Windsurf podcast mentions

Quote

Brain grid. Building with AI coding tools is exciting until the moment things start breaking. You ask for a small change and suddenly three other features stop working. AI gets confused, misses edge cases, and loses track of your intent. The problem is not code generation. The problem is planning. That is why Brain Grid exists. Brain Grid acts as your product management agent. It writes clear specification, maps UX flows, asks the clarifying questions you forgot to ask, and breaks big ideas into engineering grade tasks that AI coding tools can build reliably. It guides Cursor, Clog Code, Replit, Windsurf, and others so they deliver features that work and keep working. Founders use Brain Grid to build real AI native SaaS products without a technical background. If you want reliable features instead of fragile prototypes, try BrainGrid for free at braingrid.ai. That's Braingrid.ai. This episode is sponsored by Mesmo. If you're responsible for reliability, performance, or platform architecture, you already know the problem. Telemetry volume is growing faster than Then teams can manage it. Mesmo addresses this by moving observability upstream. Instead of storing everything and asking questions later, Mesmo processes telemetry in Motion, filtering, transforming, and enriching logs, metrics, and traces before they reach your observability backend, the result is cleaner data, reduced…

Quote

people who understand the code anymore, it's gonna be zero. Right? And in that world Well, how do we maintain it? Like what happens? How do we support it? um at any sufficient scale. Um and I think a lot of those realities are starting to be realized now. Um you started this company in in 22, um, I guess late 21. Um you know pre-Chat GPT kind of with this understanding that that what we're seeing now with vibe coding with energetic development um cursor windsurf the likes, um, you know, basically kind of anticipating. the future that we get to live in today. With that in mind, like Uh especially you said you've been working on this for three or four years. Like what it how's it been to watch the uh advent of you know ai assisted coding and uh agentic coding, vibe coding, all of this like it it, you know. It both you know, it it was it's both something that you know we You've been able to anticipate for a really talk long time, but also something that like exploded all you know, seemingly overnight and I'm curious, um, you know, when You think about it, what stands out for you as kind of the most interesting aspects of

Quote

an AI native company just has that live coding and then, you know, do what we do um or you know, implement all these complex things and then Um so um I think um because we as uh you can say I'm also leading engineering team. Um our engineer You know, some some people, you know, chose to use windsurf and curse surf, right? So There's no there's no you know, um we don't dictate on what kind of tools they use but um the commonality is that I think nobody um um you know, does um coding uh without any kind of um AI I think that the difference will be percentage, right? Percentage. For example, AI coding completely enabled our product manager. is to actually you know be back as a builder. will be able to build prototypes and be able to build UX. um using lovable Vicma or not to to uh actually lot of things are already very clear so that engineers will be able to fully

Quote

Yeah. uh it has been in the air for now many years, obviously well before L One thing I found pretty interesting is For example, different code agents take very different approaches to retrieval. from Windsurf shared an interesting perspective on how they approach retrieval. In the context of Windsurf. So they use classic code chunking. Along carefully designed semantic boundaries, embedding those chunks so classical. The kind of semantic similarity vector search and retrieval. But they also combine that with the same thing. For example, grep. They then also mention knowledge graphs. They then talk about combining those results. Doing your ranking. So this is kind of your classic complicated Now what's interesting is Borstrment. And Claude Cote has taken a very different approach. He's spoken about this quite a bit. Clock code doesn't do any indexing. It's just doing quote unquote agentic retrieval, just using the case. Using simple tool calls, for example using grep, to kind of poke around your files. No indexing whatsoever, and it obviously works extremely well.

