Max Dignan

@sqs @connorado End of the beginning. Was fun, enjoying cli with many tabs and havent had to look back 🙌

@dexhorthy @AmpCode Haha let’s get it added!

@max_dignan @AmpCode last night I added a feature request to codelayer to add `i` as a synonym for `esc` in certain contexts so I could review context trajectory with one hand and pet the dog with the other :)

The way @AmpCode requires holding the command button when sending requests makes you so much more purposeful about what you pass to context... But when you're coding hardcore while feeding a newborn baby, that becomes a bit of an issue 😂

@xlatentspace @sqs @AmpCode 100% - been using @aquavoice even when I have hands to spare

@sqs @max_dignan @AmpCode need voice mode 😊

@sqs @AmpCode Haha thanks! That mode would be perfect for me these days 😂

@max_dignan @AmpCode Congrats! And maybe we should add a Feeding Newborn Mode that is Enter-to-submit and has dangerously-allow-all enabled. :)

@snipeship @AmpCode 🎯

@max_dignan @AmpCode Sadly (or not) x.com/snipeship/stat…

Ok, I don’t think I’ll ever go back to making hand-crafted, bespoke code. @AmpCode and claude code are too good

@DrNo_21M @claude_code @AmpCode Haha 😂, not yet. Still running api credits for now. It’s working for the moment. I think it’s because context mgmt is so important, that focusing on it has the side benefit of being very purposeful with token usage

@max_dignan @claude_code @AmpCode Aye same, but I'm still on API pricing due to how the model would get nerfed early on pro plan. Are you on max plan (no pun intended) yet?

SERIOUSLY, DO NOT BLINK Now is the most important time to be staying at the cutting edge with SDLC using @claude_code and @AmpCode I’ve had multiple 2Xs of productivity - just this year Can finally focus on market and marketing - you know - what actually brings in the cash 😂

With google, all knowledge is _available_. But it requires absorbing that knowledge, then mapping it onto your task to get the benefit. With LLMs, the knowledge is also available - BUT - it also has cognition. So, you can skip the absorption step and direct it to your task

Tried out SuperClaude today for a really tough UI issue I was dealing with. It was super easy to use, tell it where some screenshots of inspiration were and let the tokens flow

Getting this kind of error. Also, the agent is going off in crazy directions that don't make sense. Wasn't doing this yesterday, may've been a regression

I think the @cursor_ai update I got this morning might've broke something...

@tankredhase @leegordon Very pretty! 😎

Preview of the new StashPay. Design by @leegordon

here sitting in a meeting, a quick cheeky session here to X to post this my laptop is sitting next to me and it’s running two agents cooking on something meetings are no longer expensive for software engineers manager time/makers time is invalidated. that’s the tweet. that’s the ponderoo.

@jordanjhamel 🔥🔥

@max_dignan ty have a few things to polish and publish an update to pypi still today

preview of my x402+CDP hackathon submission .005c for sentiment analysis, .01c for text generation - all endpoints guarded by x402 one liner in FastAPI apps. most of the work was spent in my fastapi-x402 package - works with both mainnets and testnets today

@karpathy Have a post analyzing this attack against the major coding agents: maxdignan.com/writings/the-l…

RT to help Simon raise awareness of prompt injection attacks in LLMs. Feels a bit like the wild west of early computing, with computer viruses (now = malicious prompts hiding in web data/tools), and not well developed defenses (antivirus, or a lot more developed kernel/user space security paradigm where e.g. an agent is given very specific action types instead of the ability to run arbitrary bash scripts). Conflicted because I want to be an early adopter of LLM agents in my personal computing but the wild west of possibility is holding me back.