The B2B Podcast Index
Index
All categories
MarketingSalesSaaSFinanceHROpsLeadershipCustomer SuccessAI & DataProductStartups & FoundersRevOpsEngineering & DevTools
MethodologySubmit
Best of:MarketingSalesSaaSFinanceHROpsLeadershipCustomer SuccessAI & DataProductStartups & FoundersRevOpsEngineering & DevTools
An independent project byFame
SearchBest episodesGuestsInsightsMethodologySubmit a podcast
Index/Engineering & DevTools/DevOps and Docker Talk: Cloud Native Interviews and Tooling
DevOps and Docker Talk: Cloud Native Interviews and Tooling artwork

AI Wins and Misses for 2025

DevOps and Docker Talk: Cloud Native Interviews and Tooling · 2026-02-17 · 1h 15m

0:00--:--

Key moments - from our scoring

Substance score

60 / 100

Five dimensions, 20 points each

Insight Density12 / 20
Originality10 / 20
Guest Caliber14 / 20
Specificity & Evidence13 / 20
Conversational Craft11 / 20

In this year-end retrospective, the hosts break down what actually mattered in AI tooling for 2025 rather than headline-grabbing model releases. The conversation centers on practical developer experience across the stack: terminal emulators (Ghostty vs. iTerm), CLI agents (Claude Code vs. OpenCode vs. Cursor CLI), IDE integration, and browser-based AI implementations. Viktor Farcic makes a strong case for Claude Code's supremacy in terminal workflows, emphasizing that skills (prompts combined with MCP tools) and agent patterns matter more than incremental model improvements. Nirmal raises AWS Bedrock as an option, while Bret Fisher advocates learning your available tool deeply rather than chasing the next release. The discussion reveals a critical insight: most developers don't choose their tools - enterprises do - making mastery of whatever you have more valuable than hunting for marginal model gains. They also spotlight Ghostty as a modern terminal replacement that 'just works' without complexity, contrasting it with iTerm's accumulated cruft from 20 years of feature additions.

Key takeaways

  • →Claude Code's terminal interface dominates not because of model superiority but because of better MCP integration, skills support, and terminal-native workflow matching developer mental models.
  • →The combination of tooling quality, context management, and MCP configuration matters more than raw model performance - a mediocre tool with an excellent model still underperforms excellent tooling with a decent model.
  • →Most enterprise developers lack tool choice and must maximize whatever platform they're allocated (GitHub Copilot, AWS Bedrock, Cursor), making deep expertise in that tool's patterns more valuable than chasing new releases.
  • →Ghostty proves that modern terminal emulators don't need bloat - TOML config, nerd font support out of the box, and fast rendering beat iTerm's 20-year accumulated complexity without sacrificing functionality.
  • →Model release timing and hype cycles (Opus 4.5, Gemini 3) drive productivity gains less than understanding how to use skills, agents, and MCP connections within your chosen tool.

Guests

Viktor FarcicNirmal (AWS)

Topics in this episode

Claude CodeAWS BedrockModel Context Protocol (MCP)GitHub CopilotOpenCodeCursor CLIGhostty terminaliTermWarp terminalSkills in Claude Code

Questions this episode answers

What is Claude Code and why does Viktor prefer it over OpenCode and Cursor CLI?

Claude Code is an Anthropic-based terminal CLI agent that Viktor argues surpasses alternatives because it works reliably, shows MCP connection status, supports skills (prompts combined with MCP tools that trigger via natural language), and aligns with his terminal-first mental model without requiring configuration workarounds that other tools demand.

What are skills in Claude Code and how do they differ from slash commands?

Skills are prompts stored in directories that combine context and scripts, triggering both through chat interaction. Unlike slash commands (which require explicit `/` syntax), skills trigger contextually - you can say 'create a PR' and Claude Code automatically invokes matching MCP tools and skill descriptions as system prompts, treating them as available tools.

Why do most enterprise developers use tools they don't personally prefer?

Enterprise tool choices are vendor-negotiated purchases (like GitHub Copilot licensing or AWS Bedrock contracts), not individual selections. Developers reported unanimously that they use what their company pays for rather than personal preference, making mastery of allocated tools more practical than searching for alternatives.

What is Ghostty and why is it better than iTerm?

Ghostty is a modern terminal emulator built from scratch by HashiCorp's founder with no accumulated legacy complexity. It works out-of-the-box with nerd fonts, supports italics and true color natively, uses simple TOML configuration, and renders faster than iTerm - which carries 20 years of feature cruft that makes simple changes (font, theme) unintuitive.

Does model quality matter more than tooling when choosing an AI CLI agent?

No - Viktor and Bret argue that excellent tooling with a mid-tier model (Claude Code with Haiku) outperforms mediocre tooling with a stronger model (Gemini 3 CLI with Gemini 3). The combination of MCP integration, agent patterns, skills, and UX flow determines productivity more than raw model performance.

What our scoring noted

Our reviewer’s read on each dimension, with quotes from the episode.

Insight Density

12 / 20

The episode covers specific tooling choices (Claude Code, Ghostty, GitHub Copilot, Kiro) with some reasoning about why each tool matters, but much of the discussion involves personal preference and repetitive points about model quality vs. tooling design. Several sections devolve into tangents about browsers and ads that feel padding-like rather than densely packed insights.

It's not only about models, right. you can easily have a combination. And I, I'm inventing now, so don't take this as me saying it. Kind of, you can say that, Hey, this is, I've never seen something as good as this. No, it just works.
the quality of models, when they're released, they're overhyped to begin with. Yeah. But also that people forget that it's not only about models, right.

Originality

10 / 20

The core insight that tooling and context matter more than model performance alone is somewhat fresh, but the episode largely recycles familiar comparisons between popular AI coding tools (Claude Code vs. GitHub Copilot vs. Cursor). The discussion of AI in browsers rehashes well-known security concerns. Few genuinely counterintuitive or novel frameworks emerge.

Claude Code with Haiku is better combination than, than the one I mentioned earlier. Right. not because models are better or worse, but because it's a combination of things, right?
the only VE left for those companies to finance themselves? Well, selling you Red Bull.

Guest Caliber

14 / 20

Viktor Farcic is a recognized DevOps figure with credibility in the Kubernetes/cloud-native space (founder of Upbound). Nirmal is a Principal Solution Architect at AWS with hands-on experience. Bret Fisher is an established Docker/DevOps educator. All three are practitioners with real experience, though the episode functions more as peer discussion than guest interview format. No career podcast guests or pure thought-leaders are present.

I'm a principal solution architect at AWS
Viktor Farcic of Upbound

Specificity & Evidence

13 / 20

The episode names specific tools (Claude Code, OpenCode, GitHub Copilot, Ghostty, Kiro CLI, Code Rabbit MCP) and tools tried/abandoned, but provides limited hard metrics. A few concrete details emerge: $20/month Claude Code subscription, Notion's claim that 50% revenue is now AI, one sentence about a 50-person hackathon in Kiro. Most claims about tool superiority lack quantified evidence - assertions about productivity and quality rely on subjective assessment rather than benchmarks.

I pay for Claude code. I pay $20 a month for Claude code.
they kind of gave away the fact that 50% of their revenue is now AI

Conversational Craft

11 / 20

The hosts ask follow-up questions and occasionally push back (e.g., Viktor challenging Bret's preference for Warp Terminal, or skepticism about AI browsers), but many exchanges lack sharp probing. Questions tend to be open-ended invitations rather than provocative challenges. Some tangents (e.g., iTerm settings jokes, ARC browser nostalgia) meander. Productive disagreement surfaces sporadically but isn't systematic.

Viktor, it's not that Warp's bad, it's just very, it's in a very competitive landscape
I want to push back on your comment. If you're young, I feel completely opposite opinion.

Conversation analysis

Computed from the transcript - who did the talking, and the words that came up most.

Most-used words

terminal52code50browser41tools40claude37open29viktor27different22iterm22back22models21first19model19feel18features18better18

Episode notes

I'm joined by Nirmal Mehta of AWS and Viktor Farcic from Upbound, to go through our 2025 year in review. We look into the AI tools that consumed us this year, from CLI agents to terminal emulators, IDEs, AI browsers - what worked, what flopped, what's worth your time and money, and what we think isn't! Check out the video podcast version here: The Agentic DevOps Guild has launched! It's a training + community + mentorship program for engineers wanting to learn the latest CI/CD automation and dive into Agentic DevOps. Meetups are happening now, with new course videos dropping every few weeks. Join the Guild and become your team's leader in AI for infrastructure automation Creators & Guests Cristi Cotovan - Editor Bret Fisher - Host Beth Fisher - Producer Nirmal Mehta - Host Viktor Farcic - Guest You can also support this podcast by subscribing to my YouTube channel and my weekly newsletter at bret.news ! Grab the best coupons for my Docker and Kubernetes courses . Join my cloud native DevOps community on Discord .

