
Scott: Now it’s diving back down to 18.
Scott: So I don’t know what the answer is.
Scott: I think the answer is it probably loads that into memory right away, because how much is MacWhisper using?
Scott: But MacWhisper, the whole MacWhisper is only using 2.32 gigabytes, which is less than the model that it’s using to do the transcription.
Scott: That’s bizarre.
Peter: We say right away, you mean when it starts to actually transcribe something?
Peter: Yeah.
Scott: Okay, that makes more sense.
Scott: So that doesn’t make sense.
Scott: How can MacWhisper be using less than the size of the model that’s loaded when it’s running the transcription?
Scott: I don’t know.
Scott: These are questions.
Scott: Maybe I’ll send the developer of these questions when I also send him some comments about things that he might want to fix.
Scott: By the way, I’m not complaining.
Scott: This MacWhisper, he’s done so much work on it.
Scott: The capabilities that it has now compared to what it used to are pretty stunning.
Scott: So I think Peter would agree, and if he won’t, I’ll say it for him.
Scott: I agree.
Scott: Wait, that doesn’t sound like Peter.
Peter: I agree.
Scott: No, that doesn’t sound like Peter.
Scott: Peter, sound like Peter for a minute.
Scott: Sound like John Syracuse and say you agree.
Scott: Okay.
Scott: That’s no fun.
Scott: John Syracuse never says I agree.
Peter: Because otherwise Casey’s tagline wouldn’t be, John, tell us why we’re wrong.
Scott: Oh, that’s hilarious.
Peter: So what do I need to get started with these things?
Scott: Okay.
Peter: So I have Warp installed.
Peter: That’s as far as I go, right?
Scott: And Warp, you can choose the model that you’re using in Warp.
Peter: I have a $20 for the moment.
Peter: I have a $20 Claude Pro subscription, or $20 Claude subscription, whatever.
Peter: I don’t Pro Enterprise, whatever.
Peter: Not the $200 a month one.
Peter: And I have a Google account.
Peter: And I did do a Brew install Gemini CLI, or whatever command that is.
Peter: That’s as far as I’ve gotten.
Peter: What do I do next?
Scott: Was it Brew or was it NPM?
Scott: I think it was NPM.
Peter: I used Brew.
Scott: Oh, I didn’t know that.
Scott: Now for Claude code and for the OpenAI Codex, I believe you’re going to have to use NPM.
Scott: Claude code is not in Brew as far as I know, at least last time I checked.
Scott: You have to do, I’m sending you a link for the setup, which I will also of course send to, I’m sending it via iMessage.
Peter: Claude code is a cask in Brew.
Scott: Oh, it is?
Scott: Is it Claude-code?
Peter: Claude-code.
Scott: Okay.
Scott: Well, then I would do it that way.
Peter: There’s a Claude cask and Claude code cask.
Scott: Yeah, the Claude cask, I believe, is the Mac app.
Peter: Got it.
Peter: So I set up Claude code.
Peter: Don’t use npm, as this can lead to permissions issues.
Scott: No, no, no, no.
Scott: It says don’t use sudo npm.
Scott: It says to use npm install minus g, but don’t use sudo npm install minus g.
Peter: Got it.
Peter: Got it.
Scott: Got it.
Peter: Okay.
Peter: So I can do the native binary.
Peter: I could do Windows PowerShell.
Peter: Great.
Peter: I’m going to, and I know this is dangerous, but I’m going to assume that a Brew cask install Claude code is going to be functional.
Scott: I think it should be what you need.
Scott: Yeah, I believe so.
Peter: So I’m doing a Brew install Claude hyphen code.
Peter: It’s done.
Peter: There it is.
Peter: Linking binary Claude to opt homebrew bin Claude.
Peter: All right.
Peter: I type Claude.
Peter: And I’ve been getting this warning a lot.
Peter: Let’s start.
Peter: I first started noticing this when I did fabric a while ago, but I typed just Claude and I get this warning.
Peter: All log messages before absl colon colon initialize log is called are written to standard error.
Peter: I 0000 00 colon 00 blah blah blah blah blah blah blah blah blah.
Peter: Check GCP environment.
Peter: No op blah blah blah blah blah.
Peter: So it’s something’s trying to check a Google Cloud.
Peter: Come, you know, compute, I think I don’t know what’s going on platforms other than Linux and Windows are not supported.
Peter: I get this weird error message when I do just about anything on the command line.
