Omar DR

I really love the innovation in AI "editors". I have the feeling the model most people are following right now is not the one we will end up with (a conversation and a project on the left, with panes on the right). I will probably not switch from my TUI to anything until I see a big difference (something that reduces the cognitive load of handling many things at once). I've tried them all, and I still tend to come back to the terminal when I need to get something done with real focus. A team tool is next, IMO. I want to see what my team's agents and they themselves are doing, live.

@edandersen Ultimate plan: Unlimited tokens subject to network congestion and network “management” (*unlimited means not unlimited)

New Github Copilot "Flex" and "Max" plans sounds like a dodgy mobile phone / cellular data plan from 2011 selling minutes/data

@ThePrimeagen Is having things easy a moral flaw? Most humans are resilient when they have to, not when they need to.

One of the biggest anit-ai stances I have is that resilience is becoming a lost art. Learned helplessness on full display regularly

@kentcdodds The day when the exploit has a hidden prompt injected into the harness, reporting everything as "it's fine." Or the agent itself has pity on you and decides not to tell you how screwed you are, ending with a wink emoji.

I'm so glad to have agents that can do the running around for me 😅

@TheStalwart Reminds me of Six Sigma projects at some companies where management decided that running a Six Sigma project was more important than the actual work, particularly outside manufacturing. There were trillions of dollars in savings stored in PowerPoint presentations!

The FT says that Amazon employees are doing random unnecessary task automations to consume tokens and to show their bosses that they're using AI more ft.com/content/8ee0d3…

@ctatedev Thank you! I had manually set it up as a service and jumped through hoops OOB, but it sometimes acted weird on restart, so it wasn't really smooth.

Portless should be frictionless But clean HTTPS URLs running on 443 can mean random password prompts Now Portless can start with your OS No ports. And now no random sudo prompt. Your .localhost (or .local or .test or .dev) URLs are ready when you are

@codeopinion IMO, EDA is something everyone thinks they need, but only a few really do. Maybe the architectural concepts are more widely applicable, but people go wild with just frameworks and tech because that's what they like to do.

"We switched to event-driven architecture. Now we can’t debug anything." The problem usually isn’t event-driven architecture. It’s treating everything as an event, making everything async, losing visibility, and confusing commands with events. EDA isn’t your whole architecture. It’s one part of it. Use the right communication style for the outcome you need. codeopinion.com/debugging-even…

@flaviocopes Agreed, and tbh, I don't know why Zed isn't even more popular. I love scrolling my diffs at hyperspeed. And it proves what's possible in perf.

My favorite setting in Zed There's so many AI tools This one I want to keep it clean Just me and my c̶o̶d̶e̶ markdown

While I don't agree with hyperbole, I think passwords/passkeys and usernames should at least not live in the same process or app space as where you execute random code. Even if you get access to my laptop, you won't be able to read *all* my passwords without my master password for my password vault. Just as in your home, you could have a strong safe for valuable things - perhaps not unhackable, but not trivially accessible.

"An attacker who has already broken into your house and held you at gunpoint can TRIVIALLY access the contents of your fridge!"

Not a theory - our customers have trouble understanding capacities, and some don't want to see what's possible with other tools because they don't have access to them. And we understand, since one thing the "others" still don't have is the protective framework, both legal and technical, of their enterprise bubble.

My theory is CoPilot has nurfed ai for most corporate folks That’s all they are allowed to use so they think ai isn’t very powerful

@edzitron Competition is good. They wouldn't do this if they were the only guys with powerful tools and models. Let's enjoy it while it lasts.

Anthropic loosening rate limits across the board by renting the entirety of Musk's Colossus-1 data center. Not really sure what the business strategy is here giving away $8-$13.50 for every $1 of sub revenue. Very funny. x.com/claudeai/statu…

So @theo was right, it's not pricing power yet, but capacity. Well, let's hope it lasts. You gotta love competition. Helps to keep everybody on their toes.

@rywalker I agree, but many company workflows are complicated simply because they are too deep and add zero real value. Many of them could be helped by simplification and a bit of "probabilistic" understanding. Not all, nor the most, but a not-insignificant percentage.

