# Entire with Rizèl Scarlett (and co-host Dev Agrawal) > Rizèl Scarlett demonstrates Entire’s Git-based memory for coding agents amid a wider discussion of AI workflows, code review, languages, and open source. - **Collection:** Video - **Published:** 2026-05-22 - **Author:** Anthony Campolo - **Canonical URL:** https://ajcwebdev.com/videos/entire-with-rizl-scarlett-and-co-host-dev-agrawal/ - **Markdown URL:** https://ajcwebdev.com/videos/entire-with-rizl-scarlett-and-co-host-dev-agrawal/index.md - **JSON URL:** https://ajcwebdev.com/videos/entire-with-rizl-scarlett-and-co-host-dev-agrawal/index.json - **Channel:** [Anthony Campolo](https://www.youtube.com/channel/UCpdzti0GURPfMjKzYK5FVSA) - **Original URL:** https://www.youtube.com/watch?v=UfF7uSu71Ms - **Original Label:** Watch original --- ## Episode Description Rizèl Scarlett demonstrates Entire’s Git-based memory for coding agents amid a wider discussion of AI workflows, code review, languages, and open source. ## Episode Summary Rizèl Scarlett joins Anthony Campolo and Dev Agrawal to discuss leaving Block after most of her team was laid off and joining Entire, a startup building version control and collaboration tools for AI-assisted development. She explains and demonstrates Checkpoints, which records prompts, tool calls, commits, and agent sessions on a separate Git branch, enabling search, handoffs, reusable skills, rewinds, and shared organizational memory across tools such as Codex and Claude Code. The group considers how intent-aware review could help maintainers assess large agent-generated changes without reading every line. They also debate Bun’s AI-assisted rewrite from Zig to Rust, tests as the basis for trusting generated migrations, languages designed for agents, and the continuing responsibilities of engineers. The closing discussion covers open-source security, dependency strategies, OpenClaw’s sudden popularity, Hermes, and why AI should inspire better development infrastructure rather than serve as an excuse for layoffs or closed systems. ## Speakers - Anthony Campolo - Rizèl Scarlett - Dev Agrawal ## Chapters ### 00:00:03 - Layoffs, Team Identity, and a Fast Career Change Anthony welcomes returning guest Rizèl Scarlett and co-host Dev Agrawal, who mentions that he is looking for a new role after a brief period at PowerSync. Their opening jokes about how quickly time moves in AI lead into a more serious discussion of recent employment changes and the instability affecting technology companies. Rizèl recounts enjoying her work on Goose at Block and forming a close bond with her team. Although she survived a reduction that affected roughly 40 percent of the company, her manager and teammates did not, leaving her without the group that made the job meaningful. An unexpected conversation with former GitHub colleagues at Entire quickly became an interview and offer, allowing her to leave Block for a new opportunity. ### 00:06:00 - Entire’s Vision for Agent-Aware Version Control Rizèl introduces Entire as an attempt to update version control and collaboration for a world in which coding agents can generate enormous pull requests. Traditional Git records changed files and diffs, but it does not naturally preserve the prompts, tool calls, reasoning context, and human direction that led an agent to produce those changes. That missing context can overwhelm maintainers and open-source communities. Entire’s current Checkpoints CLI listens to agent activity and connects session records to commits. It stores relevant prompts, actions, and metadata on a separate repository branch, where the information can later be summarized or inspected. Dev highlights the value of knowing whether a user made a deliberate decision or merely provided a vague specification that an agent expanded into thousands of changes. ### 00:09:55 - Codex Workflows and Preparing the Live Demo The conversation turns to Rizèl’s daily selection of AI coding tools. Although she remains fond of Goose, she has been experimenting with other agents supported by Entire and has become particularly impressed by the Codex desktop application. Anthony describes running several CLI sessions at once, with separate agents planning, implementing, and reviewing one another’s work. They compare desktop and terminal interfaces, joke about leaving fast mode enabled when the company is paying for tokens, and briefly contend with a connection glitch. As Rizèl prepares an improvised demonstration on a new laptop, Anthony shares Entire’s website and open-source CLI repository. The group also clarifies that Checkpoints has been tied to Git commits even though an earlier version of the website may have emphasized session capture more generally. ### 00:14:05 - Enabling Entire in a Fresh Repository Rizèl begins a live walkthrough by creating a new repository and running the Entire enable command with a non-interactive flag. The setup can activate tracking for several supported agents and create a private GitHub repository, although users may keep their information local if they do not want agent-session data pushed to a remote service. She explains that the first checkpoint should create a dedicated branch for session records while ordinary project code remains in the normal development branches. Because Codex is not installed on the new laptop’s command line, she opens the desktop application, points it at the demo folder, and asks it to create a Node setup README. She then prepares to commit the result manually, prompting a brief comparison of how much Git control each participant delegates to agents. ### 00:17:30 - Commits, Checkpoint Branches, and Demo Troubleshooting Anthony says he still performs commits himself because he does not want an agent to damage repository history, while Rizèl explains that she has long trusted agents to push changes. Her first commit triggers output indicating that Entire is writing to its checkpoint branch, and she opens GitHub to inspect the resulting sessions and repository structure. The improvised demo encounters trouble: the README appears empty on the checkpoint branch, the checkpoint list contains nothing, and GitHub reports branches that are ahead and behind in unexpected ways. Anthony and Dev help inspect the repository state while Rizèl tries another commit, creates a fresh branch, and asks Codex to add a simple HTML file and CSS. Rather than stall on the issue, she pivots toward a working repository that better illustrates the product. ### 00:21:16 - The Team Dashboard and Reusable Session History Using Entire’s own CLI repository, Rizèl presents the web interface for reviewing recorded work. The dashboard displays recent checkpoint activity, agent usage, contributors, and token consumption, while individual sessions expose prompts and responses with controls for hiding unnecessary detail. This provides both an organizational overview and a more readable alternative to raw terminal output. She stresses that the main value is not rereading every old conversation. Captured sessions become provenance that agents can search and reuse later, such as finding how form validation was implemented in an earlier change. A session-handoff skill can summarize unfinished work from Codex and transfer it to Claude Code or another tool. The stored context therefore supports continuity across agents instead of remaining trapped inside one vendor’s chat history. ### 00:25:49 - Turning Past Agent Work into Skills and Memory Anthony notices Rizèl’s session-to-skill project, which converts repeated agent interactions into reusable workflows. She explains that recurring tasks such as creating blog posts often require the same reminders about front matter, metadata, and folder naming. By analyzing successful prior sessions, an agent can package those conventions into a skill and apply them automatically the next time a similar request appears. Rizèl then switches to a repository with valid checkpoints and runs commands that list and explain them. The explanation reveals a checkpoint’s intent and associated transcript, though she prefers the web interface for readability. Dev characterizes Entire as organizational memory rather than memory for a single agent. Rizèl agrees, noting that its Git connection adds durable context to code changes and allows the record to reflect both machine output and subsequent human improvements. ### 00:30:10 - Background Tracking and Cross-Agent Continuity The group discusses worktree support, which lets parallel agents operate in separate Git worktrees while Entire tracks their sessions independently. Rizèl values that the product mostly disappears after its initial setup: it records activity in the background without forcing developers to learn a new daily workflow. Dev argues that this low-friction pattern often produces the strongest developer tools. They contrast that experience with systems requiring frequent configuration, including Goose, whose flexibility can become a barrier for users who simply want to begin working. Anthony then asks whether sessions remain connected when a developer switches agents. Rizèl shows a teammate beginning in Claude Code and continuing through Codex, with Entire labeling both sessions and preserving the overall work. Shared history is especially useful for teams whose members prefer different editors and agents. ### 00:34:46 - Agent Plugins and Intent-Based Code Review Rizèl explains that Entire natively supports a manageable set of agents while offering an external plugin repository for less common tools. The group uses AMP as an example: Sourcegraph’s opinionated coding agent chooses models and optimizes the experience rather than asking users to configure everything. Dev contrasts that approach with OpenCode’s extensibility, while Rizèl finds OpenCode relatively easy compared with Goose. The discussion returns to Entire’s future direction. Its experimental review command can ask one or more agents to assess a change, but the broader ambition is to organize review around intent. Instead of manually reading every line of a huge generated pull request, reviewers could compare the requester’s prompts with the delivered behavior and determine whether the implementation fulfilled the stated goal. This could reduce the burden on maintainers facing growing volumes of agent-authored code. ### 00:39:09 - Bun’s Rust Rewrite and Trusting Generated Migrations Anthony asks for Rizèl’s reaction to Bun’s large AI-assisted rewrite from Zig to Rust. She considers the accomplishment impressive and rejects the assumption that generated code is uniquely sloppy, observing that human-written startup and enterprise systems routinely contain inconsistent variables, abandoned decisions, and difficult legacy structures. Dev reinforces the point by asking critics whether they have seen ordinary enterprise code. Anthony views migrations as an especially strong AI use case because an agent can port an existing implementation while a comprehensive test suite verifies behavior. He argues that no person reads every line produced across a large engineering organization anyway; confidence comes from tests, review processes, and system-level controls. The participants suggest that resistance also comes from discomfort with change and from developers whose identity is tied to typing code exceptionally quickly. ### 00:42:31 - Zig Politics and Languages Designed for Agents Dev notes an apparent contradiction in Bun’s history: Zig was originally defended as better suited to the runtime, yet the project is now moving to Rust even though the languages themselves have not radically changed. The group discusses Zig leadership’s anti-AI position, its refusal of agent-generated contributions, and personal criticism directed at Bun creator Jarred Sumner after the rewrite announcement. Anthony wonders whether Rust is the final destination for machine-written systems or merely the best current option. He mentions ATS, a mathematically oriented functional language with advanced type guarantees, before Dev points him toward Vercel’s experimental Zero language. Zero explores syntax and semantics for agents as primary users, emphasizing deterministic inspection, easy repair, a standard-library-first design, and obvious implementation paths. Its functional tendencies suggest how languages may evolve when machines write most of the code. ### 00:46:39 - Agent-Friendly Languages and Engineering Responsibility The participants inspect Zero’s stated goals and JavaScript-like appearance while considering whether a language optimized for agents needs a stronger type system and fewer ambiguous choices. Dev sees it as a useful experiment rather than a wholly new programming paradigm, but he believes functional programming provides promising foundations for code that must be generated, inspected, and repaired reliably. Even with improved languages, Dev cautions that engineers cannot abandon understanding. Teams still need a strong architectural model, effective tests, and careful review of those tests, even if they do not manually inspect every implementation line. The conversation then returns to practical risks from Bun’s rewrite: projects built around its existing behavior may need to migrate to Node, remain on an old version, or adopt a newer Rust implementation that could introduce compatibility issues. ### 00:51:01 - Open Source, Security, and Dependency Strategy As the formal Entire discussion winds down, Rizèl shares documentation, Discord, YouTube, and contribution links. Her recent video arguing that open source is not dead prompts a wider debate about claims from Cal.com and others. Anthony argues that coding agents rely heavily on open-source code and patterns, though supply-chain risks may encourage teams to use fewer external packages. Anthony describes making his AutoShow CLI dependency-free by calling model APIs through fetch rather than installing several vendor SDKs. He questions whether forking and trimming dependencies is better than asking an agent to build a project-specific implementation. Rizèl agrees that agents reduce the historical cost of writing utilities from scratch, while also rejecting the idea that private repositories automatically improve security. Dev distinguishes open SDKs and CLIs from proprietary hosted infrastructure, a model followed by many developer-tool companies. ### 00:55:55 - Why Open Source Still Matters in an Agent Era Rizèl criticizes companies for using AI as a blanket explanation for layoffs, security decisions, or strategy changes that reflect ordinary management choices. Dev adds that increased productivity should enable organizations to attempt more work rather than dismiss people simply to maintain the same output. They see AI as a reason to improve software infrastructure, not as a universal excuse. Open source may become even more useful when agents can study several proven libraries, identify shared patterns, and create tailored implementations. Dev describes keeping local clones of many OpenClaw variants so agents can compare architectures, an approach connected to his upcoming workshop on building a coding agent. Rizèl has not installed OpenClaw on her company machine but recreated part of its experience with Goose, Discord, and remote task execution to understand the concept without buying dedicated hardware. ### 01:00:13 - OpenClaw, Hermes, and Closing Reflections The group traces OpenClaw’s sudden popularity to a combination of product novelty and controversy after Anthropic restricted related usage. Dev explains that it assembled familiar primitives—persistent memory, sandboxed computer access, and messaging through WhatsApp, Telegram, or Discord—into an approachable package. Its popularity even contributed to Mac Mini shortages, while Rizèl remains surprised at how quickly it crossed from developer circles into wider culture. Anthony and Dev prefer Hermes, citing its clearer architecture and the backing of Nous Research, a team that also trains models rather than merely cloning a popular agent shell. They close by expressing interest in Entire’s forthcoming intent-review features, inviting Rizèl to return after they ship, and discussing another possible stream. The final spoken timestamp is 01:05:15, making the total transcript duration 01:05:15. ## Resources and Links - [entire.io](https://entire.io) - [github.com/vercel-labs/zerolang](https://github.com/vercel-labs/zerolang) - [discord.com/invite/jZJs3Tue4S](https://discord.com/invite/jZJs3Tue4S) - [discord.gg/jyAENMgmb3](https://discord.gg/jyAENMgmb3) - [github.com/entireio/cli](https://github.com/entireio/cli) - [github.com/entireio/external-agents](https://github.com/entireio/external-agents) - [x.com/Teknium](https://x.com/Teknium) - [hermes-agent.nousresearch.com](https://hermes-agent.nousresearch.com/) ## Transcript [00:00:03] - Anthony Campolo And we're live. Welcome back everyone to AJC and the Web Devs. We've got third time returning guest Rizèl with us. What's up, Rizèl? [00:00:16] - Rizèl Scarlett Hey, hey guys, how you doing? [00:00:18] - Anthony Campolo Doing good. We also have Dev who is, uh, my co-host today. What's up, Dev? [00:00:25] - Dev Agrawal It's going good. I'm pretty deep into some AI and framework work these days, but I'm excited to chat more. It's been a while. [00:00:35] - Anthony Campolo Yeah. Are you on the market now for a job? [00:00:38] - Dev Agrawal Yes, I am currently looking for jobs. So if anyone watching, listening knows anyone, hit me up. [00:00:46] - Anthony Campolo Yep. So don't use PowerSync apparently. [00:00:52] - Dev Agrawal PowerSync is fine. I was like, man, I still have this post about PowerSync. I still have this t-shirt on. [00:00:59] - Anthony Campolo Yeah, that's true. [00:01:03] - Rizèl Scarlett Cool. [00:01:03] - Anthony Campolo So, um, Rizèl, um, you know, Dev was showing off PowerSync recently and he was working there for like all of like one month or something. [00:01:14] - Rizèl Scarlett And then they did layoffs. What? What? [00:01:16] - Dev Agrawal Not, not quite layoffs. Uh, no, I was there for about like 3 to 4 months. [00:01:21] - Rizèl Scarlett Okay, okay. [00:01:22] - Anthony Campolo Yeah, it felt like one month to me. [00:01:24] - Dev Agrawal Yeah, flat circle, pretty short. Yeah, although in AI years I was probably there for like 2 years. [00:01:32] - Rizèl Scarlett Yeah, that's true. In AI years, like, I've been at my job for a short time and I'm like, it feels like I've been here for a while. [00:01:41] - Dev Agrawal Yeah, I think there's an, uh, in AI we're gonna need a before Opus 4.5 and after after Opus 4.5 moment, like BC and AD. [00:01:51] - Rizèl Scarlett I know, right? [00:01:53] - Anthony Campolo Cool. So, um, Rizèl, so when you were first on my stream, I think it was towards the beginning of when you had been at Block, and we were talking about DIDs, decentralized identity, um, and then a little bit of AI stuff as well. And then the second time you came on with Ebony, and that's when you were deep into Goose. So now you're no longer at Block. Did you want to speak at all in terms of the layoff situation? Because this is happening now, like a lot of companies. So I feel like you kind of were— what were your opinions on it in terms of how it happened, how it was rolled out? [00:02:32] - Rizèl Scarlett Yeah, I mean, I will say I had a great time working for Block, particularly on the Goose side, because I was like, this brings me back to like the GitHub Copilot days a little bit, but even more advanced. Like, we're not just doing autocomplete anymore. Like, your agent does stuff for you. So that was really fun. And then my team was so fun, like Angie and Ebony and Ace and all of them. Like, I'm like, we really formed like a close bond. So when the layoffs hit, it was really shocking for us. We had, we had been doing little layoffs since the— since