Will Hopkins

Adopting Claude speak in my regular life, episode 1: Partner: Did you do the dishes tonight? Me: Yes they're done. Partner: Why are they still dirty? Me: You're right to push back. I didn't actually do them.

@glcst @carllerche Hell yeah. Sqlx integration would be cool as well! Being able to pass structs and enums through type-checked queries directly to the db would be awesome

@illothy I first have to convince @carllerche to allow me to add Turso support for Toasty. Likely will do just the standard sqlite types first just to keep it simple and to gain his trust. Then, when he least expect it, I drop this massive bomb of awesomeness with FREAKING TYPES.

Turso now supports Unions and Structs as types (experimental). Should be a lot easier and cheaper to represent complex data woth multiple options github.com/tursodatabase/…

@penberg Check out cmux.dev! Seems similar to what you’re building, also uses libghostty.

Is there an open source version of Conductor?

@thdxr my prediction for the future is thst nobody will shut up and you will keep being annoyed

@samhogan Check out freestyle.sh git service - Our APIs let agents manage branches and commit code, but you also get to keep normal git functionality (and you don’t have to build it yourself)

What if a codebase was actually stored in Postgres and agents directly modified files by reading/writing to the DB? Code velocity has increased 3-5x. This will undoubtedly continue. PR review has already become a bottleneck for high output teams. Codebase checked-out on filesystem seems like a terrible primitive when you have 10-100-1000 agents writing code. Code is now high velocity data and should be modeled at such. Bare minimum, we need write-level atomicity and better coordination across agents, better synchronization primitives for subscribing to codebase state changes and real-time time file-level code lint/fmt/review. The current ~20 year old paradigm of git checkout/branch/push/pr/review/rebase ended Jan 2026. We need an entirely new foundational system for writing code if we’re really going to keep pace with scale laws.

Programming was deeply satisfying work to me. Work for hours/days before getting the payoff of the code working well on your machine. I’m feeling so much friction now to open the editor and do this kind of task by hand, but also increasingly depressed with the nature of work in an AI assisted dev workflow. Back and forth prompting seems to eat at my soul. Need to find a balance that brings back some of the toil.

@summeryue0 @michael_kove Is this response written by AI too lmao

@michael_kove Rookie mistake tbh. Turns out alignment researchers aren’t immune to misalignment. Got overconfident because this workflow had been working on my toy inbox for weeks. Real inboxes hit different.

Nothing humbles you like telling your OpenClaw “confirm before acting” and watching it speedrun deleting your inbox. I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb.

@taheerBuilds @theo If your engineering team doesn’t understand your codebase, you don’t have an engineering team

Here’s the justification for the pricing: Opus 4.6 FAST – Your gateway to lower downtimes TL;DR – Your Claude code slop broke production, and now you need to pay top dollar to Claude Code to fix the slop it created. Because you/your engineering team don’t know what got merged in that 100k+ LoC PR, and only Claude can read it quickly via the PR you merged . Debugging this would cost the company too much time and money at this point, especially with a downed application that nobody on the engineering team even understands. What’s the fastest way to solve this and lose the minimum customer trust (due to downtime)? Use Claude Opus 4.6 FAST As advertised (and I quote): "high-stakes projects, combining impressive speed with Opus-level intelligence." Now you know why it’s expensive as hell: you didn’t read the freaking code it generated in the first place and ran it in claude --dangerously-skip-permissions

2.5x faster, 6x more expensive. Doesn’t feel worth it imo.