Full transcript

1h 15m

Transcribed and scored by The B2B Podcast Index.

Welcome to our 2025 year in review episode. I normally have these in year episodes where we try to wrap up what mattered to us this year and but this year felt different. I think it's safe to say that we are consumed all of us by AI agents everywhere in all of our tools and. I kind of knew that that would dominate any retrospect conversation.

So we, we did that even after we recorded this episode in early December. The landscape shifted under our feet as Anthropic release their Opus 4.5, along with Google's Gemini three days earlier. And these models caused a lot of my friends and tech leaders to spend their holidays, vibe coding on personal projects.

That were quite easier to accomplish than with the models available just three months ago. So in December, we recorded this year end episode. I'm joined by my co-host, Nirmal of AWS, and we're also joined by our special guest, Viktor Farcic of Upbound, to discuss everything we loved and hated about cloud native and AI in 2025 from terminal emulators to IDs from CLI agents. To AI browsers we're breaking down what worked for us, what flopped, and what's worth our time and money, and maybe yours.

We kick things off in this episode in the trenches of the command line where the battle between Claude Code, OpenCode. GitHub Copilot and all the other CLI tools rages on. And Viktor, you know, he makes his passionate case for why Claude Code dominates the terminal space. While I am mostly focused on GitHub Copilot, mostly because I'm making courses on that stuff, and Viktor is concerned from my mental health about that.

So we talk and discuss the PRs and cons. Nirmal talks about AWS Kiro and brings that into the mix. We don't just talk about CLI tools, although, you know, that really I feel like has dominated the last few months of conversations everywhere on the internet. we talk about Ghostty and terminals versus my favorite iTerm and we discuss warp terminal and how I spent some months on that and then kind of left it.

we talk all the way up the stack to IDs and into. AI browsers and why I am just very uncomfortable with any AI in my browser, at least the way that it's being implemented in all of them today, and a whole lot of other topics. So I really hope you enjoyed this episode with Normal and Viktor, and let's get into it. Thanks for being here, everyone.

yeah. So I'm just glad to see you again, Bret. It's been a hot minute. So, excited to kind of get into this end of the year, but before we start, just want to remind folks, all the opinions on the show for me are my own and not representative of Amazon Web Services, my employer.

That being said, I am a principal solution architect at AWS and, we'll probably going to get into some AWS services today. Potentially. Potentially. Maybe, maybe.

So, this show is going to be us bringing on our special guest, Viktor, and we are going to walk through the best and worst of AI tech, at least in our, our views for 2025 and then into 2026. Viktor, welcome. Hey, hey, hey. I have a disclaimer as well.

I'm probably going to say something ridiculous that is going to offend people and opinions are mine and only mine and whatever else people say in those cases, I don't care. Yes, opinions are always mine. I only say other people's opinions when they pay me to. So far, so far no one has, it's still yours successfully paid me to.

Oh no. They're always yours and they're great opinions and that's why people come. Mm-hmm. Back and back to hear us espouse our beliefs.

I influenced, I am influenced by money though. So that's my disclaimer is money will influence my decisions. Yes. which I will mention disclaimers today, when, when we get into it.

But like, I guess one I should say is GitHub provides me free Copilot, so that influenced my decision. Well wait until we get to the worst heck of the year. That's right. So, the whole point of this game, and it is a game because we could make, we could say this like, you, for every one of these tip picks, we're probably going to have someone has a, a complete opposite argument that's is just as valid.

So for whatever I love and whatever they hate, like someone else is going to think love and hate is different and that is perfectly fine. but I think like in starting in planning the show, it was, I was realizing it has been an incredibly long year. And when I think back to this just the first quarter, like what we were talking about, which is, and what we were doing versus what we're talking about and doing now is like, it's, to me, it's one of the fastest paced years in tech of my career where so many of my tools have, my behavior in using them has changed.

what we're going to talk about, tools that we tried to adopt, maybe we had to roll back, maybe, you know, failures. everything is up for grabs right now. I feel like, whether it's terminals or browsers or IDs or note takers or it's just, an onslaught of innovation and I'm kind of here for it, but also like we all just need to get stuff done. So I'm excited to hear what you all think.

I ordered, I created a mental list. Talking about the, so the theme is it's got to have some sort of AI related, which is not too hard. I, I, there's very few tools that don't have some sort of AI connection anymore. The first topic, I think, to get out of the way is CLI tools.

we're going to go to the nerdy stuff first and maybe like zoom out. Maybe that's the theme as we slowly back out. We'll, we'll end with browsers, maybe something, more consumer cited. But, you know, the rest of the world's all.

Thinking about ChatGPT and Grok and the consumer side of things. those of us that are developers and operators, like we're more and more in the terminal. I feel like this was the year that AiTerminal tools grew up. I don't know about you, but my first experience with an AI tool was when iTerm added, like three years ago, they added a ChatGPT similar experience in the iTerm, like in the actual ui, not, not a Shell CLI, but like in the UI of the terminal.

So it was kind of weird and I used it for like five seconds. Five minutes. And then I was like, this is dumb. I, I would just rather use a browser.

iTerm definitely needs AI for their settings because no human can understand how to computer. Oh, I'm so surprised they haven't done that. Exactly. Because kind of we give up on this, humanity gave up on this.

So let's put some ai, maybe it'll figure it out. Oh, that is, and whatever it does, it won't do the worst job than you will you, you would do, right? That is, that is gold Viktor, because I feel so validated that I, I, every time I open up iTerm settings, I feel like I've lost my understanding of it. Like completely every time.

Can I do a, every time I try to change a font in iTerm, like I, I don't know where to go. Can I do a non-AI pick? Just because we're talking about it? Yeah, sure.

If it's on, if it's on topic. Pick ghostty. It's absolutely awesome, it simply doesn't do anything special. Nothing.

There is nothing you say, Hey, this is, I've never seen something as good as this. No, it just works. So what is it? Explain what what it is and choose.

It's just a replacement for iTerm done by Hashimoto. You know, the Hashi HashiCorp guy? it's his, pet project, right? And even if you go to the repo, it says kind of like, this is my kind of, don't complain to me.

This is my pet project, right? So, it's just awesome. Why, why should I switch to it? Why?

What's so awesome about it? That's the thing. I cannot give you the answer to that question because there is nothing special about it. It's just fes.

I never had any trouble, I never had issue kind of how to change the theme, how to, change the font. It's just simple works. And that's it. That's awesome.

I heard it's fast, but I don't know if that's Yeah, it's lean. That's true. It's, it's, it's, it's main call, main claim to fame is speed. like, like rendering speed.

my biggest thing is it, unless it's been fixed, it doesn't have control f so I can't search the window. like a true Unix person's attitude is you don't get to search the window. that's not how terminals work. You have to grep right?

You have to re you have to put that log into a filter. I have just used control F for searching the window for decades and I, it's like a muscle memory. But I will switch as soon as that feature is available. 'cause it is a great terminal.

it is very lean and streamlined. You know what it is I, what I love about it is that. it's built now. Like iTerm shows its age because it was built over the last at, I mean, I was, I've been using it since 2009, so iTerm's probably 20 years old.

it's got every feature be, you know, because we didn't have those features, we didn't have all a true color 15 years ago. We didn't have the ability to put italics very easily 15 years ago. Like, you had to do a lot of hacks to get italics to even work in your terminal. So Ghostty just knows all that is modern and it just works out of the box.

Like there's no special config, it supports nerd fonts out of the box. it understands all these modern terminal concepts. iTerm just has this horrible history. It kinda reminds me of, what's the SSH Windows CLIent?

putty, it reminds me of like putty or, or old school. Remember we used to have like x these old school windows terminal emulators for Linux and Unix that were like, you could decide between which formats of like Unix or X term, or u like all these different terminal settings. Like they were just old school stuff. And Ghostty throws all that away, says, Hey, you got a, what is it, a YAML file or TOML or something?

yeah, TOML. Yeah, it is great. Nerd phone is, is actually the reason why I switched to it. I, I wanted to change the font in my, iTerm I ended up having to Google for more than five seconds and I said, no, no.

If I'm going to Google for solution, I'm going to Google for replacement for iTerm. That's how it started. Yeah. So speaking of terminals, this is a great segue.

so now you've got Ghostty running. Super fast, super slick. Mm-hmm. what's your favorite AI tool in the terminal from 2025?

Come on. Before, before he answers that, I did want to mention that if you're interested in terminals, Bret Fisher dot com, put this link in Bret Fisher dot com slash shell. It is my guide. It was updated this year.

it was actually in the last month. It's my guide to terminals across OSS add-ins, plugins, fonts, how to get italics working if it doesn't work. what do you use? Do you use zsh?

Do you need o Oh, my ux? Do you need Zelig? Do you like, there's so many. These are not mostly AI things.

Very few of it's AI related, but for 10 years now. I've been keeping this thing up to date. I updated every couple of years. It just got updated.

If you're a veteran to terminals, you've been using 'em for years and you know you and love your terminal, you will not need this. But if you're newer to terminals or you're learning how to advance your terminal game, if you care about the GitHub monospace font, which is my favorite fonts, a terminal Terminal, which is a replacement for tux, which uses session list UDP packets, like there's so much in here that over the years of consulting and stuff like that. So selfish plug.

There you go. Okay. What was the question? Sorry.

Let's back to the question. So now that we have this awesome terminal and I can plus one plus a thousand breadths, shell page, it's what I use as my guide to set up my shell, including the awesome demo modes that you've put in there for you know, I do a lot of presentations to customers and on stages and workshops and having that demo mode so that you're not like showcasing your whole entire right from your, from your shell, is really handy. Okay. So Viktor?

Mm-hmm. We've got Ghostty running and I just installed it while you were lighting the awesomeness of it. I've got it up and running and it boot it. One thing I already like is that you can resize the window any way you want, which is amazing.

Yeah. You can create paints, you can split planes, you can do everything. Awesome. That makes sense.

What is your favorite Ai CLI tool of 2025? Oh, there is no competition there. It's Claude Code. And I will, I dare anybody when it's terminal based.

Terminal based. I dare anybody to fight against it with me. now the, the note there is that if you're using Anthropic models, just to be clear, so if you're looking for something that is model agnostic, OpenCode is probably the. Least bad option.

but if you're okay with a Anthropic or Sorry, not Anthropic, provider and Anthropic models. It can work with Bedrock, for example. That's fine. Yeah.

now if you're not fond of a Anthropic models, then we are going to have another fight afterwards. Tell us, I mean, I, I don't think I have a debate about Anthropic models. I might have a debate about Claude Code, but what is it about Claude Code, specifically the terminal component that just beats the pants, beats the socks off of anything else? Have you tried all the, of, have you tried other tools?

Yeah, I tried OpenCode. I tried the, what is the charm thingy? the charm thing. Yeah.

The, you know, the folks behind Charm has the, have the Trivy of themselves. I tried go probably like 10 of them, and I spent, I spent this a week with each of them. and first of all, Claude Code just works. It works, right?

And all the things that you might need, assuming that you like terminals, but some people don't. It's simply there. So here's a simple example, and that might have changed, like in OpenCode, which is my close second. I cannot see the status of my mcps kind of, I don't know whether they're, they're connected and running, right?

Kind of very basic kind of, you know, table stakes. Yeah. I don't want to even discuss the usefulness of something like that. Or if I stick with the same team, I try the Cursor CLI.

And after not being able to configure, MCP may, it might, things might have changed. Just to be clear, it's at a time. I tried it after failing to configure cps. I, I googled it, went to the site and it says, oh, it's very easy to configure CPS in, in cursor, CLI, you open Cursor IDE configure it there, then close it and open the CLI and it works.

Right. Come on. then, then there are adding features that nobody else has, right? You might like, or you might dislike skills, but skills are there.

can you describe what skills are, what are, what are those? Skills are like prompts or a directory with prompts And scripts and what not. It's very simple, very, very similar to slash commands. The major difference is that, you can trigger those prompts, let's say, or commands through chat with your ai.

Right? So typically we need to execute slash something, right? Yeah. Here you can say model slash CP slash usage.

Exactly. And here you can say, hey, I don't know, like, create a PR, right? And just as it would trigger MCP, if there is a description of MCP tool that matches that it would trigger skill as well. So think of it like prompts that have the context or description, that is loaded as a system prompt so that AI treats it as a tool as well.

Got it. So it can like skills and it's not just. About itself, right? It's not just the slash commands to run Claude Code CLI, it could be anything.

Yeah. it's a combination of prompts with an MCP tools, right. Got it. And I think it got open source today.

I think the skills maybe I'm, I heard today that, OpenAI seems to have implemented skills as well. Yeah. so agent skills are open source as of today, I think. Open standard.

Open standard. There we go. there's a skilled directory, under Claude dot com slash connectors. Yeah.

I think is what you're talking about. Yeah. And now, you know, Bret, you need to stop me. I can, I can go on for 45 minutes now about code.

Well, I, I. It's apparent to me that you, you fall into a group of my friends that are such Claude Code advocates that mm-hmm. It makes me start to realize that the real secret in the sauce of whether it's OpenCode, which I've heard a lot about, which, but I've never personally used, is like we've, the internet tends to focus so much on the model's performance, but it, I feel like we're at a point where it matters more about how much of the tool you're using and your patterns around how you, whether it's skills or it's commands or it's, you know, heavily using the agent's file or whatever that agent's variation is, because they all still haven't quite settled on exactly.

Like, not everybody is all using Agents md. Mm-hmm. And what if I want a bunch of different files in different directories? Like how there, there's a lot to all of this.

What I would call like automation and context enhancement stuff, which is Outside of the model, sort of independent of the model. And, there was a lot of talk earlier in the year, especially first quarter when I was first discovering like local models and experimenting with Qwen and like some of these local things. And then I realized how inferior they are to the, to the, to the SaaS models. So I just stopped using them.

I mean, I, I, I can afford the $20 a month plans on the internet, so I, I choose to do that rather than suffer with the slower for performance local ones. So you're doing option, not everybody gets that you're doing option, but Oh, you're doing it completely wrong. You can easily avoid $20 or a hundred dollars a month subscription to something like on Anthropic, right? You just need to buy, 20 K hardware and run the latest model over there.

It'll take its problem solved a year before it gets obsolete, but it'll work well for a year and you will, you will not have to pay subscription. Yeah. Not to mention the electricity cost. Right.

So for the benefit of all that money gone from your wallet, you don't have to send your tokens to the internet. Yes. yeah. to me the takeaway here is whether you choose, whether you like Gemini or, and, and to be fair, I'm going to lump developers into two camps right now.

and that is the ones that get to choose their tools, which the three of us, especially Viktor and I, being a, a little bit more independent than Nirmal, we get to pick whatever we want. We get to pay for whatever we want. and so we get to choose what we like. Most people in enterprise positions I talk to do not get that choice.

In fact, I run a DevOps Guild meetup every month. We call high fives. four people. A whole four, showed up yesterday, so I surveyed all four.

So the, it was a universal, a hundred percent of those four all said, I like this, but I use this because it's what my company pays for. And that might be that they paid a lot of them just use GitHub and copilot because that's what their company pays for. They would talk about, like, my last job paid for cursor, so we all use that, but my new job doesn't. and the same with models, right?

Like models and ides, they pay. so I would say that a lot of people like. Your, maybe your choices aren't as infinite as Viktor's in, in mine, but, if they aren't, then your, your angle of attack is to learn the crap out of that tool so that you can take advantage of all of its specialty features. Because I think that's just as good as I, I was worried about, Gemini for a second.

Like, Gemini comes out, oh, everybody's telling me it's awesome. So I switched to the Gemini, CLI, and I switch everything. and then I'm like, you know what I mean? I've lost all my, a lot of my advantages.

So why do I care about the little edge it's given me? Because in five minutes and there's going to be a new Claude model that's going to upend it, and I can just stick with one, maybe one CLI and just wait for the model. Like, anyway, that's my theory, you know? I think that the quality of models, when they're released, they're overhyped to begin with.

Yeah. But also that people forget that it's not only about models, right. you can easily have a combination. And I, I'm inventing now, so don't take this as me saying it.

Kind of, you can say that, Hey, Gemini three as a model works better than something else. But then Gemini three with Gemini CLI, is still inferior to, let's say Claude Code with, Sonnet. Right? Right.

I, I would go as far as to say that Claude Code with Haiku is better combination than, than the one I mentioned earlier. Right. not because models are better or worse, but because it's a combination of things, right? Yeah.

Yeah, yeah. I mean, I agree. I think it's about what is actually enhancing your productivity and getting out of your way. so there is a floor to the model, right?

So like you were talking about using some of the local models. If you're getting. Crap tokens back and it's taking you minutes to get some output from your local model. That's the floor, right?

Like, like these foundational models from Anthropic or, you know, DeepSeek or others, Gemini. they're going to beat the pants off of what you can probably run on your local machine for general, like, just a general statement. maybe specific models for specific tasks that you're doing of course could be much better. But general statement, there's a floor to the model quality.

But back to what Viktor just said, and I could, I'm in in violent agreement with that, the tooling around that. Like we just talked about skills for two seconds. That seems like. A very valuable thing to get out of your way to get the most out of the model that you have access to.

And so even if you had access to Gemini but the tooling doesn't help you get out, you know, get the most out of it, then like, does it really matter how good the model is? I don't think so. Yeah, exactly. yeah, and I think that's really where people kind of get hung up right now is like chasing that incremental quality of the model versus the patterns of using these tools.

