Hamish Songsmith

This

@brankopetric00 @Jeremybtc They all are, spend a day writing a harness and you’ll see they are not that complex. Wild times ahead

@Jeremybtc claude code seemed like a simple terminal app wrapper around their models, wtfff

Anthropic accidentally leaked their entire source code yesterday. What happened next is one of the most insane stories in tech history. > Anthropic pushed a software update for Claude Code at 4AM. > A debugging file was accidentally bundled inside it. > That file contained 512,000 lines of their proprietary source code. > A researcher named Chaofan Shou spotted it within minutes and posted the download link on X. > 21 million people have seen the thread. > The entire codebase was downloaded, copied and mirrored across GitHub before Anthropic's team had even woken up. > Anthropic pulled the package and started firing DMCA takedowns at every repo hosting it. > That's when a Korean developer named Sigrid Jin woke up at 4AM to his phone blowing up. > He is the most active Claude Code user in the world with the Wall Street Journal reporting he personally used 25 billion tokens last year. > His girlfriend was worried he'd get sued just for having the code on his machine. > So he did what any engineer would do. > He rewrote the entire thing in Python from scratch before sunrise. > Called it claw-code and Pushed it to GitHub. > A Python rewrite is a new creative work. DMCA can't touch it. > The repo hit 30,000 stars faster than any repository in GitHub history. > He wasn't satisfied. He started rewriting it again in Rust. > It now has 49,000 stars and 56,000 forks. > Someone mirrored the original to a decentralised platform with one message, "will never be taken down." > The code is now permanent. Anthropic cannot get it back. Anthropic built a system called Undercover Mode specifically to stop Claude from leaking internal secrets. Then they leaked their own source code themselves. You cannot make this up.

@Agrona_ss Everyday. The hard part is I think it’s early. I don’t think people get how fucking wild it’s going to get.

We still build @GoblinRack ?

@GeoffreyHuntley @dexhorthy On an adventure I see….

@dexhorthy oh; you use a laptop? how quaint…

what is stopping you from coding like this

@0xblacklight Less attractive as a headline though

mcp does not cause context rot mcp is a protocol bad context causes context rot mcp can be a transport for bad context just as much as a CLI that's not designed with context in mind can be

@TheAhmadOsman what’s your 2c on the 6000 Pro Max Q. I run servers like r750s so the lower power factor is a big draw for me.

@ibuildthecloud Calling @TheAhmadOsman

I'm so sick of dealing with token usage. Can I just buy an NVIDIA rack and put it in my garage? I am curious what it would cost to run something like kimi full strength. Obviously a lot, but still curious.

Interesting. How does the system identify an action that is permissible versus one that is not. Let’s say I’m having a Claude session investigate a production issue. It wants to read information from a sensitive database, how does the system identify if that’s a permissible tool call or not?

You can read more in our blog post here: keycard.ai/blog/announcin…

Incredibly excited to announce Keycard for Coding Agents - no more copy & pasting credentials or approving individual tool calls. Agents get task-scoped access, so you can stay in flow and actually build. You’re only pulled in when it matters. Yolo mode, without compromise.

@Infoxicador Yes!

Biderectional human to agent interactions on a canvas Thats the new user interface! Glad we moving away from “just chat” and simple static ux components

1. Thank you for sharing how much time you’re spending on the setup of these capabilities. A lot of people don’t realise you need to spend decent effort to build something reliable enough to unlock 10x velocity. 2. Why video over screenshots? Just wondering about the storage tradeoffs and the ability for subsequent downstream agents to interpret the objects?

Finally got the agent to reliably attach a video of it using the browser to demonstrate a passing acceptance test One step closer to a real Code Factory

@Szehiroglu38 No, unfortunately the hackathon isn’t aligned with what we are currently focused on and trying to achieve.

@GoblinRack Are you thinking of participating in the Bags Hackathon? You could evaluate this amazing project there, secure funding, and $ earn some money. $Grasp

so a SOC II approved multi agent AI factory, what does that even look like; - Plane for task tracking and management - Gitlab for change management - Each agent has its own identify across every system - Log and trace capture for every interaction and step of the flow - Agent to agent interactions are captured as a part of PR reviews Its fitting agents to a human world and although its kinda fun, I don't think its particularly efficient or effective

@ChineduAni1790 @thekitze Unfortunately I had some family stuff come up so I couldn't make it. I'll think about the live coding, probably not in the next little while as im in NY next week though.

@GoblinRack @thekitze You never show us anything about this meet-up, was it cancelled @GoblinRack ?, and crypto need some hype and some activities to boost awareness and investors confident, how about doing some live coding?

Tinkerers Unite, @thekitze started pulling the community together and we have our first Sydney meetup. I’ll chat home automation and self built AI coding factories. Coming down to chat AI, tinkering and Lobsters meetu.ps/e/PS22Q/D8gG6/i

@ProductHunt @ycombinator @aaron_epstein YC Application aris-platform.com

We're teaming up with @ycombinator to get builders to launch. Schedule your launch for tomorrow, tag "YC application." and @aaron_epstein will review launches. Top ones could get a YC interview + potential funding. 👇

@DanielChesley This is worth a read about optimising AI agent workflows if you're thinking about this space: banay.me/dont-waste-you… Any chance I could join you guys Wednesday at Bar Snack to banter on this?

With each passing day, it's becoming more and more clear that the SLDC is being rewritten around agent-first primitives The companies that adopt AI coding tools are going to accelerate faster than anything we've ever seen, and those that don't won't just fall behind, they'll cease to exist

@grahams_takes Your distain for all of those frameworks is shared. "good eval data point for planner agents and orchestrators" <- exactly why I'm doing it, I want to create backpressure if it looks like a task is a bit too big, catch it early and break it up

@GoblinRack AGI bad ending: we all turn into scrum masters. just kill me now instead (this is cool and would be a good eval data point for planner agents and orchestrators)

So I'm experimenting with my agent attempting to estimate the token utilisation of a task in the same way humans had to estimate the time taken for a task. Only this time: 1. We have absolute values, there is a correct estimate at the end 2. We can deterministically track each of these data points and use them as inputs to the next estimate 3. This is all in an attempt to best size tasks for AI Agents to tackle. I'm not going in thinking this is going to be a game changer, just a fun little experiment to keep me entertained whilst I do the SOC II grind

@RAZ8ZAR never stops, I say that but it stopped last week with all the network shenanigans, but back to non stop now

$Grasp 4uUB9gLxUdh6jRMr1E5hgHEjQo7Z7gMPCjLHAm4KBAGS The Dev is still building & shipping !! 💎💎🚀🚀

@TopsealerJ Steps closer to the platform being enterprise ready so yes :)

@GoblinRack good thing for GRASP sir ?

@fredeil @GeoffreyHuntley @namespacelabs Not if you get off GH completely, GitLab + GitLab runners, + Zot all local

@GoblinRack @GeoffreyHuntley @namespacelabs Jokes on you, they will charge you anyway

If you are still using runners by GitHub, then you should switch and upgrade to @namespacelabs. a. It's faster b. Observability c. Ya can SSH into your runners to troubleshoot stuff