I was at TBD. Like, they were doing like little layoffs at a time. Um, but I was like, okay, I understand this is like a really large company because it's— I don't know if people know, like Block is an amalgamation of like Cash App and Square and all these others, and then they have all these little side projects like Boost and TBD and stuff like that. So I was like, it's understandable that like layoffs are happening, but I think that that layoff was really surprising because they laid off 40% of the company and we did not see it coming. I didn't get laid off, but my entire team, my manager, everybody got laid off. So I felt like, this is weird. Why am I here? So I just left. [00:03:55] - Anthony Campolo That's exactly actually what happened to me at Edgeo is that I didn't get fired from Edgeo, but Scott did. And then the rest of the JavaScript Jam team had pretty much been winnowed down to that point anyway. So I was just by myself. Like I joined the team to do JavaScript Jam, you know? [00:04:12] - Rizèl Scarlett Yeah. [00:04:12] - Anthony Campolo So I was just like, this sucks. [00:04:14] - Dev Agrawal Exactly. [00:04:15] - Rizèl Scarlett I'm like, the team makes it really fun. Like it makes, it makes you wanna be there. So now you're like, all right, what am I doing? And then I have like the looming feeling that maybe I would get laid off eventually. [00:04:28] - Anthony Campolo Like people would leave. Did you have, hmm, so did you already have a thing lined up at Entire before you left? [00:04:35] - Rizèl Scarlett Uh, yeah, before I left I did. It just happened to be like the layoffs happened. I had seen Entire post something, like this all happened in a matter of days. I had seen Entire post something and Entire, uh, um, the co-founders, I worked with them at GitHub because the, the former CEO of GitHub is a founder of, um, of that. [00:05:00] - Anthony Campolo Yeah. [00:05:00] - Rizèl Scarlett And then, uh, no, uh, Thomas Domkey. He's like the the most recent. [00:05:07] - Anthony Campolo Oh, she's the CEO after Nat, probably. [00:05:09] - Rizèl Scarlett Yeah, after Nat. Yeah, and then his chief of staff I had also worked with a bit. His name was Cole. And then I was like, "This project's so cool." And then like they messaged me and they're like, "We heard layoffs are happening at Block." And I didn't even know. I felt bad because they were interviewing me. I thought they were just chatting. So I was like, "Yeah, that happened." Da da da. And then they're like, "You want to work here?" I'm like, "Oh wow, okay, yeah." So then that's when I ended up quitting. I really did appreciate my time at Block though. Like, I learned a lot and I think it helps me to bring whatever I've learned over to Entire. [00:05:50] - Anthony Campolo Yeah, that's tight. Yeah, no, that's great. It shows that you really put in the work and that you're a hot commodity now. [00:05:59] - Rizèl Scarlett Hopefully. I don't know. [00:06:00] - Anthony Campolo So let's talk about what is Entire and what made you excited about it? [00:06:08] - Rizèl Scarlett Yeah, so, uh, right now Entire has one— I don't want to like reveal too much, but our whole idea is that like, um, Git is helpful, right? Because it's showing you like the diffs and stuff like that, what files change. That really like, like, um, changed software development because before Git and GitHub, people were like sending emails to each other or writing over like patches and, like, all that. It was, like, really chaotic. But now we're in this new age with agents where, like, agents are bringing in huge PRs. Maintainers are getting overwhelmed. We see every now and then platforms like GitHub are, like, having outages. Or we see open source communities are saying, like, we're closing down contributions and stuff like that because there's just too many, like, agent-authored PRs. We can't keep up with it. So the whole like premise between— behind Entire is to make version control, um, and collaboration with agents and humans much easier. Uh, right now what we have out, we're working on stuff like Instelf, but right now what we have out is something called Checkpoints. And this is— it's really simple, it's nothing crazy, but basically, um, it's a CLI tool that hooks in that has like hooks. It listens to whatever you're doing in your agent. So like if the agent started a session, if you made a commit or whatever, and then every time you make a commit, it creates a checkpoint and grabs all the like prompts that you've written, the tool calls, um, like different things that are relevant to the session, um, that you have with the agent, saves it, stores it in the— in a repo on a separate branch, um, and then Um, from there you can use that data to like, you can tell like the CLI tool to like summarize what you, um, like what, what happened in the, in the session or whatever it is that you need to do to help you to get a little bit more understanding of what, why a certain decision was made by an agent. I don't know if that was a little too wordy, my bad. [00:08:24] - Anthony Campolo That's all good. I'm trying to delete these spam comments in my, my Twitch, um, dev react. [00:08:34] - Dev Agrawal Oh yeah, that— I mean, that's a really, um, yeah, I mean, uh, that's exciting to hear. Uh, I think, uh, well, I definitely sympathize that one of the biggest problems with, uh, AI creating these giant PRs is that it's impossible to tell why it was a certain decision was made if this was like, did a user make this decision or did the user just write a really vague spec and the agent go ahead and make 20,000 changes on behalf of it? Being able to, or having some additional insight into those giant PRs is definitely very valuable. You mentioned that there are some things that you cannot yet talk about. Um, so that makes me even more excited to see what's coming on the horizon. But I'm just, for now, I'm just as excited to show what do you, uh, like what Entire is today, how we might use— like, I'm right now, I'm in the, in the background, I'm kind of setting it up for OpenCode to see how it's, uh, how it goes. Um, yeah, I would love to hear or see more about that. And, uh, I guess, yeah, before that, I would also like to ask, like, what's your Um, like kind of like daily AI usage look like? Like, are you using OpenCode? Are you trying one of the new like Claws? Um, and how, uh, how is it? How's it going? [00:09:55] - Anthony Campolo Goose. [00:09:57] - Dev Agrawal Yeah. Okay. Goose. [00:09:58] - Rizèl Scarlett Okay. I, I was a Goose lover. I still am a little bit, but I haven't used it as much because, um, I don't know. I want to try out the other ones. Oh, and Goose doesn't have an integration with Entires. So I'm like, let me use the stuff that, um, is written higher. I've been using Codex a lot. I, I, it took me a while. You have been too? Okay. I didn't believe people, but everybody was like, go to Codex. And I'm like, I don't know about all that, but then I used it. I was like, this UI is so good. Like, I really love it. [00:10:33] - Anthony Campolo Are you using the desktop app? [00:10:34] - Rizèl Scarlett Yeah, I use the desktop app. The CLI, I use that too, but To me, it doesn't seem that different than other tools. I don't know. I like, I like that. I like that. [00:10:43] - Anthony Campolo It's, it's, yeah, no, I think, I do think the app is really good. It's definitely better than Claude Code's, um, app. I still use the CLI cuz I usually am running like 3 at once. So I'll have my editor open, then the terminal at the bottom. I have like multiple ones open usually cuz I have like one writing a plan and one implementing the plan it wrote previously and one like reviewing the other's plan and just like crazy. [00:11:02] - Dev Agrawal Yeah. [00:11:03] - Anthony Campolo Yeah. [00:11:03] - Rizèl Scarlett You're doing like some orchestration going on. I love that. [00:11:06] - Anthony Campolo Yeah, totally. So you're using Codex. Um, do you turn on fast mode? [00:11:14] - Rizèl Scarlett I always have it on. [00:11:16] - Anthony Campolo Always, right? Just ripping tokens all day. [00:11:19] - Rizèl Scarlett It's not my tokens though. [00:11:23] - Dev Agrawal I think she froze. [00:11:25] - Rizèl Scarlett Oh, you got unfrozen? [00:11:26] - Anthony Campolo No, I still, I still see you. There we go. [00:11:28] - Dev Agrawal You're back. [00:11:29] - Anthony Campolo Okay. I think Dev, you, you probably froze. [00:11:31] - Dev Agrawal Okay. It's not like you didn't freeze for me. [00:11:35] - Rizèl Scarlett Companies. Okay, I don't know. [00:11:37] - Dev Agrawal All right, keep going. [00:11:39] - Anthony Campolo Everything seems fine to me. [00:11:41] - Rizèl Scarlett Okay, yeah, Wi-Fi is weird, but okay, if I'm paying for myself, then I'm a little bit more, you know, conservative with my token usage. But if it's my company, I'm like, you guys told me to use it. Yeah, but, um, you guys want me to show you? Okay, I should have been more ready. I'm on a new laptop, so very good. [00:12:09] - Anthony Campolo I'm going to share some links while you're doing that. Okay, so this is the GitHub repo entire.io/cli, and then the site itself is entire.io. Every commit tells a story. Now you can read it. [00:12:37] - Dev Agrawal Um, I, I think I, uh, visited the entire I/O website a while ago and I didn't see— I, I didn't see anything about commits. Is that like a new thing? [00:12:48] - Rizèl Scarlett Uh, what do you mean that you didn't see anything about commits? [00:12:51] - Dev Agrawal Like, I, I think I visited a while ago and it said that it was a Like it was a way to capture agent sessions across like different agents and different people and kind of like build in sharing. Uh, there wasn't a mention of like Git commits at that point. Maybe I could have just missed it. [00:13:08] - Rizèl Scarlett Maybe, maybe you even saw it before I saw it because I've always seen it. [00:13:13] - Dev