And the orientation of these tools toward productivity. Right. And I, it seems like Viktor, like what is it about Claude Code that matches your brain, your mental model of like how you want to work with a model to enhance, like development or whatever you, you know, what tasks you're doing with it. there are two answers to that.

First would be how, you know, or question that you should have asked first is terminal versus IDE. because I think that without entering into what is better, because it'll depend, but they change the way how we operate those things, right. Aren't you using both in some kind of way? I am using both but separately.

Okay. so I do have, IDE open that serves similar purpose, like Vim to be honest. but I might occasionally open a file. Literally command p open a file, that's it.

No, not even three structure or anything like that. But in my case, I feel much more comfortable with terminals simply because what they do is. I treat it almost like a person, right? I talk to it.

I, I don't need other distractions, right? I, I, I don't need the list of files. I don't need, to be clicking here and there. I don't want to click any buttons and I don't want to spend time learning all the shortcuts.

it's, it's simply there. And I think that it would be in a very similar question, like if I would ask you, forget about AI for a second, or agents, right? Do you use terminal separately from your IDE or do you use terminal inside IDE? And there is no wrong answer there, right?

But I'm, I'm, I'm a separate terminal guy. simply, it somehow feels better. and, I'm used to never touch the mouse. Or it's not, never touch the mouse, but I rarely touch the mouse, right?

it just feels so much easier, more intuitive, and so on and so forth. Now, I'm not saying that those features don't exist. When you use agent in IDE, it just somehow feels better. it's about feelings.

I don't know how to, there are some cases where I cannot give you tangible measurement. oh, I'm 20% more pro productive in Claude Code than, than in, I don't know, Cursor or whatever You hear. We heard it here first. Viktor is 20% more productive with Claude Code.

by the way, Eric Smalling's opinion, My favorite tooling is whatever Viktor's last show is about, then I forget about it and just use Claude Code. So his default is Claude Code, but he's willing to try whatever Viktor suggests. Yeah. It sounds like the winner here is the term CLI.

Winner is Ghostty plus Claude Code or OpenCode. I'm So if you can't use Claude Code, 'cause maybe your company doesn't pay for, Anthropic models, then maybe your backup plan is to use OpenCode. I don't know this for a fact, but my assumption is as much hype as I hear from open source terminal shell advocates about OpenCode, I assume it's good enough to be competitive with like Gemini, CLI or copilot, CLI or name me another model, CLI that just comes from Yeah, OpenCode is, or Codex CLI, in opinion better than any other OpenCode is the top the best, in my opinion.

Terminal based agent you can imagine. Except Claude Code that is in a different category. So it's kind of like, you know, we cannot compare soccer and basketball kind of like there, or first and second league of soccer or whatever. Kind of, if you exclude Claude Code a go with do and code, it's awesome.

Absolutely awesome as long as you don't compare it with Claude Code. So, so Bret, what's your favorite AI CLI tool? well, I'm basically agreeing with all that. I feel like I started the year at iTerm, I ended the year in iTerm, I'm not going to debate Ghostty.

I really just have one feature, muscle memory thing with the Control F and they had some other issues with that were minor issues in the ui, but those have all been fixed throughout the year. so I keep it installed. I switch back and forth every once in a while to see, and Ghostty is fine until I immediately need to search something on, like when I'm doing serious work at the shell. Probably not in my AI CLI, I'm just in the shell and I want to search now.

I will say I did, if you, if you watch videos or the live streams of us over the year, you will hear me for like three months. Not shut up about Warp Terminal 2.0 when that was released in June. That made waves and I gave another shot.

'cause I did, I gave Warp 1.0 some time. It wasn't like it was around for years, it was fast, because I think they were doing rust and claiming it was faster than other terminals. It was fast, but it feels like a young person's tool because I am 30 years of muscle memory into how pseudo terminals work.

And it is the opposite of that. And I appreciate their revising and they talk about it. They're re-envisioning the CLI, what does a terminal, what does a shell behave like? And the problem is, is the advantages don't for me outweigh the muscle memory and the efficiency of what I've already done.

So, because I know, you know ZSH idioms, I know shell shortcuts. I know all the little, you know the control. SI know you know how to command completion. I know all these things.

And Warp throws all that out the window and says, no, no. Rethink everything. and it does make more sense if you don't know the history of how terminals have, I've always had to work since green screens in the seventies, of which, that's where I came from, right? Like, maybe not the seventies, but I was using, my first job in the US Navy was managing green screens and trouble, and I actually had to repair them.

So I think that my old school muscle memory is actually preventing me from enjoying Warp terminal two better. for me warp two would be great if you're young. If you're learning terminals now, and you just want to try something new and fresh. Great idea.

Give it a shot. The one thing that broke it for me was that. The reason I had to leave Warp was because command completion for them is they break all ZSH command completion. So the only way it works in Warp is they have to add manually c command completion for each tool, which to me, one isn't sustainable.

Like there's a, there's a thousand tools, and two, I don't live strictly in a dev world. I live more in a DevOps and ops world. So I have a lot more, I think I have a lot more command line tools than a typical dev might have. You know, like, think of all the, all the, the, the Kubernetes terminal tools we have and all that stuff, and those don't have command completion and warp, which means I'm forced to use two terminals because I, I, I totally depend on command completion.

AI command completion doesn't really work well for a lot of those tools yet. And so then I have to go the, and then the minute I have to have my old terminal and the new terminal up, I want to throw one away and I throw away the new one. Yeah. Right.

Like, so. That defeats the whole entire purpose. Yeah. Viktor, it's not that Warp's bad, it's just very, it's in a very competitive landscape and I, I don't know that it has the edge it needs to sustain itself, but good luck to them.

I'm excited for their product. Viktor, go. So I want to push back on your comment. If you're young, I feel completely opposite opinion.

Okay. If you're young, that's the reason not to use it, because then you will never learn. Mm-hmm. You'll learn when it doesn't work anymore.

Kind of, you'll learn it basic. There are some basics that you need to know, no matter which technology we use, no matter how we operate, the real systems are not running work terminal. Kind of like you will not know what to do when something happens. Yeah.

Please don't. Okay. Now, if you're older, said, okay, I did my dues, I, I want to switch to something more productive and Warp works for you. Great.

Amazing. Do it. But not at the expense of never figuring out how to actually work. I mean, that's fundamental Yeah.

To all of these tools, right? I mean, this is, yeah. This is the essence of, of the challenge that college professors are going through right now with their courses, right? Like all these tools just do, and they hide what they're doing if you don't, and they're wrong, right?

Yeah. And so, there's a cliff here for, with all these things where regardless of Claude code or, or Warp or Copilot, you do have to take them a moment to understand what it's doing. And I'm, I'm assuming y'all are, are you all running your AI tools in terminal, in YOLO mode where they just have rampant authority to run any script or modify any file? Never.

No, never. I'm seeing listen to the container. Not unless it's in a container, but I don't normally develop do it in a container yet. Kyle Quest, by the way, is pointing out that war's really great for shell stuff with DevOps.

He likes it and ha and does the agent stuff and container tasking. Correct? I do agree. I love that it has native notifications and that it has a little panel that you can keep track of.

All the agents working in the background. that is neat stuff, but it's not enough of advantage over a typical multi-terminal. Like I'm an iTerm with three different windows and five different tabs and like I've, I've got my setup, I've got my process there and it was to revise myself. I completely agree with everything Viktor said.

It does not, subs warp is not a replacement for learning other. Foundational shell tooling. For me, it was about, mostly about iTerms like the way that I open tabs and shells and windows and I, I actually use, you know, command I to label terminals and I colorize them and I do all sorts of things and I have different profiles and all that stuff you're complaining about in the settings. I really heavily use all that stuff 'cause I've been using it for over 15 years.

So it's that part. That's why it's going to be hard for me to shift a Ghostty or whatever because I will lose a lot of my preferences. And the way that I, the way that I use the graphic interface of the iTerm that more so than warp. And warp is just, I feel like too, too young it has its set way, it's very opinionated and you, it's hard to deviate outside of that.

And that's what is hard for me to really commit to long term. So to me it doesn't feel like terminal, to be honest. Yeah, it feels like IDE, and nothing against IDEs here. Right.

But it doesn't feel like terminal. Kind of like I have problem with, with wasted space over there. I don't see the information I need because everything occupies so much space. Yeah.

The default format isn't compact. That's an's a setting, but it's like, it's not as good as Yeah, exactly. It's, it's very opinionated in a way that it's going for readability, not density of, of content, which can then mean your screens don't show as much. And that also was frustrating for me.

I will put in a mention of goose. Like, goose to me was better for people that aren't quite developers. Like maybe they're developer adjacent, like maybe they're people that have developers in their team, but they hear about all this agent stuff, but they're maybe not CLI people. And so they can manage multiple agents through a GUI.

