greypixel

gm Austria! 🗻

@jamesonhodge Infinite canvas is not the answer. We just need a way to batch specific windows from different apps together into a context and switch between those contexts before switching within them. As much as people love to hate on them, Mac OS spaces are very close

it's been cool to see people experimenting with infinite canvas UIs in their agentic IDEs... but the UI problem people are trying to solve (managing agents, terminals, browsers, code across multiple projects simultaneously) is the same problem tiling window managers solved decades ago: concurrent temporal streams that need fast switching, maximum information density, and zero spatial overhead. infinite canvas works when position encodes meaning, like workflow DAGs, architecture diagrams, brainstorming. but most agentic work rarely has meaningful spatial relationships. where an agent sits on a canvas relative to another agent isn't meaningful information. you just need to know what's running, what needs attention, and how to switch to it instantly. hence why VS Code's tabs and tiles still work: boring, battle-tested, and nobody has to think about where to put things that said, UI is still very personal and experimentation is great, and it's why i'm building my own agentic ide. i'm just saying don't underestimate unsexy defaults

@mattpocockuk When you forgot to add “make no mistakes”

Working on a tool that orchestrates locally sandboxed coding agents in TypeScript - Sandboxed in Docker - 100% offline: commits made in the sandbox get patched back to the host - Build complex workflows in Typescript - Claude, Codex, OpenCode It's called Sandcastle

Apps and services will provide separate agentic interfaces for the tasks that lend themselves to agentic execution. UI will still exist for humans. Some apps will also embed and expose their own agent - to use your analogy they'll be the artificial contractors and consultants hired by your artificial employees. Having an agent using a human UI will be possible, but it will be a "retrofit" solution that is clunky and inefficient.

@greypixel_ what's you vision?

Omg the example has as it setting system permissions 💀 This is fine.

@__morse That sounds like a very short term solution to me. It’s like training a humanoid robot to drive a car.

@greypixel_ I think agents will have their own computer long term. also their own accounts, credit card, etc just like artificial employees

@__morse For good reason! We should not be encouraging people to give unreliable AI models admin passwords. It is okay to have to type in a password for this stuff, it’s not like you need change these settings every day.

@greypixel_ I used this example because it's the only thing you can't do via command line on macos!

@youyuxi As someone who uses a telegram group with topics for work, trust me when I say it could be much, much worse!

How does everyone put up with Slack honestly the most confusing piece of software I have used for real time communication - Got channel invite notification from Slackbot - link goes nowhere - you need to accept it from your email! (And Slackbot doesn’t tell you) - Got a DM from 3h ago, try to reply - failed to send message because the invite to DM has expired?!

AI psychosis at its finest.

@crypto_condom A truly beautiful gift. 💝

Saturday is my mother's birthday. She is the strongest woman I know. She raised 3 smol pepes by herself after being widowed at 32. She worked two jobs for much of my childhood to always give us what we needed, if not what we wanted. This year, her birthday gift is a digital album of about 500 old 35mm slides from before her husband and daughter died. With the help of @gemini, I have been able to scan, upload & enhance all the slides so that they were photorealistic and color corrected. It is mind blowing what good AI can do if you use it well. 🍌

Limitless Agent Role Playing LARPing.

@AdamRackis Pick your battles. I think this one is not worth your energy.

"jr dev who types fast"

@camsoft2000 Yeah this is a super familiar feeling. I think from now splits (competent) software devs into two camps - the ones who accept slowing down to maybe 1.5-2x of pre-ai speeds, and the ones who gamble and turn into pure vibecoders at 10x.

I’m getting to the point with one of the projects I work on where the complexity of AI slop is becoming a real issue. While I can still happily prompt the agent to add x feature and it will do so and it will likely work perfectly, the code is just getting too complex and fragmented. Agents love to copy and paste and keeping patterns DRY is a real challenge. The agent will start diverging all those copy and pastes until you’ve got loads of similar but slightly different blocks of logic. Again it all still works and solves the problem I’m after. But I just can’t get any kind of consistency anymore, the code is a mess and I just don’t have a handle on it. I want a clean unified architecture but agents just code with tunnel vision. The project is now too big and complex for an agent to fully reason with and too big and complex for me to reason with. The only real solution is a complete rewrite. Maybe this is the way things will go. Code will just become disposable. I don’t really want to care about the code and to be honest I don’t but I do care about consistency and maintainability and the AI slop is hurting those very things I do care about. I know some will say “I’m holding it wrong”, use x,y,z skill, tool whatever and already use tools and anti slop skills, plans, docs, etc but the outcome is the same. Vibe coding something into existence is truly magical. But turning it into a mature product with months of iterations is painful. I can’t even hand code this thing because I don’t understand the code anymore and I’m too lazy to try and code myself because I’m addicted to AI. So what’s the solution, either start again and accept that’s just the way we have to roll, or just carry on fighting the slop and accept each new feature will take longer to implement than the last. I’m tired. I’m addicted.

UI is here for the long term but I don’t think we click buttons as much. Agents will interact for us directly on the site. Probably voice first, with the option to type. Think “add the ingredients for pancakes from this recipe to my basket - enough for 8 people”

I bought a gesture and returned it after a week. It has a slightly forward sloping seat and for me it just felt like I was constantly fighting the chair to stay in it. I came to the conclusion most of these ergonomic chairs are way overhyped and overpriced for the majority of people

@BenjDicken Steelcase gesture and it’s not close

What's the best $2k office chair and why? Aeron vs Embody

@thomasjeans @karpathy @RhysSullivan Disagree. At the moment you cannot trust the output enough not to read the implementation (unless you have extremely thorough tests, which you also have to read). If you can’t review the code effectively then you’re just going to end up as a vibe coder

the paradox is that we are trying to impose our individual design preferences and in some cases style on code that realistically will only need to be understood and maintained by agents most tech leads have learned this lesson with human teams at some point, you let the doers earn a level of autonomy within the guardrails

Thank you Sarah, my pleasure to come on the pod! And happy to do some more Q&A in the replies.

i'm honestly amazed by the valley big wigs, who used to be engineers, are now doing trivial shit with their agents as a hobby, and thing what they do applies to SWE. this also means they will have extremely bad judgement when it comes to investments. quite hilarious.

If you cancel you @AnthropicAI subscription because of this be sure to let them know the reason. Time to give @cursor_ai's new model a try!

Thank god for this. Imagine if all my keys and secrets being sent to china bloated my context window, that would be disastrous

Jfc.