Agrawal No, I saw it through your profile. I didn't, uh, that's how I discovered like, oh, that's a thing. [00:13:19] - Rizèl Scarlett Interesting. I can't remember. I don't know. Okay. [00:13:23] - Anthony Campolo Yeah. [00:13:23] - Rizèl Scarlett The checkpoints are always connected with the commit situation. So I guess I could— I don't have a demo plan, but let me— I'm going to just freestyle it, I guess. [00:13:35] - Dev Agrawal Is it a freestyle stream? [00:13:37] - Rizèl Scarlett Yeah, it's a freestyle stream. [00:13:38] - Anthony Campolo Always freestyling. [00:13:40] - Rizèl Scarlett Let's see. Share screen. And— Okay, I'm just gonna do entire screen. I don't know, I was trying to just get like my terminal, but that won't happen. All right, so okay, you guys see my terminal, right? [00:14:05] - Anthony Campolo Yep. [00:14:06] - Rizèl Scarlett Okay, cool. So I'm just gonna show you guys what it will look like to like set, like set entire up and maybe like create a checkpoint or something like that. Hopefully this goes well. Hopefully the demo gods are with me. But let's say I'm going to just create a new repo. I'll just call this demo AJC or something like that and I'll go into that directory. Then since I have Entire already installed on my computer, I'm going to run entire enable -y. I didn't spell entire right, but basically I'm saying that I want this directory to have entire, like track my sessions. When I do -y, that does the non-interactive for me. If I didn't do -y, it would ask me questions, but it set up a private GitHub repository for me. It enabled all the agents. Usually you would pick like, oh, I want to use Claude Code or AMP or Copilot CLI or whatever. But I'm like, I switch between agents a lot. And enable all of them for me. And then it pushed a private repo to GitHub for me. So I can show you guys that too. GitHub.com/blackgirlbytes. And you don't have to push it to GitHub. Like you could stay— oh, I'm not signed in here. Sorry. You could stay Um, local. You could keep everything local, um, but this is just for demonstration purposes. I don't mind if things are on GitHub for me or on remote. Some people are like, I don't want everything that I'm doing immediately on GitHub or at all, whatever it is. But it basically, um, pushed like a— like the setup configurations for me, and it should Oh, it didn't yet. But once I make my first checkpoint, it should create another branch for me. And that branch is where all my agent sessions and different checkpoints and stuff like that will be on there. I can now go back to my directory and I can type Codex or whatever it is that I want to type. Codex is not on here. Okay, I could have sworn I had it, but I think it's because I switched laptops. I'll go to the app then, um, and I'll just find this folder, Documents. Where did I put this in? I think Agent Experiments and Demo AJC. All right, and I think I can just tell it like create a README. Let me zoom in more for you guys and make this little thing. Create a README or something simple that explains I don't know how to set up Node. And then we'll do that. For usually I tell my agent to make commit for me. But just for the purposes of seeing like entire working, I'm just going to like get— I'm going to literally make a commit, like manually do it. [00:17:30] - Anthony Campolo One thing I still do manually, I don't have my My agents mess with doing any commits because I don't want to mess anything up. [00:17:37] - Rizèl Scarlett Really? Oh, I always have. Even like early on, somebody like laughed at me on a livestream. Like, it was like 2025 or something. And I was like, git push. No, I didn't say git push. I was like, push this. And they're like, what are you doing? I'm like, I trust my agent. All right, so now if we go to git status, we see something here. And if I say git add all, git commit -m, Adding a README, it should git push. See, it's saying it's pushing it to this branch called entire checkpoints v1. And then it's probably not going to be anything interesting because we just made a README. And if I refresh, we see this branch here existing and we'll see Um, if I switch to this branch, we'll see the sessions. Not much really went on. And then we have a UI as well. Um, let me show you guys it, that it's easier to like visualize things. Um, well, actually, before the UI, I'll keep showing you guys something in the CLI. Um, if we're in the CLI now, I can run something like entire checkpoints list. I guess I don't have any. I got to make one more thing. Checkpoints will appear here after you save changes. Let's do one more change to see if that creates a checkpoint for us. I don't know. Create a simple hello world app. I don't want anything that's going to take long. HTML app. [00:19:22] - Anthony Campolo It seems like what you pushed up, the README was empty. But it's saying here your README should have 102 lines. [00:19:29] - Rizèl Scarlett Interesting. What? Hold on. Maybe that's why I didn't create a checkpoint. I didn't really check, obviously. [00:19:38] - Dev Agrawal Might have pushed it up to main. [00:19:41] - Rizèl Scarlett No, but it's okay to push it to main. Even if the— [00:19:48] - Anthony Campolo what's it called? [00:19:48] - Rizèl Scarlett Even if the checkpoints are tracking what I'm doing on the other, And shouldn't it have been pushed to main? [00:19:56] - Anthony Campolo It did push it to main, but in the branch that wasn't there. You can see none of the code's there either. [00:20:06] - Rizèl Scarlett Interesting. Okay, let's try again. Let's try again. [00:20:12] - Anthony Campolo It looks like it wasn't tracking the other branches, that it was 2 commits behind and 1 commit ahead. [00:20:19] - Rizèl Scarlett Where do you guys even see that? Let me read it. Hold on. [00:20:23] - Anthony Campolo I saw it in the GitHub UI. Oh, okay. [00:20:27] - Rizèl Scarlett I see this. Oh yeah, it did say 0 files changed. Okay, let's give it— let's, let's try one more time. Thank you guys. All right, so I'm going to go ahead and try this one more time. git add all, git commit -m adding HTML file. If not, I'll just show you guys one that already has it. And figure out what happened, or I'll create a new. Let's create a new branch too and see if anything else changes. Checkout B, new test branch. Okay, and Codex should be on that. All right, create. I don't know what else. Add some CSS. [00:21:16] - Dev Agrawal Almost the least interesting part. Just make it look good somehow. [00:21:21] - Rizèl Scarlett Yeah, well, um, we're not gonna check the, the site anyways. All right, but, um, while that's happening, I'll show you guys some actual— like, some things that actually happen in the UI from other repositories. Um, Okay, so let's say we look at the entire io-cli. That's like our open source repository. So this like CLI tool that I'm using, it's open source. You can see, um, an overview and it shows us like how many, um, like checkpoints or we've had in the last 7 days or whatever you want to change it to, whatever filtering. You could see like the different activity. This is like a team overview type of thing. If you were like, oh, how's, how's my team doing? Or how's myself doing? Whatever, right? Um, The— we've been mostly using Claude Code. Unknown would be like you just weren't tracking your session, um, and maybe it was just like yourself that did it and not a particular agent. Um, and then here's like the top contributors. I've mostly been doing untracked. And then like tokens that are used and stuff like that. I think this to me is just like if you want that like overview of what's going on. But then you can also see— and let me make sure admin mode is off. Perfect. I don't want to reveal anything. You can also see the sessions that have been going on, just a little bit of a better view. I'm just going to pick what Stefan has been doing. No clue what this is. It'll load. But basically you can see he made this prompt that says, based on this, I'd like to switch the auth to the new token in the CLI. You can see the responses that Claude did. This does look like a lot to see. Some people are like, why would I want to really read through some of my prompts or the responses. You can, um, like toggle on and off certain things like the responses or the prompts or whatever it is that you want to do. But to me, it's not really the value of having the entire session and just going back to reading it. It's the, the value of being able to go back to like leverage it later. So like we have like different skills and stuff like that where I can say like Let me use the search skill and I'll be like, there was one point where I implemented, I don't know, form validation or I don't know, some type of thing that I can't remember exactly how I did this or exactly what commit this happened in. Then my agent will use that skill and run different commands in entire and be like, here's how you implemented it, let's try this here. Or there's a session handoff skill as well. I want to be able to show that, but the session handoff skill, Basically, you have something going on, um, in like Codex, right? And then maybe you for some reason want to switch over to Claude Code. You can tell it session— I want to do a session handoff. It'll give a quick summary of what are the things you did, where you left off. You paste that into Claude Code or OpenCode or whatever it is, and it's able to resume the work from there. So to me, it's not like— when I show people this, they're like, Okay, this is just like the history. I don't really go back and read my sessions. And that's true for a lot of people, but it's more like that provenance of being able to use it later on or being able to run. Hopefully we can do entire explain. And I, I don't know if I kept messing up on this. Watch me have done something so dumb. Uh, let's make sure. Adding files. Git push. [00:25:06] - Anthony Campolo