But they don't need to look at, they don't need to stare at code, but maybe they need something other than five different ChatGPT windows open, you know, they need something like that. And that's where I feel like Goose might have an edge anyway. Normal. Tell us what your, what your PIX are.

yeah, so I'm going to pick, and this is, you know, my opinion, I'm going to go a little bit different and say, AWS kiro for both the CLI, formerly known as Amazon QCLI, or QCLI and, the Kiro IDE. not that it's, necessarily like, you know, the best thing since sliced bread, but the idea for the CLI, it's just, I enjoy working with it. It has those slash commands. It.

Gets your MCP, it's just like straightforward. And for the operations of EKS clusters, AWS environments, it, comes with some of those MCP servers if you want out of the box, for AWS documentation for the work I do and the, and the customers I support, it has been great to just be a better operator. And also, you know, I have to understand what my, what my customers are doing and how they're using our tools. And so I live in, in that world.

Of course, it helps that it's available to me as an employee. I do enjoy Claude Code a lot and I. I wish there was like this. I think we'll get there.

But all these cons, like, there's like little pieces of different tools that I've used. I used Cline earlier this year. Love that a lot. and especially to be able to dig in to some of the details in its thinking.

but out of all these tools, Claude Code has a lot of paradigms and then Kiro IDE has like the spec driven development, the idea of really thinking through how you're going to use the tool and the project management. And it feels much more like how you would actually use this in an enterprise. and also it's, or it orients you toward being successful with the output of these LLMs, which I really appreciate. And we can get into spec-driven development and all that a little bit more.

At some point in the future, I wish all of these things, like all the awesome bits from all these tools would just be in one, and there would be just like some kind of, there, this wouldn't even be a debate anymore because we've just like settled on how these tools work and it's, it gets out of your way. But for me, you know, at at Reinvent, I, I helped host a hackathon on a bus where a, I think 50 developers were using Kiro to build apps in six hours. and they didn't really run into many issues, with random teams of people.

They just met getting together and building an app. And I think that's the power of like, the spec driven development. but I'm excited to see where these tools go. I'm of course, hedging my bets here.

I'm excited to see where these tools go when they're bridging. The age agentic, non-deterministic components of, these tools with the deterministic normal code that we're writing. and to me, like the SPECT driven development, there's another paradigm, A-I-D-L-C where you're planning through, you're thinking through, what the project's going to be, you're, you're going through it at, in a task orientation, makes sense in terms of bridging that. And also, I think we're going to start seeing a little bit more formal verification or formal methods of verifying what the output of these tools are.

Are what you asked for. And if you go the spec driven development route, you have a framework for being able to take advantage of that in the future. So that's my answer. so, when you say Kiro IDE, that's web?

No. So there's a Kiro IDE, it's called Kiro and it's its own thing. And then there's Kiro, CLI, which is the, it used to be called QCLI, but it was rebranded to Kiro. Is Kiro IDE just a another vs.

Clone a VS. Code clone? Yes, but it's, there's some. M modifications to it for sure.

Yeah, it looks pretty, it looks heavily modified, but yeah. Yeah. All right. By the way, my pick was actually, my pick isn't what everyone else should use.

It's the picks were like, what I'm using and then what I'm using in right now is Copilot CLI. So, okay. and it's not because I chose it over Claude Code because I thought it was better. It's because it's what is free for me.

And it is, I'm influenced by money and it, I do like that it's integrated with VS Code's IDE. So I can have the conver, I can start a conversation in the CL, I continue that conversation with the context in the IDE and then go back to the CLI like, I like that. so I do that a lot, but that's just me. Can I, can I make a suggestion to you, Bret?

Can you open a GoFundMe? create a GoFundMe. I pay for Claude code. I pay $20 a month for Claude code.

No, no, no, no. You did. Wait, I didn't finish. I mean, I think you need GoFundMe hand fund me because I think that people will churn in because you definitely need money to pay a psychologist after all the drama you were experiencing this year.

Right. So I used copilot, CLI with, yeah. Okay. When you ignite term, I said, okay, you know, things happen, but then you added copilot.

CLI cannot. Definitely that trauma, it's too far needs to be, it's too far. It needs to be resolved because it'll count you for the rest of your life. So.

So it's a quick advice, express cry for help, professional, professional. So we're going to have a private therapy session where Viktor is going to walk me through my horrible Copilot experience and why Claude Code is so much better at it. But that is too long for this podcast, so don't take my advice. I'm just saying it's what I stuck with the most this year over all the things I tried.

I played, I tried codex when suddenly, you know, whatever 5.0 came out. Everyone was like, oh, Codex, Codex CLI, blah, blah, blah. I tried that.

Like, but I'm using very basics. I'm not using all the features and throwing, you know, I'm not using the latest functionality in there. So I, I understand. I'm not necessarily code course, target course you're using basics because you cannot find features in iTerm and compiler doesn't have them.

So of course you are fair, all. So, before's, before we turn this into the only thing we talked about with CLI tools, I'm going to change subjects on us. we touched on ides, so visual editors. Yes.

Another topic. We've got tons of them. Almost every CLI has a companion IDE. I feel like at this point, maybe not Gemini.

I don't know what Well, Google has oh yeah, windsurf now, now, right? Gravity. Oh, no. Anti-gravity.

Gravity. Who's got windsurf now? No anti windsurf. Google.

Google. Also, he has people from Windsurf, not windsurf. Windsurf turned is turning into, well, we think like the winds surf's team is building antigravity or something like that. So who knows?

I tried that for five minutes and went Yeah, it's like all the others. So, yeah. I kind of already answered that, but I would say Kiro. I am productive with it.

I like the spec driven development. I'll say that again. we should have another show, Bret, on like these paradigms, like A-I-D-D-L-C spectrum development. I like, it's, it's just straightforward.

it's easy to get started and use. and I just watched a bunch of developers in a time constrained manner be very productive with it. I can't say a percentage of productivity with it, but you know, that's a good test. and I'm biased because, I get it for free from AWS as an employee.

Alright, so you're saying Kiro. so by the way, with Kiro, CLI, you would be using what terminal before you just installed Ghostty. What was before Ghostty? Just, just terminal?

iTerm. iTerm. Okay. Yeah, I mean, I won't, I will refrain Frain from saying it.

'cause it sounds like you're going to be a convert to Ghostty. So, yeah, I am. I have Ghostty running now for, what, 20 minutes and it's. The, I'm already thinking I'm going to like it.

It is, it is definitely the one for you if you're not Mac too, because like iTerm is only Mac. and true. You know, so you're, and if you don't have high expectations feature wise, just to be fully transparent, right. You will not get iTerm fanciness or anything like that.

It's just a terminal. Yeah. I don't, I don't think I take advantage of all the iTerm fanciness because it's hard to take advantage of all of it to understand it. Back to what you were saying, Viktor, I have plenty of developer friends.

Yeah. I have plenty of developer friends that are terminal people, that are Mac terminal people. So like when I see them running terminal, I lose my shit. Ghostty is for them, right?

Because it's, it's definitely faster than the, my understanding is it's definitely faster than Mac terminal and it has a lot of much nicer built in themes. The ones that are built into the Mac terminal, are mostly garbage. so yeah, it does it, and it supports, like you said, it supports nerd fonts. One of the best things to come out of terminals is nerd fonts in many years, so it's great.

Viktor, what is your IDE of choice? I, I'm not sure that you've me mentioned it. None. None.

Ooh, I, so I, I do have, IDE open, all the time and I, the only thing that I do in that IDE is command P to start typing the file that I want to open. Open it if I want to check it in more detail. Right. I use it for that exclusively.

I don't use IDE for anything else. I'm not saying that they're bad, just can, terminal based agency is very spent. Almost all my time, occasionally look at the file. So if I, whichever choice I would give you, I cannot justify this that is looking at the file is better in one over the other.

Right. So I'm not against them, just I don't use them. So this half of the year, this half of the year. So, so what if you needed to like rename the top of the read Me.

Would you use like Neo Vim, I mean, you're in the shell, would you just use like a, a Vim variant or like, I mean, if you needed a one line edit, you wouldn't ask an AI to do that. Right? would you just open up VS Code for that? Would, do you, would you use Cursor for that?

No. First of all. I would just say I to do it. Let's start with that.

so you're having a AI edit, one line changes. Okay. See, I, I can't get it to be that precise, so I'm faster than the AI when I need to change one line. So, look, I can seriously issue throughout my life that I'm like speaking with people, and now all of a sudden that's okay because they don't talk back to me.

I'm always right. Kind of like, and I want, I want more of that. So you like, you like your terminal to be sycophantic. Yeah.

No. So actually what I do is that in, in my terminal in ghost, I have different types, for different projects. Sometimes I work in multiple projects at the same time, which is badness, but bear with me. And each of those top has two pains up and down.

one pane is Claude Code, another one is just pure terminal. When I need to do those things, like vi going back to your question, And Bret, change one line. Yeah, I would use vi, right? Execute some command.