Peter: So somewhere I have to have something.
Peter: Yeah, somewhere I have something like configured to call something that I’m not using.
Scott: Either that or the people pwning your machine or have it configured.
Scott: So I got, I just sent you a link to the Homebrew formula page for Cloud Code.
Scott: And it has some interesting information there on the updater.
Scott: It says, Cloud Code’s auto updater installs to.local bin cloud and not to Homebrew’s location, which is fine.
Scott: Cause I assume that’s where I didn’t check, but I’m sure, well, no, actually I’ll have to look.
Scott: Anyway, it is, it installs it there.
Scott: It says it is recommended to disable the auto updater, which is a good idea because the auto updater always fails anyway.
Scott: And then instead use brew upgrade, hyphen hyphen casc cloud code to upgrade cloud code.
Scott: So I think that’s probably good advice because it seems to me, I installed cloud via the NPM route and its auto updater always fails.
Scott: So I don’t see any point in using it.
Scott: Warning, all logged messages before ABSL are written to standard error.
Scott: Oh, this is interesting.
Peter: With disable auto updater equals one.
Peter: Where do I put that?
Peter: Is that in a config file or is that an environment variable?
Scott: What are you talking about?
Peter: The instructions you set, it says, its updater installs to tilde.localbinclod and not homebrewslocation.
Peter: It is recommended to disable the auto updater with either disable auto updater equals one or blah, blah, blah, blah, blah.
Peter: Well, disable auto updater equals one.
Peter: Where does one put that?
Peter: Is that an environment variable?
Peter: Is that going to a config file somewhere?
Peter: And if so, where?
Scott: Yeah, that’s a good question because I don’t have, that’s not where my Claude install is.
Scott: So I don’t, let me see, where is Claude?
Scott: Which Claude?
Peter: Or Claude config set auto updates equal to false.
Peter: I think I can do that.
Peter: That’s easy enough.
Scott: Yeah.
Peter: That’s what I’ll do.
Peter: And, oh, look, usage.
Peter: Oh, I know why it’s failing.
Scott: Ha ha ha.
Peter: I aliased Claude to a Fabric call to use Claude as its model.
Peter: So when I type Claude, that’s not what, it’s not what I think it is.
Scott: That’s why.
Peter: Yes.
Scott: And Fabric is throwing you a message.
Peter: Yes, that helps.
Scott: Gosh damn it.
Scott: You told me you were done with that guy.
Scott: You said, I’m at war with his entire community and I will fight them.
Scott: That’s exactly what you said.
Scott: Peter, get that off your computer.
Peter: I never said anything like that, but okay, good.
Scott: I heard it.
Scott: Everyone who listened to this podcast heard it.
Peter: Yeah, I know, but not everyone listening to this podcast has a traumatic brain injury like you do apparently.
Scott: No, it’s a finger injury, damn it.
Scott: It’s a finger injury.
Peter: You keep telling yourself that.
Scott: It’s a tendon.
Scott: I don’t have that many tendons in my brain.
Peter: All right.
Peter: All right.
Peter: So now I got to remember where did I have my, which is my Z login file or is it my Z profile file or my Z?
Scott: Or is it your Zsure C?
Peter: Yeah.
Peter: So I aliased Claude to Sonnet.
Peter: Interesting.
Peter: And Sonnet was a call to say, use, you know, call fabric, blah, blah, blah, blah, blah, with Sonnet module.
Scott: Do you have a B in your Sonnet bonnet, Peter?
Scott: Your wish is I was dead.
Scott: He’s like, why did it have to be the finger?
Scott: Why not the head?
Scott: Just, you know, somewhere.
Peter: Yeah.
Scott: Why couldn’t you have been the neck that snapped?
Peter: Fabric aliases that I made.
Peter: So I made one called Gemini, one called Sonnet, one called Lama, and one called GPT.
Peter: So instead of saying, like, fabric, use this command, it was just, you know, I would just say Claude, and it would say, okay, I’m now going to run fabric, do whatever Peter wanted me to do with fabric, but I’m going to use this model.
Peter: So I could easily switch without changing the default model every time.
Peter: But I’m not using any of those anymore.
Peter: So I need to, I need to back these up.
Peter: I don’t want to get rid of them all together, because I might use them at some point.
Peter: And these were kind of cool, you know, functions.
Scott: But Peter back the fabric pass right up there.