Quote

Thought provoking stuff, definitely worth a read. Employ a peon Today, I clicked the copy to clipboard button on the brew install command approximately. Approximately 0.3 seconds after landing on this Peon Ping website. What could possibly be so compelling? Quote. Game character voice lines the instant you can see. Your AI agent finishes or needs permission. It works with Claud Code, Codecs, Cursor, Opencode, Kiro, Windsurf, Antigravity, and more, never lose flow to a silent. Terminal again. With ninety-five plus sound packs and counting. There's a game character voice in here for everyone. I sampled a bunch, but I'm sticking with the defense Default Warcraft 3 Orc peon for now. Ready to work. Give it a try, there is a 0% chance this won't bring some joy to your work life. Ladybird adopts Rust. One popular segment Of our Lady Bird Pod with Andreas Kling and Chris Wanstroth was when Andreas told us They were leaning towards Swift as their C replacement. Well, well, well. Turntables. Quote. We previously explored Swift, but the C Interop never quite got there and platform support outside the Apple ecosystem was limited. Limited. Rust is a different story. When we originally evaluated Rust back in 2024,

Quote

A kayaker built Creek Intelligence, which predicts when rain-fed whitewater creeks are runnable, and a parent built a toddler behavior chart rendering as an exploding universe called Jude Stars. In terms of challenges people ran into, there is one clear infrastructure gap that the whole field is screaming about, and that is memory. A meaningful number of the submissions are are effectively elaborate workarounds for agents forgetting everything between sessions. MIS uses 50 plus markdown brain files. Synapt reported that their agents kept forgetting what each other were working on. Carrier file is literally a text file you paste into any AI to help with context. OpenBrain shares one MCP memory server across clawed code, cursor, and windsurf. All of these Hacks, markdown files, knowledge graphs, vectorDBs, copy-paste text is kind of the diagnosis of the big problem facing the agent ecosystem, which is the memory problem. In terms of who is building, the median builder here is probably not who you'd guess. Partially that is of course because of the wide nature of this audience, partially it's because Agent Madness might have represented a different type of opportunity that non technical builders might not usually have had. Still, we have paramedics, glaciologists, kayakers, restaurant operators, sailors. Sales leaders, people who are domain experts and can now use software to do…

Quote

It's like, great, I want to go make a rapid change on the website. It's already live in production. I've come in here. said Jim and I run this and I'm looking at these three of them have a live preview as soon as I'm like this one is the right one you like it too I can send you the URL right now I can't merged and it's done. It's done. You're still going to go through your CID pipeline. So if you've to figured any checks and balances, that's still going to happen, but it just makes it quick and yes you could go into Hirscher Windsurf V But you're sitting there having to say deploy, check, and the workflow brings you out. out what we find is once you deploy the first time, you can iterate right there. And it is a little bit different for traditional developers because we're used to working in our IDE, but when you're but not as new hot and I'm calling myself one now, not as new hot front. and devs were just like sweet here go do an agent run yourself Maybe you don't merge, but you can show your stakeholders and run and just makes that feedback loop incredibly quick in that. that process frictionless. And I do think you're going to see more and more of that. And why wouldn't you? Because I always said and we've always said, there's no better testing than in production. And it also changes the way you think about A B testing. Now with this, you can go really take it off to the ra…

Quote

using AI on their machine, either in their IDE or on their C LI with clock. code, right? People are doing it where it's relatively local. They're still much more in the loop than I expected the people to be. They're not not just treating it like a coworker who opens PRs, right? And so that was just a bet we made. that AI is just not there yet, or maybe developers are not there yet, that we put a lot behind that. turns out like people who are much closer to the traditional workflows, companies like Cursor or and windsurf, they were saying we're just gonna like take the existing ID and slightly add some more AI. to it, that worked much better than trying to leapfrog straight to like fully autonomous agents, right? So that's part of the things we got wrong. The second part was just like is a dedicated AI for software maintenance, what the world really needs. And at the time I just came to the conclusion that I don't think so. Like I don't think that's the right paradigm. We had enough developers that were saying, hey, We don't really need your products because we're just gonna have Cloud Code do this. And like maybe it's Not as good. Like maybe it's gonna be fifty percent as good, right? Or sixty or seventy percent is reliable, but that's enough that like it still is doing enough automation that we don't need to have a totally separate tool for it. And I got to the point that, ye…