Sure. [00:25:07] - Rizèl Scarlett Because what I want— git checkpoint. Oh, not git checkpoints. Entire checkpoint list. Still, I don't know. I'm making a really dumb mistake on stream. But what I do like doing is entire Checkpoint explain. I'll just use a repo that already exists and I'll do that instead of making my own. cd dot dot. Yeah, this is not— [00:25:36] - Anthony Campolo I'm looking at the blog here and I'm seeing you have something. What was it? That session to skill that turns repeated agent sessions into useful workflows. That sounds cool. [00:25:49] - Rizèl Scarlett Yeah, that's a skill that I made. This is a new computer, so I don't have that skill installed, but I use that skill because Or I made that skill because I'm like, I keep doing the same things over and over again. Debra, there's a lot of stuff that are like repeated workflows. Like, okay, I always create a blog post and I always have to tell my agent, did you forget the front matter or the metadata? Like you messed up the folder naming structure or whatever. So I'm like, just look at how I've— look at all the past sessions that I've had where I set up blog posts. And make sure and create a skill out of that. Anytime I say create a blog post or write this doc, it uses what it learned from there. To me, I'm like, it's so cool to have this history and to be able to use it afterwards. If I do entire checkpoint list, and this will be the last thing I show you guys, sorry. Now you can see in the CLI the different checkpoints. These are all connected to commits that happened. I don't know what step I'm doing that's wrong. Like, as soon as I get off the stream, I know I'm going to be like, wow, that was really dumb. But let's say I want to do like entire checkpoint explain. I think you could do that. And it should tell me like what went on in that specific checkpoint, what the intent was. [00:27:10] - Dev Agrawal Yeah. [00:27:11] - Rizèl Scarlett So here's the intent. I guess this person was doing like some kind of review or something like that. And here's the transcript that happened and stuff like that. It's a little hard to read, which is why I often use the UI because it's much easier to just like quickly go through. But this is kind of like what the Checkpoint CLI is. That's it for me. [00:27:33] - Dev Agrawal Nice. [00:27:34] - Rizèl Scarlett Yeah. [00:27:34] - Dev Agrawal So it's like a memory for your entire organization instead of like how some agents like I think OpenClaw and Hermes, they have expose like a session search tool so that the agent can look through its past history, see what— see how it approached something. Or it can also— what's up, Chris? Long time no see. And it can also take your current conversation and package like whatever workflow it went through into a skill and then later recall that. But this is that concept but taken to your entire team, which means likely that there's a lot more useful knowledge in there. There's a lot more insights and a lot more utility. [00:28:18] - Rizèl Scarlett Yeah, exactly. That's a, that's a really great way to, to put it. It's, uh, I hear some people say it's like, um, adding memory to, to Git and your agents. Um, so that, that's a really great way to put it. Okay, Chris. [00:28:34] - Dev Agrawal Yeah, the Git part is important because like most memory systems are like, just look through the session history. But for coding agents, the session history— I mean, in a way that you can look at the session history and see what code changes have been made, but Git is simply a more reliable way to look at code changes, especially like if I as a developer go in and make some tweaks, I definitely want that in the memory that, hey, this is the code that the agent wrote, this is the improvement that I made. And that's what I want in memory, not like this is the code that the agent wrote. [00:29:08] - Rizèl Scarlett Yeah, exactly. I think it's also helpful for— I guess this doesn't happen as much when agents were a little bit less developed. Sometimes I would feel like I want to go back to a change from the past. You know what I mean? I'd be like, never mind, undo that. Then it wouldn't undo everything or do the wrong things. So I'm like, having Git as part of the workflow, I can just run something called entire rewind and it can go back to a specific commit in time or specific checkpoint in time, um, which I really— I feel much safer with that. [00:29:51] - Anthony Campolo Yeah, for sure. That would definitely be something that I would find useful. It's saying here it works seamlessly with get work trees. Each work tree has independent session tracking, so you can run multiple AI sessions, different work trees without conflicts. You can also run concurrent sessions. [00:30:10] - Rizèl Scarlett Yep. That part's been pretty great too, because there's a lot of people that are working like, or doing parallel workflows and stuff like that. So having those, like having it be able to go by work trees is great. Like, I think the nice thing is like Entire is something that really just works in the background, which is why you guys saw me kind of like flop a little bit with the, the, um, like setting it up because I'm like, I don't often— it's not like I'm setting it up often, right? As I set it up, it's working and I just keep moving on. It doesn't like interrupt the regular developer workflow that I have. I don't have to learn anything really that new except for that like initial setup and it's pretty much good. [00:30:53] - Dev Agrawal Yeah, I really like that aspect of it. I think some of the most useful product dev tool ideas are going to come from, uh, like things that can be set up once and do almost all of their work in the background to improve things rather than, uh, me as a user having to constantly prompt it to do something. [00:31:13] - Rizèl Scarlett Exactly, exactly. Or things that need so much like configuration. I love Goose, but I think like one of its— to me, I don't know if other people feel this way, One of its downfalls was that it needs so much configuration to do certain things. And I'm like, that's kind of a turnoff for people. The pro of it was that like it can do so many things, but I'm like, people don't want to be doing setup every day. Like they just, they just want to get their work done, to be honest. [00:31:43] - Dev Agrawal Reminds me of Webpack. [00:31:45] - Rizèl Scarlett Yeah. Yeah. [00:31:50] - Anthony Campolo So since it works with different agents, if you say you start with, you know, Codex and then like you switch over to Claude, will it save checkpoints from both and have them be kind of unified, or will they be like separate from each other? [00:32:07] - Rizèl Scarlett That's a good question. Let me see if I could find it. I mean, this project, it will, it will save it from each, but I want to see it if I show you guys what it looks easier for y'all to see. I want to see if anybody started in a session and then switched over. Okay, maybe something like this. Um, yeah, I think this might be a good example. This is from somebody else, so hopefully I am not wrong. But entire CLI— let me just share my tab. But like right here you can see that, um, my teammate, um, Nina, she started in Claude Code here and then she switched over to Codex. So it, I mean, it was a new session technically, but she's continuing the work there. So it'll just label it as that and continue the, the session for you or continues the work for you. Does that, is that answering the question? [00:33:10] - Anthony Campolo Yeah, for sure. And that's what I was thinking would be, you know, really thing that would make entire valuable is if you're someone who uses multiple agents. And, um, because then it's like, you know, you could be in Codex, you're like, oh, I remember this one session, you go back like, oh no, that was in Claude, so you gotta switch to the different things, like find that session. So if this keeps it all in one, that would be really useful. [00:33:30] - Rizèl Scarlett Yeah, yeah, I think, I think that part is great. Go ahead, sorry. [00:33:34] - Dev Agrawal Yeah, especially in teams where you're, you're almost never going to convince like 7 people in a team to use the same agent. They're always Like, uh, one person's like, oh, I really like Cursor and Composer, and someone else is like, oh, I, I'm just gonna stick to my Claude, it works. So it's nice too that, uh, regardless of what kind of like my— the team chooses, they have this like shared platform. [00:33:59] - Rizèl Scarlett Exactly. I like that too, uh, that it's agent agnostic. You can see everything where you need to see it. Um, we— I think also that I like is like Um, we support a specific amount of agents natively because that's how much like we can manage as a team right now. It's still a small team, but we do have like this external agent plugins repo. So for people— hey, how you doing? Um, for people that are like, I want to use like maybe my like particular agent that like maybe is not like the most popular or whatever, or it's gaining popularity, like you can feel free to create a plugin for that. It's super easy. Like we have AMP, we used to have Pi as an external, and now it's— [00:34:46] - Anthony Campolo it's actually— I don't know what AMP is. [00:34:49] - Rizèl Scarlett AMP is, um, made by Sourcegraph. It's a coding agent made by Sourcegraph. It's really, really opinionated. Um, I don't use it often, so I don't know that much. I know they had, they had ads at one point. Like, to be able to use AMP for free, you had to watch ads. [00:35:06] - Anthony Campolo AMP is a frontier coding agent built for leading models and what comes next. [00:35:12] - Dev Agrawal I think their main— [00:35:14] - Rizèl Scarlett Yeah, go ahead. [00:35:14] - Dev Agrawal I think their main thing is that they don't let you select the model because they want to pick the model and then optimize the experience