Peter: I just need to back this right up, back it right up there.
Peter: So anyway, all right.
Peter: Well, that’s a little bit of troubleshooting that was helpful.
Scott: Trouble shouting.
Scott: It’s always trouble shouting.
Scott: It was trouble shouting.
Scott: There should always be shouting involved with the trouble.
Scott: Yeah.
Scott: So okay.
Scott: So now you got Cloud Code installed, you can do that.
Scott: And actually, I’m thinking about uninstalling the NPM way and installing the Brew way, because I kind of like that.
Scott: I hope and believe.
Scott: I don’t have anything custom really set up in Cloud Code.
Scott: I’ve got, for projects that I’m using it for, I’ve got cloud.md.
Scott: I’ve got all the files local to the projects that I wanted to look at.
Scott: So I think I should be fine.
Scott: So I think I’ll do that.
Scott: I think you just taught me something, because last time I checked, when I installed Cloud Code, there was not a Brew cask.
Peter: So that explains it though.
Peter: That weird error message that I was seeing, I was seeing on Fabric because I was literally calling Fabric.
Peter: So it is something that happened in, I was just seeing Fabric.
Peter: So that’s why that happened.
Scott: Okay.
Peter: So now, and the funny part is too, is I just typed Claude and I got that error message.
Peter: This morning, I typed Gemini and I got that message because again, that was also at least Fabric.
Scott: Why do these AIs hate me so much?
Peter: So now I’m going to just type Gemini and see what happens.
Peter: Oh, this is totally a different experience than I had this morning.
Scott: Oh boy.
Peter: I typed Gemini and now I get this ASCII Art Gemini with a prompt and cool stuff happening.
Scott: That’s the actual Gemini CLI, I bet you.
Peter: Right, which I’ve never seen, because I was just seeing fabric waiting for me to prompt it to do something.
Peter: And now I’ve just typed Claude, and I have welcome to Claude Code.
Scott: Look at that!
Scott: You know, screaming, screaming, Measler doesn’t sound the same as Cone!
Scott: But I would still like it to see as Claude Measler.
Peter: I have just chosen the Claude Code dark theme, colorblind friendly.
Peter: Thank you very much.
Peter: That was very, that was very cool of you guys.
Scott: You’re not actually colorblind, you’re just wearing those glasses with purple lenses, and it’s filtering some colors out for you.
Scott: Hey, I have a question.
Scott: What’s Measler’s first name?
Peter: Daniel.
Peter: All right, so now I’m prompted to log in with Gemini, and it opens up a browser and says, who do you want to be?
Peter: I’m clicking.
Scott: Who do you want to be?
Scott: Who do you want to be?
Peter: I’ve got a number of Google accounts, I’ll use this one.
Peter: All right, Gemini Code Assist and Gemini CLI are now authorized to access your account.
Scott: Beauty, mate!
Peter: You can now close this window.
Scott: Good, close that window.
Scott: Don’t ever look at that window again.
Peter: So now I am back at a Gemini prompt.
Peter: And all right, cool.
Peter: Meanwhile, Warp is saying, hey, try AI coding in Warp.
Scott: It does.
Scott: Every time I say Claude and I launch Claude Code, it’s like, hey, have you heard the good news about AI, AI, Warp?
Peter: It’s so needy.
Scott: It’s so needy.
Scott: I like to spread my tokens between the two.
Scott: So I do use the Warp AI agent.
Scott: The thing that I am trying not to do is switch back and forth between them on the same project, because I’m afraid that they’ll fight each other.
Scott: Even though I’m using Sonnet models for both, even though I’m using Sonnet models for both, I’m sure that Warp AI does something different because it’s just calling the Claude APIs.
Scott: It’s not using Claude Code specifically to do its Warp AI agent.
Scott: It’s acting as the agent, and then it’s using the Claude model to do the processing.
Scott: So there is a difference.
Scott: It’s not the same as just running Claude Code, even though I’m using the Claude model.
Peter: But the persistence would be like it’s spinning out code on the disk, and then you’re basically…
Peter: Is that not like switching between two different programmers?
Peter: You’re like saying like, my programmer buddy Claude generated all this, and now I’m handing it over to my friend Gemini to take over from there?
Peter: Is that what you’re describing?