Still do it. So I just switch up, go up and down, or full screen if I need only one of those panes and that, that's my workflow. All right. I'm, I'm putting nine and I don't recommend it to everybody, just to be clear.

Kind of, I just freaky that, that's all. Yeah. Well, and you know, like, it, it's a purist approach. and I sympathize or, or align with that because there are days where I'm like, I don't want anything on my screen.

I just want a giant terminal window. And I don't want to leave this terminal window. So if I'm editing something, it's going to be in Neovim. which, you know, I go back and forth between VS.

Code and Neovim, Kyle Quest. And Chad is asking about anyone using Zed. I have Zed open right now. It, I've used it as a text note taker, like it is, I'm actually tracking our show notes with it right now.

it's super fast. It replaced my sublime. text editor, sublime Text, which that replaced whatever was before VS. Code.

The Adam editor. So Zed is my go-to random editor of, of text. But I do see a lot of people using Zed, full time, especially in the Go world. They, they tend to like Zed as a Golang editor, but Zed is awesome.

If I would be using it, I would be using Zed to be honest. If I would bad installing it, I had it for quite some time. What is really, and I suspect that next year will be the year of Zed, because what makes Zed special, forget about AI for a second, right? Is that it is IDE that is built or being built for collaboration.

That's its main strength, right? And the more we are adopting AI agents, what not. The more I can see how that can be a competitive advantage of Zed, right? Mm-hmm.

Because you can argue, hey, how much do I collaborate with others in my daily workflow and how much I work alone? But now that we are moving up to different degrees to to agents, I think that collaboration is happening now all the time. So if something is designed for collaboration, that makes perfect sense, even though it might not be the type of collaboration, how it was envisioned from the very start of that. That's a, that's a good take.

I like it. Yeah. I've actually been sad I wasn't in a team to use Zed because when they first came out and that was one of the first features was like automatic GitHub login, automatic collaboration with other people. And I was like, I can't get to really use any of these features 'cause I'm not like pair programming.

Or you can do that with Bret. You can do that with me. Okay. Just hit me up.

Okay, normal. And I can, we'll, a z session sometime, maybe we should, should go online and see. We should see how many people live can join our Zed hacking session. We'll be Z friends, we'll be Z friends.

Sounds good. There we go. AI browsers. Ooh, thumbs up, thumbs down.

And if it's thumbs up, which one. Go Viktor down. he's thumbs downing all AI browsers, which I'm also thumbs downing all AI browsers. Wait, so Bret, okay, I've got to call you out.

This has been a running meme in our show with you for years now, where we almost had not a single show where you did not mention ARC browser, right? It was like, I love ARC browser. You, you were like an unpaid spokesperson Super Bowl ad for the ARC browser. What happened?

What's going on? It is not an AI browser. It has AI features, right? So, okay, so what does that mean?

Like, I'm not a, I'm still running Firefox and, and Chrome. So yeah. what is the difference? What, what is an ai?

Because you're. You're like our, you're like our, our baseline. You're like the normal dev pre 2023 or 2022. and, and Viktor chime in here real quick.

the difference between ARC and an AI browser, which is why they changed and made a new browser that was not arc, is ARC has AI functionality that I love, like it'll rename downloads to a more, it'll look at the file contents and then rename the file name to something that's not tax rebate. Document PDF 0 0 1 2 2 1 1 under underscore five, like it'll rename it to Bret's tax return, whatever, right? it has other little AI features like it will auto name, it'll rename the tab. To what it thinks that page is about versus what they will call that tab from the website.

Right. It'll do all that sort of stuff. It'll automatically combine different tabs into folders if you want to organize things. So it'll, it'll sub, it'll create groups and groupings based on the contents, but I can't talk to an AI about a website.

Got it. So it's like a browser. Oh, so, so those, that's not an AI browser. What you're talking about is using AI features to make a normal browser more useful.

Right. Essentially that's what they started with. Yeah. Okay.

Go Viktor. No, I, I don't know what would be a definition, but I see a browsers as being, you know, it's a similar like MCP where AI can use tools so you give it Right. Okay. Similar kind of basically AI can use your browser to do things for you.

does doesn't work, right? ai browser definition. It has control of the URL bar from an AI prompt and likely can see the complete and it can see the complete dom and it can control the window. So we can create new tabs, it can delete tabs, it can read the page, it can summarize the page, it can change, it can decide what websites you go to.

It can open up five pages at once. That's an AI browser in my definition. Got it. So for example, if I'm building an app in Kiro and I want to, it's like a web app and I use like a playwright MCP server mm-hmm.

That will like open up a headless browser or whatever it is and click buttons and you know, test that website as q and a is that playwright MCP a AI browser? It feels like that, I mean, it's probably not okay, but feels like that? Yes. Yes.

Okay. Basically, AI can do stuff using a browser. So is it a browser with just a really good API that is attached to an agent that knows that API? Well, you have you have an AI dropdown for Gemini in your, Chrome browser.

Do you not use that normal? I, I don't, mostly because security, I, I mean, sure. Yeah. Well, I feel that there are two major use cases over there.

One is that, use AI to make sense of the, of some information that is available online. You know, that's the example that you brought from mentioning, oh, I have seven tabs open and I want sum this, stuff like that. Right. And the other one is to do things for you, like, Hey, I want to book a flight.

Ah, the first one, I feel I don't need a special browser for that. That's what, you know, ChatGPT, Gemini and all this stuff. Yeah. I get a place where I type something and I get the answer kind of like, I'm not sure why I need a browser for that instead as opposed to a tab.

Right. does the same thing now, if somebody would ever convince me that AI can book a flight that matches what I really need, and that would take less time than if I do it myself, I will adopt AI browsers that same moment. but at the same time, like what you're talking about, like for example, booking a flight, that's just a stop gap between just because the flight, the Air Airline website doesn't have any API at some point. You, I mean, you could go into a terminal and say, Hey, I want to book a flight.

And if the airline had a good API and an MCP or an agent, then it doesn't matter. Like it's so the, the AI browser in those use cases, it's just a stopgap until there's like better integration. Maybe now we are talking, now we are coming to the crux of it, I think because you both put thumbs down on AI browsers and I don't know why. Well, yeah, let's, yeah, let's, let's clar, let's clarify our why.

My why Security nightmare. Everyone keeps jail breaking them. They will eventually, we will have announcements of major. Problems We've already seen that they're incorrectly buying things.

They're not buying the correct things. They're, when they shop for you, they're screwing up. They, right. They're the fact that an AI can see everything I'm doing online, including every password I'm putting in, and like, it is a security nightmare and there's no way around it.

And I don't know how we're going to solve that. So until people just pretend that it's all fine and ignore security, which is what maybe people that are using today are doing, like the DIA browser, which is the Arc's replacement from the browser company. I mean, these things, there's no way to keep them safe. I don't know how we would ever do it in our current con and at least in the current world of AI and MCP and everything else that we're living.

So it sounds like to me, like if I, if I was a scam artist or someone designing phishing, th this is like a Yeah. Dream. You can jailbreak, this is a dream come true. Yeah.

You can jailbreak them and people keep doing it. Yeah. I would get someone to download these browsers. And then you're off to the races essentially, right?

Oh yeah. Yeah. Yeah. Actually, I, I, I don't even understand what they're trying to solve, to be honest.

And why browser? Why not APIs that al you just mentioned kind of, basically, right. If you want, let's say that I want you to find me five top options for a flight to Whenev wherever, whenever. Right?

Right. why would browser AI browsers do it for me if no GT is perfectly capable of doing it? Yeah. I want, I want the unless API to find, unless I wanted to.

Yeah. Unless I wanted to book it for, for me. Right. Kind of like, Hey, here's my password and user and here's my credit card, do it for me.

And I, I already have enough trouble trusting, those that are, should be trusted, like Amazon with that. Right Now, I'm not trusting with my credit card number, some malware. Right. So, and that, that's not the scammer.

Right. And that's not like the main. Friction point, like purchasing, click this, like purchasing the ticket isn't the major part of finding an, finding a flight. Yeah.

It's the finding the flight. That's the major part. It's not like the credit card information and your, you know, like your passport IDE and your profile. that's a solved problem.

Yeah. Right. Exactly. And so to me, I would love, like, you know, a, a direct MCP feed into the Sabre API and like Google Flights and Flight Aware and you know, kayak or whatever those tools are, and they should just return JSON with all the prices and schedule information and call it a day.

Right. Like, but the key here I think is that. the services like Gemini and, open ai, open AI and whatnot, they're already capable of doing that without my browser. That's the point, I think.

Yeah. Right. Kind of like you're perfectly capable of finding that flight. Why do you navigate my browser to do it?