around that every time. So that's kind of their angle that they want to give you the opinions so that you don't have to spend a bunch of time configuring stuff. Um, like I spend a lot of time with OpenCode and I'll be honest, it's a great app, but out of the box it does very little, which is almost like, uh, for on purpose, uh, because like the plugin system is really good and you can add things. Um, but AMP is like, uh, okay, you, you're not gonna need to configure. You're just gonna get the best thing out of the box that we have built for you with the best model, the best prompts, the best tools. And you just go and use it. I think it was way more popular a few months ago, but it's kind of— I've stopped hearing about it. [00:36:01] - Rizèl Scarlett Yes, exactly. And for those ones that were like, listen, not everyone maybe knows about this agent, but maybe you use it, just make a, make a plugin for it. And making the plugin super easy because we have like a skill that like does it for you. But that's so interesting that you have that experience with like OpenCode because I thought I like that it's like No configuration. Maybe just like in comparison to Goose, like to use Goose, you need to configure everything to start. But like, I was just like impressed. Like, I'm like, I open OpenCode, we got Big Pickle. I could just start using it. [00:36:37] - Anthony Campolo Yeah, that's true. [00:36:41] - Rizèl Scarlett Yeah. [00:36:43] - Anthony Campolo Yeah, I found OpenCode didn't take much for me to get it up and working. Um, you just gotta have like wherever the newest model is and just let it go YOLO. [00:36:55] - Rizèl Scarlett Yeah, I like that. Oh, another thing I should say about Entire. Okay, I don't— and, um, like I told you, some stuff are in stealth, but I've already written about this in a blog post, so I'll just show you guys this part. We're starting to rethink code review as like— I mean, I kind of like touched on it a little bit where like, yeah, I was curious about the review command. Yeah, well, even though— yeah, we have an entire review command. That one's, that one's interesting. It's still like under work, but the whole idea is like you can have an agent review your code. You can have multiple agents review your code by running entire review. But this, we're like going to— we're hoping we can completely evolve how code review happens and lean more towards like an intense review. Being able to see what the agent did, did the agent complete what the intent was from the prompt? That way you're not reading through long files of code. You're like, all right, the user had this prompt. How accurate or how close was the agent's? I'm trying to figure out how to not reveal anything. Accurate was the agent's PR based on these particular prompts. So we're trying to like shift the way that code review happens so it's a little bit lighter for open source maintainers and anybody who has to review code. That's it. [00:38:29] - Dev Agrawal Amazing. Anything that you can do to, to make me spend less time on GitHub.com, I'm all for it. You don't like it, especially the code review tab. Or the code review page? No. [00:38:45] - Anthony Campolo I started using the GitHub PR extension in VS Code, so I have a PR now. I just look at it in my editor because I can get my actions logs and just copy paste it right into my, my agent. Because that was the one thing that every time I have a failing, failing staging test or something, I have to go into the UI, go into the action, get the logs, you know. [00:39:05] - Rizèl Scarlett Oh, so smart. I didn't think of that. [00:39:09] - Anthony Campolo Yeah, uh, I want to get your thoughts, just some like, you know, of the moment things. Um, what were your thoughts on the, the Bun Rust rewrite? [00:39:18] - Rizèl Scarlett Oh, a lot of people, a lot of people feel upset about that. I'm like, go off, like, if that's what you want to do. Like, I thought that's, that's pretty impressive, um, that they were able to do that. I know a lot of people were like, oh, I bet this is just slop, blah blah blah. But I'm like, I don't know, I feel like even some human-created code is pretty sloppy. [00:39:43] - Dev Agrawal So yeah, my question to anyone who talks about slop is that, have you worked at an enterprise, right? Have you seen legacy code or code that was written 2 years ago by random people that don't no longer work at the company? [00:39:59] - Rizèl Scarlett Exactly. I'm like, I've worked at startups where like the company was only around for like a year and a half. They hired contractors. The contractors, it was crazy. Like we had like all these shared variables, but the variables were spelled differently different times. I'm like, everything's always sloppy in my opinion. So I don't, I don't consider anything AI slop. What was you guys' thoughts on, on that part on, um, the Bun rewrite? [00:40:26] - Anthony Campolo I think it's cool because, um, I've been thinking for a while that AI would be really good for migrations because it's just porting from one thing to another. And if you have a complete test suite, then you have a lot of assurances that what you're porting is still going to work pretty well. So I thought that was a really great example of how you can do that at scale. I thought it was funny when people were like, how could you do this? How could you just push all this code without reviewing it, they're like, it's a million lines that no human has ever read. And I was just like, but this is like, software engineering is itself meant to like solve that problem. Because if you have like 1,000 people working at a company like Google and all of them push up a PR in one day, that's like 1,000 PRs worth of code and no one has read all of it. Everyone has just read a piece of it. And the way you ensure you don't break the system is by having like these tests, like all this stuff to verify that it works. Like not because someone can actually physically find every single bug in the code by reading it. [00:41:26] - Rizèl Scarlett Yeah, that's a really good point. That's a really good point. Yeah, I think people are— they don't like change. That's what I think. [00:41:35] - Anthony Campolo Also, people who have a lot of their identity wrapped up in being really good at something like Rust, they feel threatened by it. This is why I think people like ThePrimeAgent who were really anti-AI for a while, because his whole thing is like, He can code super fast and like he just, he just works like a madman. But it's like, that's what AI is actually really good at, is just coding forever. So if you're someone who just like can crank out all these Vim shortcuts, like write, you know, 100 lines of code super quickly, that's no longer really a useful skillset. And now the skillset is more how do we think long-term about our projects? How do we architect them correctly? How do we decide what features are actually worth building versus not building? Because we can build any feature we want now. You know, it, it really actually bumps things up a level to— we can think about things on a higher level, which to me I think is, is really cool. It's, it's all good stuff. [00:42:31] - Rizèl Scarlett Yeah, you're right. It did hit a lot of people's ego. One time I tweeted that, like, I'm like, I don't— I never really cared about typing the syntax. Like, I keep seeing people say they really miss, like, syntax and stuff. I'm like, I just like the dopamine hit of like, I just created this. So true. Yeah, yeah. [00:42:51] - Dev Agrawal I feel like the Rust rewrite, I mean, it makes sense from, so from a certain point of view. And, uh, the, the one thing that maybe it slightly contradicts is like when Bun was started, a lot of people were asking like, why is it not in Rust? Why is it in Zig? And Jared had like some good things to say about Zig at that point that, oh, Zig has some features that are better Especially for this specific use case, and Rust doesn't. So we're using Zig. But now, like, here we are 2, 3, 4 years later, and now it's like going back to Rust. I don't think the languages have changed that much. So it's probably something else. And the other thing that I think is pissing people off is that I think maybe in the last few months, Zig has kind of taken a very anti-AI stance. Um, but like they moved, I think they moved away from, did they move away from GitHub or something? Or was that someone else? [00:43:46] - Anthony Campolo Well, they said they're not accepting basically PRs from like agents or whatever. And, and Jared swore that their decision had nothing to do with that. And I was like, eh, that it had at least a little bit to do with that. [00:43:58] - Dev Agrawal Exactly. Yeah. CompTime was, is what I was thinking of. [00:44:01] - Rizèl Scarlett Mm-hmm. I didn't know this about Zig that it was anti-AI. [00:44:07] - Dev Agrawal Yeah, Zig team has been, uh, has taken an anti-AI stance. They also, like, ever since, uh, Jared moved, uh, Bun over to Rust, uh, the Zig kind of leadership has been like, eh, Jared was never a great engineer anyways. He learned a little bit and then he wrote Zig, and then, and then he put more focus on the business and startup side. He's not— he didn't spend too much time learning and coding and becoming an actual good engineer. So Blah, blah, blah. That there's— [00:44:35] - Anthony Campolo yeah, class. [00:44:36] - Rizèl Scarlett That's— they don't need to say that. Oh my gosh. [00:44:40] - Dev Agrawal Yeah, yeah, it's, it's been brutal. Um, but I am pretty sure that the AI stuff almost certainly has something to do with it, even if Jared is not going to speak, speak out loud about it. Also, in general, I've stopped trusting what I see from Anthropic employees because a few days later they're probably gonna rug pull Um, something. No, Jared is amazing, but Anthropic, I just don't like that company anymore. [00:45:08] - Rizèl Scarlett I was