Scott: Well, I don’t know about Gemini, but like when I’m doing it between Warp AI using a Sonnet model and Claude code, yes, I tell them both, look at the Claude MD file, which has A, all the things I want in the program, what the app is supposed to do, all the instructions, and also I have it check mark as it does the to-dos, and we go down the list of to-dos, it check marks those, and then I also have a separate progress.md, and I tell it to put the changes that it made in there.
Scott: So besides the Git commits, I also have a list of changes made in progress.md.
Scott: Now, I haven’t looked at progress.md for a while and made sure that they’re not screwing it up by just saying, oh, we added these features instead of, here’s specifically what I changed.
Scott: But anyway, when I switch, yes, I tell the new agent, go read those two files and, you know, tell me that you understand before we proceed any further.
Scott: Yes, so in theory, yes.
Scott: And all of these, like the Claude code, the Claude subscription and the ChatGBT subscription, and I’m not sure about Gemini, but they all have memory now.
Scott: So they will remember past conversations.
Scott: I don’t know how much of that applies to Claude code.
Scott: I don’t know how much that applies to their vibe coding interfaces.
Scott: Don’t think it does.
Scott: I kind of feel like the memory thing could be a bad thing, though, because I don’t want it to remember everything that we talked about.
Scott: I only want it to remember specific helpful stuff, and this is why MCP servers and giving it context, like giving it your Notion files, is helpful, because those are specific things that can go look at and query.
Scott: But when you’re having a conversation with AI, sometimes you’re wasting time.
Scott: You’re saying dumb things, it’s saying dumb things back.
Scott: You don’t want it remembering that when you’re asking it new stuff.
Scott: So I’m not 100% sure how you can filter the garbage out of those memories.
Scott: It’ll be interesting to see.
Scott: Don’t remember the dumb stuff we talked about.
Peter: Well, it’s funny you mentioned that, because in my specific use case, like whatever I put down in Notion, if I write notes in it, then that is like, you know, canonically what happened in this fictional universe.
Scott: Right.
Peter: The transcript is made up of what people said, right?
Peter: And that’s not always what happened, because they could say like, oh my God, you know, Martian Manhunter has betrayed us.
Peter: Well, in reality, no, Martian Manhunter was, you know, playing as a double agent, and he actually betrayed the bad guys by thinking, you know, tricking them to thinking that he was on their side, right?
Peter: But yeah, I could very easily think, oh yeah, and it turns out Martian Manhunter is really a bad guy.
Peter: So it’s, I could see it having a hard time if I say, like, this is all, all these things over here were written from a user’s perspective, but everything over here really happened.
Peter: And I can totally see it convoluting that.
Scott: Yeah, absolutely.
Scott: Peter, you say that whatever’s in there is what actually happened in the fictional universe.
Scott: But if you remember in the Bible, it says that God spoke things into existence.
Scott: How do you know that you’re not the god of this existence, this reality that you’ve created, and you spoke it into existence, and now when you say that something happened there, it did happen there.
Scott: You don’t have to say this fictional reality, Peter.
Scott: You’re too humble.
Scott: Own it.
Peter: Okay, I am god here, so saith the lawnmower man.
Scott: Yeah, exactly.
Peter: Now, that was a classic movie about hacking and stuff, wasn’t it?
Peter: No, not really.
Peter: Virtual VR.
Peter: That’s what it was.
Scott: VR, yeah.
Peter: I was going to say not AI, VR, yes.
Peter: So, yeah.
Scott: All right, well, we’ve talked a long time.
Scott: We have a specific chore to do for some future edition of Friends with Bots, apparently.
Peter: We trouble shouted a problem that I had with my fabric install.
Scott: Hey, if nothing else happened in this whole hour of podcasting, that was worth it.
Scott: I don’t know if it was worth it for the listeners.
Peter: That was great.
Peter: So anyway, I set up my Gemini.
Peter: Now, Claude is now saying, hey, what are you using Claude account with a subscription?
Peter: I say yes.
Peter: Opening browser to sign in.
Scott: Yep.
Peter: Okay.
Scott: See, this is great.
Scott: You don’t have to use API.
Peter: Login successful.
Peter: Press Enter to continue.
Peter: Enter.
Peter: Do you want to use an API key?
Peter: Yes or no?
Peter: Recommended.
Peter: Well, no, recommended.
Peter: I’ve already paid.
Peter: I’m not going to.
Peter: Let’s say, okay.
Peter: And Claude can make mistakes.
Peter: Good to know.
Peter: You should always review its responses, especially when running code.