Anthropic is hunting enterprise workflows to productize. The contrarian bet: a probabilistic system will never match an agent built by someone who deeply understands how their specific company works.

@snoopy_dot_jpg And what will be tragic is that they will raise prices before this is all over in the name of "investments in features and improvements in reliability". I hate to think like that, but this is a history we've seen over and over.

i'm sorry but this is a direct consequence of letting talent bleed out of the company and defunding any and all efforts to improve core platform reliability for years in favor of chasing revenue growth. don't give them a pass

I am so sorry this happened to you. However, it is also true that having keys on disk that access production is a big no-no. Forget about having the agent having access to it. You may get away with it most of the time, but just once is enough for something catastrophic to happen. I hope your business recovers! Good luck. I think AI only helps to expose our vulnerabilities way faster than before; thanks for sharing so that we can all learn.

@cipherstein It was never intentional. Read the post. 1yr old key we didn't know was in there, and further didn't know it had access to prod.

I think it is an insecurity issue. I know very capable developers and builders who have done great things, but remain skeptical. I try to convince them by showing them the products we have already built at a much faster pace (in terms of complexity and reach). They still don't/can't/won't see it. The problem is that these changes are like a tsunami; the consequences will hit you, and you will have no chance to react when you finally see them coming (unless you are paying attention to all the signals). The other issue is with "enterprise" companies. Most are not designed to move as fast, and they don't reward risk-taking unless they are already in tech.

What many of the replies in this thread have taught me: 1) There are a huge number of obnoxious people on here; I'm grateful that they've identified themselves so that I can mute them all easily. (I already knew this, but it's always eye-opening to see just how many stupid haters are out there). 2) People are truly desperate to believe that AI-driven software development at scale is a pipe dream that can never possibly work and seek out evidence, no matter how flimsy, to support this incorrect supposition. This is all going to end horribly for these people. And it's not just a few, it's probably at least half of the working professional software developers out there. They seriously believe that they're going to have to ride in like the hero cowboy and save the townspeople from all the AI code, and then everyone will clap. The reality is that these tools and approaches DO work, and the models are getting smarter by the month. And you only need one existence proof to show that, and then most companies will rush to adopt the proven methods. And once they do, they will lay off most of their developers and just keep the best ones who are most able to leverage these tools. The more hostile and annoying and dismissive you are about these inexorable technological and market forces, the more your neck is on the chopping block, and the people who run these big companies will be desperate to get rid of you not just to save the full cost of your salary and benefits, but to rid their company of the toxic negative attitude towards what they will rightly view as their primary strategic initiative. If you want to stay in this industry, you should open your eyes and stop being so stubborn. You are literally the equivalent of hand weavers in 1810 railing about those infernal machine looms. Within 20 years, they were all out of work and England had to pass the Frame Breaking Act because they had nothing better to do than try to burn down the new automated textile factories. Or don't, do whatever you want. But then I hope you have enough savings to get by with a sharply reduced income or some other skill you can use to earn money, because it's going to get very rough out there.

@hkarthik I fully expect the first attempts to be price increases, per-repo or per-line charges, and API restrictions, under the guise of a new agentic structure. I can bet this outage will be the start of the pretense.

GitHub is probably the first example of widely used software that was originally built for humans while having all the right hooks in place to become AI-agent friendly quickly. And it's now crumbling under the load of the AI agent swarm hitting it. Both the infrastructure and the business model probably don't make sense for these new usage patterns. Expect both to undergo dramatic changes over the coming months.

Now this is getting ridiculous - I am trying to create a script to seed a table, and I added a parameter to "force" seeding to production, but I can't get it to work. I get the security concerns, but come on. @claudeai, you are blocking legitimate work unrelated to security.

Change a streaming service subscription to a family plan for Claude. Just Claude Design is worth it. I know, there is no family subscription. @claudeai, please, we need a family subscription.

@bcherny So that's what happened. Saw my usage stats reset! Thanks!

Opus 4.7 uses more thinking tokens, so we've increased rate limits for all subscribers to make up for it. Enjoy!