Quote

We just people just type in what they want and get that set up. And immediately I thought about well, Zapier has something like that. And it's not very good because every Every time I've tried to use it, it I type in the prompt, it generates something, and I can see the workflow. And half the time it doesn't work. Um and another kind of example was where um you know using clawed code and using it inside cursor or windsurf or whatever your favorite ID is. That was kind of pretty cool, right? So you got your code, you got this window, you can start to do interesting stuff. And then when Claude um Anthropic release clawed code for the web, I was like, no thanks. Like my head for a while couldn't get around how do you do this when you can't see it? the code. You just have a prompt to do this all right. And now that I've been using it for a while, it's actually pretty cool. And it starts to make sense. So I'm just curious like why why did you take that route and you know what what was the reaction like what did it make it easier to sell the product Yeah, so I will say, um, I think the only real reason we have this advantage is not due to any like intelligence. It's literally just because like Z Zapier, Zapier, and NADN, what they have to do is they have all of this infrastructure that they've already set up, right? Like they have two, they have two solutions when

Quote

And one of the other interesting aspects too that I forgot to bring up earlier is there are many styles of agents. And there are coding agents that operate aut autonomously. is something like a cursor or a windsurf or even these cloud codes. and GitHub CLI and Gemini Cli and I'm wondering if you can talk to some of the juxtaposition I think we had this debates back and forth internally, uh both kinda for our own I think ultimately you need to have both modalities. So there There should be code first modality and most engineers think that that's the main one and that that's gonna stay there forever. I think they're overestimating the importance of that modality and the longevity. And then there's gotta be agent first'cause as you're starting to manage this fleets of agent and They're work more complicated workflows, both sequential ones, and then parallelizing multiple. Multiple agents. Again, the whole industry is not yet unlocked. Very simple technique which ensembling right in machine learning we always use ensembling so you should be for complex problem you should

Quote

is that you are looking to make faster, cheaper, more reliable and If that part is sufficiently narrow on the task that is trying to perform So I'm looking at like right use cases very often have that in terms of like re ranking and priority. advertising and all of that stuff, like then you probably ought to fine tune a small model. Um I I think I have That uh Cursor and Windsurf have at least part of their workflow. That's using a fine-tuned model, but I'm not sure. Yeah, I think the point on narrowness is is definitely really key. I mean i it's tempting in some cases to try to imagine creating like the fine-tuned model for our company that does everything for our company. And that in my experience is not the way to go. It's much more about just nailing down With clarity, you know, what is the desired behavior on something that really matters. A good example from the reinforcement fine-tuning docs that I was reading this weekend was just Like from healthcare, some sort of doctor notes, you know, translation. transcript of the appointment to like a diagnostic or I think it's even a billing code, which is like Like super gnarly stuff and obviously accuracy, you know, really super important there. So that kind of thing often I think will work and and hopefully will push the frontier of like what people can

Quote

papers using AI tools daily weekly and 50% of the code going to production was written by AI. Those are two wildly different scenarios and it's important to figure out where you are. We at DX have a tool that runs on the file system layer. So it can track in any IDE in the terminal, whatever code completion from AI tools as well as copy paste and then track the human editing and then do that on a per line basis at the commit level so that we can look at the ratio of of AI code to human authored code in a given commit, and then we can track that commit all the way to production. So there's like pockets of tools that could do this before like Windsurf had some pretty good telemetry. into what was going on. But before we just sort of had acceptance rate, which is, you know, accepted suggestions and really no insight into what happens. even like before the commit, after the acceptance, and then definitely after the commit and into customer facing environment. So very excited about this. It's technically very cool. But yeah. It's an industry first. Is this a part of the exact question? Is this a part of the DX platform or is this an open source project? What is this specifically? Sorry, that should should not have been my first reaction to that. It sounds really cool, but it does so the scary part to me is it sounds very Very brothery? Yeah, like severe. And unfortunately, we could…