Peter: And due to prompt injection risks, only use it with code you trust.
Peter: Excellent.
Peter: I thought you were going to be doing the code, Claude.
Peter: Wait, I’m confused.
Scott: Well, here’s the thing.
Scott: Don’t forget, you can connect it to MPC server, MCP server, MPC, MMPC.
Peter: MPCs.
Peter: Well, MPCs, that’s from my names.
Scott: You can connect it to MPCs, like Elon Musk is fond of calling everyone an MPC.
Scott: You can connect it to MCP servers, which I haven’t looked deeply into it, but I believe that most security experts will say that MCP servers can be a security risk because the way they authenticate or don’t, and also because of what they can contain.
Peter: So next thing.
Peter: So here I’m seeing my Gemini prompt, and it says, tips for getting started.
Peter: Ask questions, edit files, or run commands.
Peter: Be specific for the best results.
Peter: Create a Gemini.md file to customize your interactions with Gemini.
Scott: See, this is where…
Peter: Or slash help for more information.
Scott: Right.
Scott: Just like it will look for Gemini.md, Cloud will look for Cloud.md.
Scott: I don’t know if OpenAI will look for GPT.md or whatever their codecs looks for.
Peter: What’s the structure of those?
Peter: I mean, they’re markdown files, but what else?
Scott: They’re just markdown files.
Scott: I think you can just put whatever you want in there.
Peter: It’s like you’re prompting it?
Peter: You’re just saying this is…
Scott: Yeah, you’re saying, like, for example, if you want it to be a project-specific file rather than a…
Scott: Like, I don’t have a global…
Scott: Do I?
Scott: I don’t think I have a global Cloud file.
Scott: Maybe I do.
Scott: Maybe I created one somewhere.
Scott: But I put them in my project and I say, here’s what this app does.
Scott: Here’s the features that I want it to have.
Scott: Here’s the, you know, stuff like that.
Scott: But I think that you can create a global one of those as well.
Scott: And I think that you can give it specific prompts in terms of, here’s how you behave.
Scott: Here’s what you do.
Scott: Here’s what I use you for.
Scott: And by the way, you should be able to see these data stores through these MCP servers or something like that.
Scott: You might not need to tell it that, but I think you can tell it that.
Peter: So Gemini, by default, it says, you are running the Gemini CLI in your home directory.
Peter: It’s recommended to run this in a project-specific directory.
Peter: So do I quit out of Gemini, make a new folder CD into that, and then run Gemini again?
Peter: Is that how I do that?
Scott: Yeah, you can do that.
Scott: Or open another tab, or open another pane in that tab, or whatever.
Scott: But yeah, make a new directory for your project.
Scott: Like, I have scripts in my scripts folder, and then sub-categorize from there.
Scott: I have my Mac apps in documents, Mac apps.
Scott: So if I’m working on a Mac app, I go in there and I have a Xcode folder full of my Xcode project, and in there, it’ll let it make its Cloud MD file.
Peter: And apparently to quit out of Gemini, you press Ctrl-C twice in rapid succession.
Scott: It’s kind of interesting.
Scott: So in Cloud, you can just type exit in its little fancy prompt.
Peter: I typed exit and nothing really happened.
Peter: I typed quit and it’s like, betting farewell for now, but it didn’t actually quit.
Peter: And then I hit Ctrl-C and it’s like, hit Ctrl-C again.
Peter: I hit Ctrl-C again, nothing happened.
Peter: Then I hit it, Ctrl-C, Ctrl-C, and then it actually quit.
Scott: Have you tried powering off your computer before?
Peter: I was trying to power off Gemini.
Scott: All right.
Scott: Well, next time, we can talk about what we’ve done with these in the meantime, because you have given me some things to think about, because like I said, I haven’t done a lot with workflows, like setting up a workflow that I can just run and know that it’s going off and doing something.
Scott: I have connected different apps to the Clot app and the OpenAI app so that I can ask it questions about those things, but I haven’t set up any great workflows yet.
Scott: So that is something that I want to do where it’s not vibe coding for me, but it’s doing some other tests that I want it to do.
Scott: So yeah, I have that on my to do list and you have, I don’t know what you have on your to do list.
Peter: And I don’t remember if we talked about this on the podcast, but I definitively at this point know that the reason that my finder specifically in like dialogue boxes with my browser and preview was hanging and pausing for sometimes up to at least seven, but sometimes up to like 30 seconds at a time.