Yep. I think it's just a way for the AI companies to lock you in. I mean, it's no different than Google owning Chrome and I, I feel like it's just another lock in play and I don't, I do have hope for the Zen browser. I do have hope for the Zen browser.

It's not an AI browser. it's taking all of the ARC features, putting 'em on top of Firefox. I really wish it wasn't Firefox, but it's got the Firefox Mozilla engine. but it, it feels a lot like arc.

There's someone in chat who's like, yeah. but ARC is no longer actively developed. ARC is being maintained. There's a difference.

I'm perfectly fine with ARC's existing functionality. I don't like a couple of the bugs. It has like, especially window management bug that it has on Mac. But everything else in it has performed great for me for years now.

So I don't need new features or features expanded. I just need it to stay stable and secure. And they've been doing that with regular weekly chromium updates all year long. So kudos to them for shelving a product, but keeping it safe for the existing users that still use it.

I bet you they still got a half a million daily users on the, on the browser that they've shelved, they've probably got more daily users on it than they do on the, on the DIA browser. My, I'm guessing, but until we have like something Arc replacement, which is probably then. Zen still got some little limitations in corks that I'm not quite ready to leave a completely for it yet. I don't know about Viktor if you're trying something else besides Arc or if you're No, I'm back, back to Chrome just because I have no patience now to, to, know, I fair, I'd like to go back to your previous question, kind of what they're doing that, the many vendors have a real problem and that's, that they don't control the platform, right?

You cannot really do real business, because Apple controls your mobile phones, right? they say what you can and cannot do. Google controls your browsing experience Chrome. Right?

if you don't control those things, you cannot really go as far as you want. And I'm a hundred percent sure convinced that that's why. they're releasing those browsers because if you are convinced to switch from the platform, core platform that you are already using to theirs, then they can do the stuff they would normally, wouldn't be able to do. Right.

That's the game. Not because you benefit, I cannot imagine the reason. Why would you use open AI browser, whatever it's called. Is it come or something?

I don't know. It makes no sense except when you look at it from that perspective, that kind of, if we manage to convince you to switch to our platform, then we are talking business. That that's the whole game. Yeah.

That's my conspiracy theory. Theory, yeah. I don't think it's a conspiracy. I mean that's just, that's like a classic vertical integration play, right?

Like, I think that Arc will go down as like one of the major. sad stories of tech, of a fantastic product with an upward trajectory and a clear potential market for people to pay. I wish they would've, they made the same mistake. Docker did.

They didn't charge early enough. No, they still don't charge. There was no opport. I would've paid them money for years, five, 10 bucks a month for access to those AI features.

Like, but you and five others? Well, maybe, but they could have expanded that market. Okay. 10.

10. Okay. Don't fight me on that. Let it be 10.

Yeah. But, they, and they could have explored additional third party, you know, like additional functionality that, and, you know, they probably could have done like everyone else and said, here, we'll give you access to the models for our, you know, we'll be the gateway and we'll take a cut off the top like everyone else is doing with the IDEs. But the, the crazy part to me, Arc died because of Silicon Valley. it died because of VCs.

VCs need their money back and there wasn't a clear path to a $500 million annual run rate kind of thing. So they couldn't figure out how to monetize it in a way that they could've figured out how to monetize it. If they had a 10 or less people team and they just needed to make 10 million a year or something, but they couldn't figure out how to monetize it at scale, which is the problem with VC money. So now they've got to make another terrible browser that's, that's not going to make them that 5 billion either.

Without VC money, they couldn't have made those movies that cost like Hollywood production. Yeah, well, they might be right, by the way, they might get bought by another company for a billion dollars, just like, well, they already did. They got bought. So they want, that's their thing.

Like they got bought, I keep forgetting. They got bought, by Atlas, Atlassian. So they made their money. The VCs got paid back, presumably.

And so to me, like I kind of wish that maybe because they've already been bought, maybe the DIA features will just be added to Arc. I don't know. Or maybe they'll add the ARC features to DIA and I can turn off the AI. I'm not really sure, but at some point the AI browsers are like, it's a question of how much will Chrome have to compete with the AI browsers and are we going to get a worse chrome in the future?

Like, are we going to get a Chrome. That we all learn to hate because it's AI first and we don't want it, and we just want to turn off the AI. I've already got friends that work in the AI space that are like websites. Could be, remember how SEO hackers used to put little tiny texts in a, in a font at the bottom of the page that would seed the search engines with information to help them their rankings.

And they would put it in small fonts with the same color as the background so that the human couldn't see it. But the, well, the same thing is going to be happening with ai, where we're going to end up with jail breaking texts at the bottom of your website. That would, that if I land on your website, now you're controlling my browser. And, I, right now there's seems to be no one that's able to avoid that attack.

which is a pretty basic attack. Same with like, you know, MCP pony and, and a lot of these things that we're, We're seeing is it's not, anyway, we could talk about browsers forever. last, last call. Wild card, like what other AI tool did you try and throw away this year?

Or you tried and you adopted? It could be a SaaS, it could be like, I use a notion AI more than I did a year ago. they keep iterating and making it better, better. So now my notion AI can, search all of my stuff in a very.

Nervous. I like it has access to my email and my documents and my go, my GitHub and it has access to everything and it is a little concerning. I just hope their security team's great because it has a lot of access to my stuff. The only thing it doesn't have access to is my passwords and my browser history.

That's about the only thing it doesn't have at this point. But the AI is great. It can do crazy things like, Hey, can you tell me where in 2025 did I interact with Nirmal? And it will find all the pages and, and it'll summarize conversations.

It'll say You were on this many shows with him. You had these, this, these different email conversations with him. You have these documents that you shared with Nirmal. Like it, it does all that in five seconds and.

That for me when I can't find where, where did I put that thing? Is it a Google Doc? Is it a notion page like that has fundamentally, for me this year changed the behavior of the way I work, and it's a hotkey now in Mac. So I can just hit Command K and search everything in Notion, or ask the AI questions about my notion and my Google Docs and all of it all at once, and I, I do have to pay for it.

It turns out that when Notion announced, they were doing a, a stock thing with their employees on Monday, they announced this, they kind of gave away the fact that 50% of their revenue is now AI, the cost of the, AI plans on top of the regular plans. So they have successfully monetized AI so that now basically a majority of their customers are now paying. Double what they used to pay in order to have access to the AI on top of the Notion features, which I think is a, a huge win for a, a non-AI company to add AI functionality as a feature center and to be able to actually show a lot of profit from that.

I think it's a pretty cool thing for them. So that's my pick. I'll throw it in Wild cards. Anyone else got one?

It can be Dev, it could be nine devs. You want to go first? No. Go for it.

Go for it, Viktor. Okay, so, you said yeah, what you picked, and what you abandoned, right? Yeah. I mean, I, I already answered what the abandoned there is.

Id none. I, I was a VS. Code cursor user all, all the way. And that's what I abandoned, right?

now the pick Code Rabbit MCP. Okay. That's absolutely amazing, right? 'cause I love the idea of, you know, I create a PR.

I get review as, in parallel or without the human review, right? I get review and Code Rabbit reviews are absolutely amazing, right? Not going to try to compare them now, but the reason why I say MCP is because I don't have to leave Claude Code. I don't have to go to GitHub PR to see what it did and then copy and paste it back to Claude Code to say, can you fix this?

And so on and so forth. I just say, create a PR go. I go and fetch coffee or something like that, few minutes later. Okay.

What are the issues? tell me what are the issues and let's sso, let's pick which ones to resolve. And I continue my workflow right in, in my agent, whichever I'm using, right? so, it's absolutely amazing that it's happening outside of Claude Code, but still works in IT or cursor whatever key, right?

Whatever you're using. So Code Rabbit MCP specifically, absolutely amazing. All right. I haven't tried it yet, and I've been mostly focused on building some GitHub courses this year, and I've been mo mostly focused on the GitHub functionality.

And I, I've been very curious to, to see some of these sort of, GitHub workflow automation tools outside of GitHub and see how superior they are to the native offerings. So Code Rabbit just jumped to the top of my list. I envy you, man. I really envy you because you're so used to, to copilot that now you're going to enjoy so much, all the other things, and your body is so low now.

Maybe that's the secret to life. Yeah, exactly. Expectations, low expectations. Exactly.

Exactly. Low expectations. There we go. That's funny.

Yeah. You, you, you've been working your whole life in mainframe. Let me introduce you to cloud, right, Bret, you got to open up Copilot and say, how can I raise my expectations on AI tools? What, what tools are better than you, Mr.

Copilot? that would actually be a great test. I'm going to ask, KIRO, CLI like what? IDE what AI IDs it would recommend.

Right. What do you recommend, right? are these biased? I don't think that it is.

I think it's just going right to sonnet or whatever, you know, whatever model it's connected to, but it's still, it's still trained on user, you know, on public data. So it has all the Reddit posts and all those things. I'm not sure how much it, even the authors of models can avoid that. Yeah.