gonna say it's probably powers above him and he just gotta go with it, you know? [00:45:14] - Anthony Campolo Yeah, Peter had, um, an interesting comment on Twitter when he was responding to something I said. He said, um, some languages are faster to prototype, so pick ones that's faster prototype, get the architecture solid, and then port it over to Rust. It's kind of like how he was kind of justifying it. And it's like, to me, I'm just like, but if the agent's writing it anyway, it's like, is Rust the ultimate final end state? Or is there going to be some other language that it could write that would get even more benefits? So I was actually asking ChatGPT, I was like, what would be, what is out there that would give you the same kind of memory safety or whatever from Rust, give you like better performance or something like that? And it was like, no language exists, which is like that. Like, that can't possibly be true. So I kept digging. Down. And it eventually told me about something called ATS, which is called the Applied Type System, which is a multi-paradigm, general-purpose, high-level functional programming language. That's like a dialect of ML. So it's like a— has something to do with like more like mathematical proof solving or whatever. But apparently that can have speed that would be comparable But also can protect you from like certain memory stuff. So I don't know, I'd never heard about it, but I feel like there has to be some better paradigm out there at some point that's going to come around that you can make. [00:46:39] - Dev Agrawal I think, yeah, I think Versailles answered that question. You might wanna check that out. [00:46:44] - Anthony Campolo Zero Lang. [00:46:45] - Rizèl Scarlett Oh, is this the, uh, Chris Tate or something like that? Yes. Okay. I could see that. I did see that. [00:46:53] - Dev Agrawal This is their take on like building a low-level language like Rust that's better optimized for like AI agents that write code, which basically means a slightly better type system. [00:47:07] - Anthony Campolo A programming language for agents explores what a programming language can look like when agents are primary users from day one. The aim of the language is easy to learn on the fly, deterministic to inspect and repair, standard library first, explicit enough that most tasks have one obvious path. Let's see. It's got a function. Looks kind of like JavaScript to a certain extent, but not— and with like a more functional kind of bent. [00:47:43] - Dev Agrawal Yeah, I think the overall kind of, uh, it's like all signs point to functional programming of some sort. [00:47:55] - Rizèl Scarlett This is cool though. [00:48:00] - Anthony Campolo Fascinating. [00:48:01] - Dev Agrawal Yeah, I mean, it's an exploratory, like, experimental thing. Um, I don't think it, it's like super different from an existing language to become like a, uh, like a thing of its own, but I think it's an interesting exploration of like How do we optimize certain languages for AI that writes code? I have my own like kind of side explorations into that idea from different angles. Yeah, but it's nice to see that you can also maybe remake a low-level language. So instead of writing C++ or Rust, you can give agents a better language. I think overall it doesn't really— like, regardless of what language is being used, it doesn't save us as engineers to, like, from reading and understanding every, every bit of it, almost every bit of it. And obviously we don't have to do all of it, but we still need to retain, like, a good high-level architectural understanding of everything, make sure there are good tests in place, make— which means we do have to read the tests, unfortunately. Even if we don't read the code. [00:49:14] - Rizèl Scarlett Yeah, that part's important. [00:49:19] - Anthony Campolo Huh. This is really interesting. Glad you told me about this. [00:49:24] - Dev Agrawal Of course. Yeah. And I guess like one last thing on the Bun Rust rewrite is that a lot of people who bet on Bun for their projects like OpenCode, it's like most of them kind of either have to now move to Node or they have to be stuck on the current Bun version forever. [00:49:40] - Anthony Campolo OpenCode already said they were moving to Node, didn't they? [00:49:45] - Rizèl Scarlett Exactly. [00:49:45] - Dev Agrawal But like, that's another reason why people are pissed off about the Rust rewrite, is that, oh, now they have to either move to Node or get stuck on the current version or adopt like this super experimental Rust build, um, that's almost certainly going to have issues. And that was when Bun was already facing a lot of issues, especially around Node compatibility. [00:50:06] - Rizèl Scarlett That's fair, that's fair. [00:50:08] - Anthony Campolo Yeah, I'll be curious to see how many issues there actually are, because a lot of people are just kind of like assuming there would be. But another person was saying they're like, it's moving from a memory unsafe language to memory safe language. So any things that were old that could have like broken were already in the old code base. So it's like, so I don't know, we'll see. I don't know enough about Rust or Zig to really say for sure. So, but, um, We'll see when the bug reports start rolling in. [00:50:42] - Dev Agrawal Nice. But yeah, it was really nice seeing what Entire is doing. I'm excited to see your take on code review and intent review. [00:50:51] - Rizèl Scarlett Yeah, I can't wait. Yeah, just keep watching us, see what other things we keep creating. Yeah. Thanks for having me, guys. [00:51:01] - Anthony Campolo Yeah. So what are the links people should check out? Like we shared entire.io, the homepage. Entire.io/cli GitHub. You said there's a Discord as well? [00:51:11] - Rizèl Scarlett Yeah, there is. It's, it's, it's not— it's a random string of numbers, but I put docs.entire.io in here. And then here's the Discord. Come through, guys. I'm chatting in here and having fun. Hold on, I'll put this on unlimited and generate a new link. [00:51:32] - Anthony Campolo Yeah, you guys, I grabbed the link from the homepage and put it in there. [00:51:36] - Rizèl Scarlett Great, thank you. Um, what other places you guys should be? If you wanted to check out the open source repo and contribute to, you can go here. There's this, and we have, um, what's this called? [00:51:55] - Anthony Campolo External YouTube with a short you just put on. Open source is not dead. Sorry about that as our last thing to close out because that's awesome as well. [00:52:07] - Rizèl Scarlett Yeah, I want to know what y'all think. I'm trying to do a spicy take a week. Oh, I mean, it wasn't too spicy, but lightly spicy. [00:52:15] - Anthony Campolo Yeah, because I mean, I think it was someone from cal.com wrote an article about how open source was dead or whatever. And I mean, to me it felt, I feel like it's the opposite because the agents write with open source code. It's like you ask an agent to write something and it's gonna pull in open source dependencies. So I thought the just take didn't really make much sense to me. I think if you're looking at like the supply chain meltdown, you could be like open source is in risk of dying because it's gonna be too risky to use open source packages. So it's like, You know, I actually right now the AutoShare CLI, I've made it completely dependencyless. So I'm not using like the OpenAI npm library or the Anthropic one or, you know, Gemini's one. I'm doing everything just through fetch calls to the APIs. And so that's, I think, where things are also. And then I'm also seeing people say like, you should just fork your dependencies. And I'm like, why don't you just like write your own from scratch? I think makes more sense. They're saying you should fork a dependency and then like pare it down to just your uses. And it's like, but then you're stuck with all those legacy decisions. I think it makes more sense just to build your stuff for your project directly and only pull in a dependency if you really, really need it. But with Bun now, there's like, it covers so many things. It even covers rudimentary image manipulation now. [00:53:40] - Dev Agrawal Yeah. [00:53:41] - Anthony Campolo Yeah. [00:53:41] - Rizèl Scarlett I mean, that's the thing. I think we needed dependencies back then when there was no agents, it's like, dang, I don't want to write this all from scratch. But you can literally have your agent just write, create dependencies from scratch really quickly. I think it's interesting. Was Cal.com the one that was like, they're getting security breaches or something like that, which is why they're going private? There was like some company that was like, we're not— we're gonna go closed source because like, we were unable to keep secure. And I'm like, that sounds like a you problem. Like open source is the perfect place to make sure that like your tooling is secure. Just because you make it, put it in a private repo is not going to stop it. Like we look at GitHub, they just got a security breach and their repos are not public. [00:54:31] - Dev Agrawal Yes, that was Cal.com. They took, I think their position was that they had their entire infrastructure open source, which includes like their backend and stuff. And that's the thing that, I mean, a lot of other companies are also, like, especially DevTools, they are open source as well, but they don't open source everything. So, for example, like, PowerSync open sourced the SDK and these, and like source available service. Clerk also like open sourced their SDK and stuff, but the actual infrastructure was not. So, cal.com's argument was that the infrastructure is not, should not, shouldn't be open source. Which is also kind of consistent with how other companies do it, where, um, like the entire CLI is open source but the backend is not, correct? [00:55:18] - Anthony Campolo Yeah, like, yeah, that's what I do with— yeah, that's what I do with AutoShow. Um, so there was Calcom, it was definitely not— they had actual vulnerabilities. It was a convenient excuse in my opinion. [00:55:33] - Rizèl Scarlett Honestly, everybody's using AI as the excuse for things. I just want to say that. Yes, like even with layoffs, they're like, because AI— like, we're doing amazing as a company, but because of AI we have to lay them off. What? That's just— you're just using the excuse. Every single thing you can't blame on AI, it's just you. [00:55:55] - Dev Agrawal Yeah, the worst part is that, oh, the, the AI increases productivity, so instead of having more people do more things we're just going to let people go so that we can keep doing the exact number of things that we were already doing because we have— we don't have any ideas on what else we can do. [00:56:10] - Rizèl Scarlett Exactly. It's just silly. But, um, back to the open source thing, I do— I think, I think what you guys are saying, it's just the supply chain issue. It's just— this is why everyone's like, oh, they're gonna try to build their own GitHub or whatever, because we recognize we need better tooling. Somehow to support this new age of coding. And I think that's okay. Like, it doesn't mean open source is dead. We just need to change a few things. [00:56:40] - Dev Agrawal I feel like open source is maybe even more important because if I have my agent— if I'm going to have my agent write a dependency for me, I don't want it to do it from scratch. I wanted to look at a few existing libraries that already do the job correctly and then copy that to my, like, to my, uh, or like customize it to my, um, needs. [00:57:00] - Rizèl Scarlett That's a good point. [00:57:01] - Anthony Campolo Yeah. [00:57:02] - Rizèl Scarlett Look at the examples cuz agents are really great at following patterns. [00:57:06] - Anthony Campolo Yeah. [00:57:08] - Dev Agrawal Yeah. Every time I, like, I, I, I have so many repos where like I've cloned a bunch of random stuff. Like I have a folder where I've cloned every possible variation of OpenClaw so that I, I can have an agent like go through them and maybe find some like common patterns and things like that. Um, yeah, I'm actually giving a workshop at KCDC called Building— Build Your Own Coding Agent. Um, so I've been— I've also been spending some extra time going and looking through a few of them. Uh, yeah, it's gonna be fun. [00:57:38] - Anthony Campolo Have you used OpenClaw, Rizèl? [00:57:41] - Rizèl Scarlett No, I have not. I, I, I was nervous to use it on my company computer. I was like, what if I get in trouble for something? But I kind of like used Goose to like create my own. So like, I mean, it was just like an experiment. It's not like something I continued to use, but like I had Goose's backend and then like I would like type stuff in Discord and then I would have it like execute stuff for me. So that was like my little, like my FOMO moment when I wanted to use OpenClaw. And I know everybody bought Mac Minis, but I'm not the type to hop on a trend like that and spend my money. But yeah, have you used it, Anthony? [00:58:25] - Anthony Campolo I wrote a blog post where I compared OpenClaw and Hermes, so I used it to the extent that I needed to, to just kind of understand how it works. It seems like Hermes gets you a lot of the same benefits, but it has some kind of additional cool stuff in terms of like memory. And to me, it also just seemed, it seems simpler and like it was kind of more a more clear architecture. And, um, there's like a whole team kind of working on it. And, um, yeah, the OpenClaw, it was, I wasn't really sure because, you know, the guy got hired by OpenAI. So it says like, what's that going to mean for the project? And yeah, yeah, it seemed like a thing that was kind of, I can, I've got a good sense for these things now where it's like when something gets so popular so fast, and everyone was talking about it. I'm always just like, there's an idea here that has caught fire because it's something that is important and unique and different about it, but it's probably going to filter out to other stuff and competitors are going to come around or something else, do it better. And that this thing is going to be just kind of a moment, like a flash in the pan. Qwik is a good example. Who gives a crap about Qwik anymore, you know, like the, the front-end framework. So I think, um, I, I felt like this was not something that I needed to like put a whole bunch of time and effort into. [00:59:52] - Rizèl Scarlett How did it even get popular so fast? Because I was following him like making OpenClaw and I was like, oh, this is cool that he's making what he's building. And then all of a sudden one day everyone's like, I installed OpenClaw. Even like celebrities and stuff like that were like, I'm installing OpenClaw right now. Like, what, what was that trigger? I don't know. [01:00:13] - Dev Agrawal Honestly, I'm pretty sure it started by— I'm pretty sure it started by Anthropic banning ClaudeBot and OpenClaw from their, from their, uh, Max plan. [01:00:27] - Rizèl Scarlett And then everybody started paying attention. [01:00:29] - Anthony Campolo Yeah. [01:00:30] - Rizèl Scarlett Okay. Yeah. Cause he changed the name to MoltBot or something. Odd. Okay, good point. [01:00:37] - Dev Agrawal Yeah, I think so. That was the trigger. But I think, uh, from like the product use, like utility perspective, I think it's also like, like all the— like it, it was basically like a, a set of primitives put together in a certain way that was not done before. Um, like memory channels and sandboxes are the main things that come to mind. Like Like host it in a computer, let it do things, give it its own memory, and then talk to it over WhatsApp or Telegram, Discord. So it was like, like these are things you could have done. You could have set up yourself easily. You could have like spun up a Tailscale thing and used like a UI on your phone. But like just talking to it over Telegram, Discord, WhatsApp, and it was like a, I think that was like the main innovation. And then the rest of it was like the existing primitives kind of just put together. Um, yeah, it's a— and then like obviously the Anthropic blowup and all of that. Um, and I'm sure a lot of stores just ran out, completely ran out of Mac Minis. It blew up. [01:01:44] - Rizèl Scarlett Everybody was going crazy with that. It's still, it's still sold out, like Mac Minis. Like we tried to buy one for a hackathon to give people And it says sold out. That's funny. [01:02:02] - Dev Agrawal Uh, yeah, I, I've been like just using Hermes for the past few months. I'm pretty happy with it. I tried OpenClaw for a little bit like a few days ago. It just like didn't quite work the same way. Maybe I should have invested more time into it, but I'm back on Hermes again. It's pretty nice. [01:02:20] - Rizèl Scarlett I haven't tried Hermes. I should though. Everybody keeps telling me. [01:02:25] - Anthony Campolo Yeah, Hermes is cool. The team behind it is very cool. It's, um, Nous. I don't think that's how it's pronounced, but N-O-U-S. And, uh, News Research. News, that's right. Yeah. And, um, Technium, another guy's name. Yeah, Technium, who is the best follow in terms of the AI space. I've been following him for a while. [01:02:46] - Dev Agrawal Oh yeah, there's a lot of OpenClaw clones that came out. But I think the unique thing about Hermes was that it came from a model lab or like an AI lab, um, where most other like OpenClaw variants are just like random people trying something out or building like a smaller, more secure version somehow. Um, but yeah, Hermes is like, they, they kind of like do their own AI training as well. Um, so they're the ones who built this, uh, That's cool. [01:03:21] - Anthony Campolo All right, we could probably start wrapping it up. Sweet. This was super fun. Yeah, I'll be curious to see kind of where things go with Entire. It seems like it's still pretty early days, so got a lot of fun things coming up. And yeah, I'll probably check it out, especially once the review functionality starts getting built out, because that's something that I really want to spent more time, um, is figuring out what are the best kind of review things around. [01:03:50] - Rizèl Scarlett Yeah, definitely. Um, yeah, it started in like February, so we are really early. Oh wow. But, um, I found, I found the Checkpoints tool itself really useful so far, and I, I myself can't wait for like the, the review to see it. I'm like, hurry up guys, but I know it's a, it's a lot. To create in in a short amount of time. [01:04:14] - Anthony Campolo Nice. [01:04:14] - Dev Agrawal We're gonna need you to come back here and talk about the review once that review features once those for ship. [01:04:20] - Rizèl Scarlett Yeah. [01:04:21] - Anthony Campolo GitKraken review just launched. [01:04:22] - Rizèl Scarlett I just got something. A little competition. [01:04:26] - Anthony Campolo I've never used GitKraken before. [01:04:30] - Rizèl Scarlett Me neither, but it looks great. Like the UI. [01:04:35] - Anthony Campolo I always just like the name. Get Kraken's a good name. All right. Hey, thanks, Chris, for hanging out. Justin as well. Anyone else who is in the chat watching, really appreciate you all chatting and being here. Dev, when do you think we'll stream again? Do you want to do this next Monday or you want to take a week off? [01:04:58] - Dev Agrawal We can do the next Monday. I have— I suddenly have a lot of free time. [01:05:02] - Anthony Campolo Yeah, yeah, um, we'll, we'll decide over the weekend because, um, I might have some stuff to show. [01:05:09] - Dev Agrawal Yeah, me too. I'm gonna have a bunch of stuff to show, most likely. [01:05:15] - Anthony Campolo Okay, sounds good. All right, well, um, check out Entire, um, give Rizèl a follow, and we will catch you all on the next one.