Peter: I thought it was Backblaze that was causing it.
Peter: I thought it might be Sophos or Sentinel One.
Peter: 100% now it was OneDrive.
Peter: The OneDrive client for Mac was causing this.
Peter: And today it happened again.
Peter: I tried to download or tried to open something in a browser.
Peter: And it was just like hanging beach balling.
Peter: I went up to the OneDrive icon, said quit, boom, instantly unfroze and did what I wanted to do.
Peter: And I’m thinking, you know, most of the files that I’m having this problem happening with are not work-related OneDrive.
Peter: It’s all my personal stuff.
Peter: And now I pay for the two terabyte iCloud subscription.
Peter: So I’m tempted just to move all my personal files out of OneDrive over to iCloud and let that problem just go away.
Scott: Yeah, I will say that the only magic remote cloud thing that looks like a folder on my Mac that I have is I have iCloud.
Scott: That’s it.
Scott: I don’t have, and that’s built into the OS.
Scott: So I don’t have, I don’t, I never did that with Dropbox, or I did originally a long time ago, and then I stopped doing that because I heard that they were basically rooting your Mac.
Peter: Yes.
Scott: And I never did it with OneDrive, even though I use, I have stuff in OneDrive, I don’t use that.
Scott: And so, yeah, I’ve never, I’ve never done that.
Scott: I’ve always figured it was a bad idea to let those guys try to worm their way into my finder.
Peter: And like, can’t you just go?
Peter: It’s so typical, though.
Peter: It’s like, it shouldn’t be that hard to just periodically pull this folder, send an update.
Scott: Well, I guess the other thing is, when you do that on Windows, there’s no performance issues.
Peter: Huh.
Peter: Why is that, you suppose?
Scott: I don’t know.
Scott: Do you think it’s because Microsoft owns Windows?
Peter: No, it couldn’t be.
Scott: Do you think it’s because they own OneDrive?
Peter: No.
Peter: There’s no correlation between a company owning two technologies and its stuff working better or worse.
Peter: See also AirPods.
Scott: Well, I was going to say, in the case of Apple, we know that to be 100% true.
Scott: See?
Peter: That’s what I’m saying.
Peter: You’re right.
Scott: Peter, how do we tell the loser, I mean, how do we tell the listener?
Peter: I’ll tell the losers how to find us.
Peter: Dear losers.
Peter: How do we do it when we get lost?
Peter: You’ve already…
Peter: Dear listeners.
Peter: Listener, I’m thinking, loser is just a contraction of listeners and users.
Peter: Dear losers, I will now tell our tens of losers.
Scott: Oh, if only we had tens.
Scott: Oh, if only, Peter, you are an optimist.
Peter: Tens of losers.
Peter: If you want to find us, you have already found us at friendswithbrews.com.
Peter: Well, maybe you didn’t.
Peter: Maybe you found a podcast feed.
Peter: Maybe a friend sent you a link.
Peter: Maybe a fellow loser sent you a link to our podcast.
Peter: But you can find us at friendswithbrews.com.
Peter: From there, there are links to Scott’s personal website, my personal website, our social media handles, active or inactive, you know, otherwise, whatever status.
Peter: And yeah, that’s how you can find us.
Peter: You can send us feedback.
Peter: You might be able to hit us up on the socials from time to time, depending on our moods.
Scott: I will see if somebody says something to me.
Scott: I won’t get a notification, but next time I check, I will see that somebody did that.
Peter: Yeah.
Peter: So, you know, so yeah, send us a note, give us some feedback.
Peter: We love it, especially we love it.
Peter: We, we, the two of us love it when you criticize Scott and point out all the things that he said wrong on the podcast.
Scott: Yep.
Scott: Yep.
Peter: Yep.
Peter: And we absolutely love it when you tell me, you know, sorry, when you tell us how Peter was right about everything.
Scott: So yeah, yeah, I think so.
Peter: Absolutely.
Peter: Yep.
Peter: And if this makes it past the edit, it’ll be a miracle.
Scott: I think that we should call our listeners bruisers.
Peter: There you go.
Peter: That’s, you know what?
Peter: That’s the best thing to come out of this episode.
Scott: Yeah.
Peter: Bruisers.
Peter: That’s our show title.
Peter: Gemini and Claude make bruisers.
Scott: Make bruisers.
Peter: Cool.
Peter: On that note, you know what to do.
Scott: Tell your friends.