Apple Watch gotta be my most regrettable purchase. Why tf can I not get audio playback controls on the watch face? Why can I not use it to control my Mac, only my phone? This thing sucks

A small moment that genuinely scared me. A friend of mine was stuck on a simple bug. Instead of reading the error or looking at the code, his first instinct was to open ChatGPT. I stopped him and asked, what’s the error? He said, I don’t know, I was about to ask AI. I told him to just read the error message. Nothing fancy. Just read it. He hesitated. Then actually looked at the code. Twenty seconds later he said, oh… wrong variable type. That’s what scared me. Not that he used AI. But that thinking wasn’t the default anymore. I’m all for using AI to code faster. But there’s a big difference between using AI to speed up your work and using it to avoid thinking at all. I see this everywhere now. Every bug becomes a prompt. Every confusion gets outsourced instantly. But the struggle is where you build intuition. That’s how you learn when AI is wrong. If you never debug simple things yourself, you won’t know what to do when AI hallucinates on harder problems. We’re confusing efficiency with dependency. Use AI, absolutely. But read the error first. Try the obvious fix. Think for thirty seconds. Because the moment you stop solving small problems on your own, you slowly lose the ability to solve big ones.

@AdamRackis That being said, models trying to take shortcuts is still a constant source of frustration for me

@AdamRackis For complex tasks I like to have it make an implementation plan (plan mode for priming the session, just normal requests after that) and iterate on it before implementing. Then you pretty much just have to catch it when it tries to deviate from the plan or take shortcuts.

Where are my gainfully employed software engineers who use AI tools (vibe coders just sit this one out - here’s some crayons to play with) Am I using it wrong, or is it just faster to do smaller / non-obvious tasks by hand, vs writing a clear prompt and then verifying?

@canvardar Huge news for people who pick languages based on groupthink

remember when everyone was talking about rust vs ts? ts won again. rust hype has died

@jasonzhou1993 Can’t decide if this is bait or not because it feels obvious, but give Neovim a try!

I want a terminal first IDE Nowadays, I mainly open Cursor to view file changes made by Claude code But its terminal experience still not great Anyone building this?

@om_patel5 The guy clearly is way out of line but ya gotta admit, login being redirected to localhost when people are paying is pretty bad…

someone just tried to destroy my startup and reputation over a 30-second bug fix despite immediately fixing the issue, he > called it a "complete scam" > threatened chargebacks > threatened to contact stripe to "remove our integration" > said he'd spam "every hackernews article and reddit post" > even threatened to contact my professors at MIT (???) all in a span of minutes after purchasing a subscription i offered a full refund immediately. no questions asked. problem solved, right? wrong. even AFTER processing his refund, he continued with threats and hostile messages. some people think they can bully you because of your age. to other founders dealing with hostile customers: > stay professional always > document everything > offer refunds when appropriate > don't let threats shake you building startups is hard enough without people trying to destroy your reputation over a 30-second technical glitch that was immediately fixed. looked him up and he's actually a "respected founder" in the community. imagine being so entitled that a 30-second technical glitch makes you try to destroy a 16-year-old's entire future.

@peer_rich Fuck PG&E, all my homies hate PG&E

new tax proposal: NPS-based wealth tax if you‘re a billionaire and you created a product people love you get a wealth tax exemption but if you run a company with terrible NPS like Comcast or you became a billionaire by (legally) scamming people (crypto, SPAC, etc.) you should pay a heavy wealth tax 🤷‍♂️

@thdxr They’re also too conservative about complexity - some things are just complex but the models will always try some sort of shortcut instead of doing what is correct (even if there is a known correct solution)

you might have noticed lately some models are too conservative about bash output and will append `| head -n 100` to running your test suite this sucks because your failures might be at the end and then it has to rerun the whole thing because it's baked into the model it's difficult to undo this behavior there needs to be way more thought put into which layer problems are solved, putting everything on the model isn't the right approach

@craigzLiszt There can be a healthy middle ground where everyone is happy and your non-slopper engineers don’t have to suffer!

@craigzLiszt Sure, until there’s a bug and the codebase is so messy and poorly documented that debugging it is a nightmare, or you have to onboard someone new. Shipping fast is great but if things aren’t maintainable it will inevitably be a bottleneck imo. It’s just a matter of time.

I'd rather have a codebase full of ai slop if it means we can ship 10x faster