So I'm bringing that up because, this is like a burgeoning field, right? Which is the AI optimization, right? We have search engine optimization. This is the wild card.

This is one of the wildcard answers. Okay. So we have search engine optimization, right? Like how to make your results show up.

we're already there and we're going to continue to have to understand how to make sure our stuff shows up in. M responses in models. And so like, you know, we, you know, how does Bret Fisher's and Nirmal Mehta's Agentic DevOps fm podcast, show up when you ask Claude Code? Hey, I'd like to listen to a podcast that tells me more about how to use agents for DevOps.

and that whole entire field is very nascent and burgeoning. there's no major tools right now that I know of. I haven't done any digging, major digging into that. But, right now it's even hard to understand what percentage of people are discovering your, you know, your blog post, your podcast, your content through ChatGPT, through Claude, through Perplexity and that visibility and tools that will help.

Bring that visibility to light are going to be the one part of the future for sure. you don't have a particular product there. I don't have, I don't have a product there, but it's a product category and it's something to kind of keep an eye on because it's the other side of what we've been talking about, right? Like if we're using all these tools in the terminal, in Claude Code, they're being influenced by something to give you that response, right?

There's an opinion there and how they deform opinion that's going to be optimized in the future. and it's something, so that's what I, when I asked you about AI browsers, I thought that's where this, that conversation was going to go. And I'm kind of surprised that wasn't part of the response I got from y'all because to me. The idea behind like an AI browser would be to then sell the influence, on the other side of the market to that person that's using that browser.

Right? Like, if that makes any sense. Yeah. So there maybe what you're talking about, like, so a two friends of mine, have a company, a startup called No Atoa, and they do what you're talking about in terms of like helping marketers.

It's, it's a tool for marketers, right? For people that care about SEO and they help you understand what your brand is doing in the different models and the different chat bots and, how that correlates to your old school Google, SEO search stuff. And, they give a pretty great, email. It's not quite weekly, but you can also, you can just get all this at Noah to, but, yeah.

I wonder what ad blocking in the future is going to look like. can, would an ad blocker be like, unbias the response from an LLM future? So we have ads now showing up in go. Like they just wrote, had a whole email about this like last week or something.

Was the, that ad showing up in the chat? Chat bots now? Mm-hmm. That, that's already happening.

And Google was first as they should be. ChatGPT is next. They're planning on it. and so the problem is, the big problem here for me is that it's one thing when you're just searching the internet, but people don't treat chatbots like they treat the internet search.

Like people are having therapy sessions in the chat bot. Can you imagine having a therapy talking about some deeply personal topics and then you see an ad for whatever. you know, some prescription drug like in the middle of the chat response because that's kind of how it's showing up is like, Hey, here's my opinion. But also did you know about, like that's super cringey and it may backfire.

So I'm guessing that that's part of the reason why maybe we don't see ads yet. 'cause it seems like it's been taking a long time for, so are we going to have in, you know, in, in this IDE of the future, is it going to, there's, is there going to be a banner that says, why don't you go grab a Red Bull while I fix your code? kind of thing? Or are there going to be comments in the code about like, you know, this would be a lot easier if you used X, Y, Z service instead of Java or something.

You know, like, is that where we're headed to? We are, but that's partly our fault. And this goes back to the conversation about Arc browser, right? We are so used never to pay for anything.

Yeah. Right. Yep. People don't want to pay for, for open ai.

I mean, professionals like we are, we are paying, right? But a normal person is not, is not paying for it. just as normal person would never pay for a browser, right. And if we insist on getting everything for free, what's the va?

What's the only VE left for those companies to finance themselves? Well, selling you Red Bull. Yeah. Yeah.

Kind of. We need somehow to change as people are customs to kind of, Hey, I'm going to pay for this. Do you remember the day that Gmail showed up and announced ads next to your emails and how everyone was up in arms for the invasion into privacy of them reading our email, which they were already doing. They were all already doing it, reading our email, and then providing ads next to them, Th that was 20, 20 years ago.

And that's completely normal now. Like it's totally normalized to have ads right next to my email about what was in that email. Nobody makes a mention of it. It's fine.

This is the next wave, right? This is the next wave. Is it You're having private conversations about personal topics, healthcare topics, interesting, whatever. And you're going to have ads and you're, and people are going to have to just start to get, be okay with it.

And Yeah. And well, the rumor right now is that, ChatGPT, even on the, or Claude or whoever it was, that even on the $200 plans, you're going to see ads, like everyone's going to get ads. Because one of the arguments, one of the problems with ads is if you can pay for ads to go away, you're the best customers are the ones that are paying from the go away. So when you're an ad seller, you really want the best customer.

You want to be able to sell those customers, which is why you could never get rid of ads in Facebook. 'cause you, you know, anyway, I, I'm not sure that Anthropic will do that. Maybe to some extent, but not much simply because the approach that Anthropic takes compared to open AI from the business perspective is very, very, very different. Right?

Sure. Yeah. Anthropic is an enterprise based company. They're selling to enterprises.

That's the revenue stream, right? Open AI is trying to be end user type of service company. Right? B2B, like your mom and pop Yeah.

My, my parents type of users, right. Or my daughter or whatever. so they will definitely do ads because their users don't pay Anthropic. I mean, I know my company's paying for it.

many companies are paying for it, and if they're not paying for a Anthropic, they're paying for Amazon or for, Azure or something. Right. And I, I don't think that ads would fly there. Yeah.

So that was my hot take. kind of just like something to look out for is, Is like the wild card. if you'll take it, it's not a real tool. I'm not sure what to write.

'cause our lists are mostly tool lists, so I don't know what to add there. But, ad block, LM Unbias ad blocking tool of the future can now reduce it to two words. Yeah. oh.

Like, LM Block. A SEO. Yeah. There we go.

L lm ad block, anti anti AI SEO or something like that. but yes. Well, I've, we we're, I think we're at the end here. Yeah.

We, we could keep going on tooling forever. but I will leave this here 'cause I feel like this is a really good list and any of these topics could have become, became their whole hour. Yes. I.

So thank you so much for being here. This was our Christmas episode. I don't even have Christmas music like I usually do. I usually have the room set up in Christmas.

This was a minimal Christmas year, but, thank you both for being here. And you can find Viktor on YouTube at, at DevOps Toolkit. And Nirmal is on LinkedIn. And all of this stuff will be in the description below, So, thank you both for both being here and we will see you again in the new year with more live streams.

And before we leave, I just wanted to thank all the viewers, listeners, supporters that we've had over this past year, and years. Bret and I can't do this without you. Please tell your friends, family, colleagues, Random people that you meet on the street, the people that you see at conferences about both of our podcasts, our live stream. I thoroughly enjoy the conversations that we have in the chat.

It seems awesome to get to, to see from everyone what y'all are doing, your questions. And, we want to make sure that we can continue to do this in 2026. So, check out our Agentic DevOps podcast, the one that we launched this year. We'll probably be having way more conversations about it next year as all these tools, maybe this list of tools will be completely different in six months, probably, possibly.

Yes. And like, and subscribe and all those things. if you didn't know, I'm making courses, links below into description. See you soon.

Bye everybody. Bye. Cheers.

Related episodes across the Index

Other episodes covering the same guests and topics, from across The B2B Podcast Index.

  • Utilizing AI internally to iterate faster and empower smaller teams to upskill w/ Vivek Raghunathan #263The Engineering Leadership Podcast · on GitHub Copilot96 / 100
  • AI Was A Waste of Time, Until It Wasn't with Megan BoshuyzenMaking Sense of Martech · on Claude Code91 / 100
  • How Organizations Can Thrive in the Human + AI Era with David ChestnutThe Edge of Work · on Claude Code85 / 100
  • Episode 018: Season 2, the $75 Consult and the Frankenstein StackAI Tools for Practicing Lawyers · on Claude Code84 / 100
  • AI for Engineering Is Leaving the Demo PhaseAI Across The Product Lifecycle Podcast · on Claude Code83 / 100
  • Late checkouts, AI agents and the future of guest communication with Cole Rubin of ConduitMatt Talks Hospitality: Real conversations for innovative hoteliers · on Model Context Protocol (MCP)83 / 100

More from DevOps and Docker Talk: Cloud Native Interviews and Tooling

All episodes →
  • K8s Maxxing with AI-Native Platform Engineering Stack with OpenChoreo59 / 100
  • Docker AI, what’s new with MCP, Agents, Sandboxes, and more85 / 100
  • Backup S3, Google Drive, iCloud, Notion with Plakar87 / 100
  • Your Images are Out of Date (probably) - The Silent Rebuilds problem81 / 100
  • Find Your K8s Happy Path with RawKode Academy80 / 100
Explore the best B2B Engineering & DevTools podcasts →
All DevOps and Docker Talk: Cloud Native Interviews and Tooling episodes →