jaythegeek

OVERRATED: running tons of agents in parallel; working on too many things at once; perpetual context-switching; opening lots of low-quality PRs that may never land. UNDERRATED: using one or two agents at a time; focusing on the task in front of you; thinking deeply; finishing stuff; making your code works in prod.

@noisyb0y1 What were the neighbours doing there?

A regular American developer bought $1,400 worth and stacked seven Mac Minis on top of each other and connected them with metal cables. Neighbors thought he was building a mining server. His wife thought he'd lost his mind. He just didn't want to pay $15,000 a month for a dev team. On the screen - a diagram. Seven Mac Minis connected via Ethernet working as one machine. EXO framework distributes tasks between them automatically. 11.44 TFLOPS each. Together - more than most cloud servers that companies pay thousands for every month. He paid $1,400 for the hardware once. 38 agents from GitHub, 156 skills. A system that learns from session to session and in two weeks writes code just like he does - but seven times faster because it runs on seven machines in parallel. A task that took a junior dev 10-12 hours - the tower closes in 20 minutes. One founder with this setup ships a product like a team of eight people. For $20 a month instead of $120,000 a year. This 7 Mac Mini setup helped him win the Anthropic hackathon and make $26,000 without a team.

@businessbarista @da_fant My take is that you need the brain next to the harness for it to be truly effective. Working on a solution called Jeff for this.

Someone is going to build a worldclass “Brain” for enterprises & make a stupid amount of money. Why? As @da_fant said, “coding w ai is solved bc all context is in the git repo. knowledge work is difficult bc context is spread out. an ai system that creates a git repo w all context for a knowledge worker will be able to 100% automate the work.” When companies talk about being data ready for AI, this is what they’re implicitly saying. Engineering has been prepared for this moment for a long time because of the deterministic nature of code, the centralization/versioning of data (read: GitHub), and AI tools that are largely build by engineers for engineers. But for the rest of white collar work, there’s a TON of catching up to do to properly harness the power of the technology. The big challenge here, and why no one has truly cracked the code for "an ai system that creates a git repo w all context for a knowledge worker" is because unlike code, most knowledge is 1) distributed, 2) unstructured, and 3) unverifiable. It's distributed: transcripts live in Granola. Documents in Notion. Customer Data in Hubspot. ERP. Emails. Slack messages. Random spreadsheets. SOP docs. Etc. Etc. Building an ingestion engine that connects to all of your disparate data sources and auto-updates based on the shelf-life of the data is the first, and frankly, easiest step of the process. Next, it's unstructured: let's say I want to create a proposal for a potential client. To nail the proposal, I want it to pull important information from a variety of sources. The specific asks & background from our initial sales call. Previous proposals to anchor ourselves to a proven format. And completed sprint boards from Linear, so the pricing & timeline in the document is grounded in truth. Whether it's a thoughtful filesystem (a la Obsidian) or an OpenClaw-esque memory structure, the brain needs to be great at self-organizing in a thoughtful schema. This is very hard, especially if you want to build a generalizable brain that can be shaped to an array of different enterprises. And finally, most knowledge is unverifiable: writing a function, running a unit test, and seeing if the code works is easy. It works or it doesn't. Using AI to accelerate your content creation process is highly subjective. What is a good/bad idea? Is the content in your voice or not? Does it feel like slop or novel? Answering these questions are both difficult and non-verifiable. That same system described above doesn't just have to be great at organizing & forming coherent relationships, but it also has to be great at self-improving based on feedback from the user. Memory systems (like those introduced by OpenClaw) are great to a point, but as you scale the corpus of data within your company's brain, things like compaction and cleaning become wildly important to avoid the needle in the haystack problem. Someone is going to figure out how to solve this problem, and when they do, not only will they make a shit ton of money, but they'll be robinhood for knowledge workers, enabling non-engineers to enjoy the sort of leverage that only technical folks have felt for the last few years.

The Claude Code Desktop app is an affront on software. As developers, we should be offended that they chose to ship something this awful. Rushed out my video because I feel like I'm going insane.

@theo Totally agree with this. Every release is rushed, bad implementations and bugs where they just shouldn’t be. If you ain’t using your own shit though…

I have feelings about Opus 4.7.

@sama Sorry that happened to you. What a fucked up world we live in.

I wrote this early this morning and I wasn't sure if I would actually publish it, but here it is: blog.samaltman.com/2279512

@obsdmd this is just mesmerising🫠 @beingjefftheai actively learning from an ingestion cycle

It’s hard to get good videos when your machine keeps stalling on an 11 hour ingestion run. 🤣

Jeff goes to school. Learning cycles, ingesting 40k+ docs and images and it just flippin’ works. Insane. This isn't OpenClaw. It isn't a MemPalace. It isn't another vector database with a trench coat on. This is the result of months of research. Ingestion cycle after ingestion cycle. Pouring over summaries, articles and synthesised learnings to make sure everything in his head is actually accurate. Because a brain full of half-truths is worse than no brain at all. Over the next few posts I'll take you inside exactly how it works. The ingestion pipeline. The promotion from raw notes to wiki. The git-backed memory that follows him from my laptop to my tower and back. The super lightweight harness I now use instead of Claude Code. How our business uses @beingjefftheai everyday via @SlackHQ Thanks to @obsdmd for the visuals ❤️ Thanks to @karpathy for inspiring me to make it better. 🔥

@bcherny i am getting a lot of this output from tool calls and subagents right now. It is causing extortionate token usage too. Latest version of CC installed, happening on Linux / Mac @claudeai

your db query is slow? just add this! boom, now you are ai-native instead of lazy!

Key takeaway from Anthropic announcing Project Glasswing: American firms get access to Claude Mythos... Europe and the rest of the world can jog on... We ALL rely on software and packages not covered by these US based companies. I'm an engineer, I do not dispute that we need to close the holes before releasing to the public. But this is not that, this is essentially creating an even bigger gap in cyber security and stifling innovation at the geographical layer as well as making it a really tough pill for startups and small businesses to swallow. Let's be brutal, Microsoft Outlook doesn't work on Earth let alone in space, don't get me started on Teams. Apple have destroyed what was a functional and beautiful operating system with sloppy design and implementation, Anthropic can't keep Claude online, the five nines are out the window... I could go on. So what is the answer for Europe? We run a severe risk of falling behind in both cyber security and software development in general if these companies have access to models that are light years ahead of ours. Really interested to hear your thoughts on this.

@AlexFinn Completely agree with this. Was the first thing that popped to mind when reading the post

Good news: Anthropic just revealed Mythos- the most powerful AI model ever made Bad news: you'll never be able to use it I get it. It's so powerful that it could exploit cybersecurity But I hate it. I don't love that a company gets to hand select who gets to use the best intelligence. The companies who get access to Mythos will have a distinct economic advantage against those that don't That feels unfair I'm more of a fan of democratization of intelligence. This feels like an opportunity for OpenAI to release something as powerful but put it in the hands of consumers. Trust the consumer by default. Sort of like with the OpenClaw situation Another reason to root for open source

@peter_szilagyi I have had my work account "banned" for over a year, have spoken directly with account managers. Still cannot get in... Have had some fun back n forths with the AI email agent lol

Well, fuck Anthropic. I've bought a 3 month Claude Max sub to a friend as a gift. Sent it to them. 10 days later, their gift is GONE from their account. No trace whatsoever. I go to Anthropic to request a refund: - I can't it's not my gift. - They can't, it doesn't exist. Oh, and you have NO WAY to contact a person who understands the problem, you can only talk to a fucking AI whose job is to get rid of you. It just closes the convo with "End." after you explain what's wrong.

I’ve been building Jeff, an always on harness, with a TUI for coding and a desktop app for having semi-autonomous or fully autonomous task driven development… Gonna write a few posts on the setup, the brain / memory, the harness and so on, here goes… The Desktop Flow A task moves through the lanes, each lane is its own agent (or not) and can be configured to use any harness such as Claude Code or Codex or Jeff’s own harness. Each lane is represented as a thread that can be viewed, taken over, forked or audited. Some lanes are special and handle git worktrees or auto publishing a draft or listening for an event emitted such as a review failing and sending it back for fixes. Where it gets interesting, Jeff is also in Linear as an agent that can have tickets assigned to it, which then processes them in a fully automated manner. Jeff is in Slack too, answering questions, debugging on behalf of FDEs other engineers, running reviews and actively managing the flow of tickets assigned to the "office". Jeff also has a brain, one that I have iterated on since last year and is pretty damn close to being what @karpathy described in his recent tweets. More on this soon. I’m exhausted but proud of what I’ve built so far. Thanks to @steipete @theo and many others for inspiring me. @beingjefftheai

#aurora Peak District, South Yorkshire

@beffjezos The video is edited. Look at the screens, they do not match what is happening!

Humans are already cybernetically enhanced. Not a single person in the crowd opted out of leveraging their perception + memory augmentation known as their smartphone. The notion of what is a human is already diffusing into bio-techno-hybrid territory twitter.com/RaclureOne/sta…

@owencm Detecting an image within another image.

How are people finding the GPT-4 with Vision API? What’s not working for you? What are your feature requests? We’re planning next chunks of work and would love your input!