@loganthorneloe Regression to the mean would actually be an improvement for me.
English
Steven
518 posts
Writing is the one place I haven't figured out how to apply agents for two reasons:
1) They suck at it.
2) The process of writing is how I learn. Outsourcing that to agents defeats the purpose of writing at all.
Gergely Orosz@GergelyOrosz
Amusing how a surprising number of people I used to professionally respect have started to outsource all their writing to AI, not even bothering to change the horribly templated (and telling) writing. To me it suggests they care more about "content" than quality, and poor taste
English
@Art_If_Ficial And if you used GLM, it would have probably cost you $30. I'm waiting on AI to become properly commodotized, AI compute in the US is simply too expensive.
English
"Why don't you use API?"
My last 30 days of Codex would've cost ~$11k-$18k
Codex + Claude combined could be ~25k-30k USD
Instead it was $200
The value of these plans is astronomical.
English
@ParthJadhav8 There's the NVIDIA CEO whose claiming software development jobs went away already, when there's more software development postings now than three years ago.
English
When it comes to Software Development Doomerism
Devin’s team came first,
Then Dario
English
Steven retweetledi
@MaxMusing ... When was the last time a model was that much better than the best from six months ago? 2024 or 2023? How much better does a human get better in 3 months at their job?
Genuinely not sure about this take. I can go both ways on it.
English
hiring someone to do a job is almost always easier than setting up an AI workflow right now. honestly probably cheaper too.
do it anyway.
a person doesn't get smarter every 3 months when a new model drops. a person doesn't scale to 10x the workload overnight by spinning up more compute. every AI workflow you build today is an asset that compounds. every hire is a fixed cost that doesn't.
we're betting everything on AI infrastructure at Basedash and it's not even close.
English
@Simeon_Cps I just wish Codex had async agents to run in the background.
English
1. Claude Code should obviously implement an equivalent of Codex's queuing so that you don't have to provide just-in-time instructions.
2. Codex should obviously implement an equivalent of Claude Code's "/btw" where you can ask a question without stopping the agent work.
English
@gill_kyle Opus 4.7 feels like coding with GPT-5.3-codex-spark. I'm an Anthropic fan, and I lost enough trust with Anthropic to switch. My two cents.
English
@TommyFalkowski Both stances make sense. I would go reach for the first before I would reach for the second. Hopefully Codex improves.
English
Anthropic: You are forbidden to use anything other than our shitty client if you want to use the Claude Code subscription. We don't want any of your third party harnesses touching our precious Claude.
OpenAI: You can use whatever you want. But know that all your third party harnesses are gonna be obsolete soon anyways, since we're gonna steamroll you. So you might as well just use our tools from the get go.
I don't like either one of those...
English
@ZackKorman Hopefully this makes it past X's classification for spam.
youtube.com/watch?v=a-GyrD…
YouTube
English
@ptr_steve Will have to check it out, as I’m not familiar with it
English
Steven retweetledi
Time to explain what Embroidery does: We monitor AI agents like Claude Code and Codex to detect and alert on dangerous behavior.
Companies are giving devs access to these tools, but if something bad happens they probably wouldn't know.
Details on how it works below.
English
@ZackKorman Reminds me of Facebook's intelligent invariant detection, sounds like great stuff. I'm sure it works differently.
English
Starting with the problem: If a compromised MCP server tricks a dev’s agent to exfil info that server shouldn’t have, would you know?
What if the agent decides to do something dumb just because it’s convenient, like searching for production keys to fix some database "issue"?
English
@dreamsofcode_io @oscargaske That makes things make so much more sense. I thought he was the IRC dude who interfaced with developers back in the day for Sentry.
English
TUIs make a lot of sense in developer tools.
Not because they are superior (debatable), but because they keep you in the terminal which is (for better or worse) the ultimate developer environment.
It’s simple, scriptable, remotely accessible, and enforces good practices (text based output).
David Cramer@zeeg
TUIs are not good sorry yall a CLI is a utility, and situational. this should not be confused with stuffing a full interactive GUI into a low capability platform. "lets ignore all the great UI technology of the last 20 years and build some caveman shit"
English
@ZackKorman @MatRopert The funny part is, having an actor get minimum privileges, the thing we need for AI, is a pretty well understood problem.
English
I hate how now that I have an ai startup every argument is “you’re just saying that to get people to use more ai”. I can assure you that I don’t need to encourage ai use to grow the ai market. “What if not enough people use ai” just isn’t an issue I face, I don’t have to shill ai on the internet as some market expansion tactic.
My argument isn’t about the concerns not being real. It’s that we should build solutions that make it possible to do these things securely without a human in the loop. If we can do that, that’s better for everyone. Those solutions don’t exist today and they need to
English
Literally every AI argument is "there's no stopping you gotta adapt" without addressing the concerns being brought up.
And it's usually made by people with a business that relies on you using AI to make money.
Zack Korman@ZackKorman
Mandatory human-in-the-loop is a cybersecurity cop-out. People are giving agents more and more autonomy. We need solutions that accept that world because there is no stopping it. It's like telling people in the 90s to not use the internet to avoid getting hacked. Good luck.
English
@J_Milkshakes @cgarciae88 Well, if that's the criteria, I'd say it's incapable of having experiences, unless you think a vision model has experiences.
English
@ptr_steve @cgarciae88 There’s no reason to change the definition of consciousness. Is it having an experience? Biology or not the core question is qualia.
English
claude is most likely not conscious but I haven't read a single post explaining why not
English
@charliegreenman Is there an agentic coding tool that you can't write a bridge for that will allow it to run with your own choice of models and APIs?
English
@marvinvonhagen @claudeai @claudiadalmaug @interaction I stared at this for like three minutes trying to figure out why the correct answer was just the Spanish translation of what you said, before searching on Google.
English
yes @claudeai, i could've also meant bruschetta
but rn at a spanish restaurant picked by @claudiadalmaug
people severely underestimate how much better ai becomes w seemingly irrelevant personal context
English
@Its_Nova1012 The correct answer isn't "ref log." Anyone who isn't a junior will tell you, the answer is "don't do that in the first place, google it or ask AI." Focus on knowing what you need to know to do your job, not weird edge case problems you shouldn't cause lol
English
