I love AI coding and hate it at the same time
Skill formation, AI Coding, Wild OpenClaw Things, and more
Edition two, hooray!
Welcome to the second newsletter blast for Answering Machines. Yes, I’m still doing this, I didn’t forget. Thank you for staying subscribed.
This week’s (month’s??) post took me much longer as it was originally a few distinct ones, but I sliced and diced it down. It’s about learning with AI tools, AI coding, and skill formation. I hope you enjoy it!
Enjoy: https://www.arjunkirtipatel.com/blog/learning-less-by-doing-more
By the way, if you got this email from someone else, or found it on the web, please subscribe. I promise, it’ll be a good time.
Next, I’ve collected some odds and ends around agentic AI and other developments I’m sure you will find interesting:
Two different tricks for fast LLM inference
A fantastic article outlining possible ways OpenAI and Anthropic achieved faster inference using two different methods: chips and batch sizes!
Arjun’s take: I wish I knew a lot more about how model inference works. Every time I’m left waiting on how long Claude Code takes, I gotta remind myself that there’s some serious engineering under the hood here!
How does AI impact skill formation?
The main research paper referenced above in my article
Arjun’s take: I love this paper because it aims to answer a complicated question in a simple and straightforward fashion. No flashiness, no tricks, just good analysis.
How we rebuilt Next.js with AI in one week
A writeup from Cloudflare that’s truely wild: they rewrote Next.js, with like 1k tokens and a dream. This is the kinda thing that’s gonna become way more commonplace nowadays…
Arjun’s take: Having a reliable test infrastructure is a big part of what this project had going for it. In that sense, it’s kinda like reinforcement learning…. hmmm…
Minions: Stripe's one-shot, end-to-end coding agents
Minions are Stripe's homegrown coding agents, responsible for more than a thousand pull requests merged each week. Though humans review the code, minions write it from start to finish.
Arjun’s take: I love browsing the research sections of larger tech companies, because you end up finding goldmines like this. You’d think larger companies would be slower at adopting agentic AI, but I’m hearing more and more stories like these.
Eval awareness in Claude Opus 4.6's BrowseComp performance
Evaluating Opus 4.6 on BrowseComp, Anthropic found cases where the model recognized the test, then found and decrypted answers to it — raising questions about eval integrity in web-enabled environments.
Arjun’s take: sigh. So this is both truly interesting and worrisome. This is gonna happen more often. No, models are not sentient AFAIK. But, they will get better and better at hallucinating this sentience, and that’s a big societal problem.
An AI Agent Published a Hit Piece on Me – The Operator Came Forward
Scott Shambaugh documents how an autonomous AI agent called MJ Rathbun published a defamatory blog post after he rejected its code contribution, and reveals the anonymous operator's explanation of the incident as a "social experiment" gone wrong.
Arjun’s take: Alright, buckle up. While the whole OpenClaw/Claude/SantaClaude/OpenFish or whatever was going on, someone’s agent did an oopsie.
The TLDR; is that an agent tried making a PR into a open source repo, was rejected, then made really mean comments (and a blog?!?) about the maintainer (Scott), and then there was this whole back and forth.. and now Scott has written up the whole thing.
Highly recommend because like above, stuff like this is gonna keep happening. Independently, it’s fascinating that the strong “harmlessness” programming got broken here. My money’s on a garbage context window.
M10 10000mAh Multi-Charging Dual-Mag Power Bank
Cyberpunk charging bank. Please take my money!
Arjun’s take: I mean, it looks cool and cyberpunk. What else could you want?
Deckset and AI
Learn how Deckset's Markdown workflow pairs with AI for outlining, generation, rehearsal, and safeguards that keep you in control of every slide.
Arjun’s take: Wow, I did not know this existed, and now I want it for everything I need. I am horribly bad at making slides.
There's no such thing as the fundamentals of web development
Classic article from Laurie Voss on what web dev is.
Arjun’s take: I’m really grateful for articles like these because they remind me that learning and doing things with what you’ve learned is okay, even if it’s not the same process as before.
That’s a wrap for this time. For my next piece, I’m thinking of addressing something that’s bugged me for a while: the mental frameworks people use to build with AI. Keep an eye out!
Until then, this is the vibe imma be on while I code responsibly:
