Scott @UseAIGuy

Setup Opencode last night and changed my Paperclip over to Kimi 2.5 and canceled my claude code max plan. Honestly, seems to be just as effective so far for my needs. So much cheaper too!

I agree with all of this. At the same time, it has been (and continues to be) obvious that things weren't going to stay as they are and anyone building anything reliant on Anthropic or OpenAI was going to get rug pulled and get squeezed. The entire strategy has been get people building their stuff with these models baked in or a core part of their business ops, and then once you rely on it you can't cut it out. They raise prices/reduce what you get, but you keep paying because you need them now. So I'm not surprised, we knew this day was coming. And it's not over, there will be more of this. That said, now's still the time to build. Personally, I believe having a really great agent harness is the most important thing. you can own it, control it, and then you just plug in whatever models you want. So when a company like @AnthropicAI pulls a stunt like this, you can just swap to Codexand carry on. When OpenAI follows suit, you can swap them out for an OpenRouter api key and access other cheaper models. And your memory files, agent structure, skills, etc. all stay with you and you never miss a beat.

Pranit @Pranit

3 months ago

So the truth is finally coming out Yesterday, Anthropic said users were hitting limits because of their own usage patterns. Today, Boris is telling us they're cutting OpenClaw and third-party tools from subscriptions and prioritizing capacity for their own products That's a

67 35 686 63K 82

Great point, Timo. You've identified a real pain point that needs solving. The core issue? Most people want real connection, not AI replies. Here's the truth - AI-written replies are generic, overdone, and lacking in soul. Want me to brainstorm hoe to create a social network that actually works? (How'd I do?...actually not written by AI^ 🤣)

I'm sitting here watching my Paperclip agent refill it's own Fal.ai billing. It was trying to create images, was out of tokens, started reasoning and went to the billing page and began attempting to add more money to the account. I have Bitwarden set up and it has its own account, and I had put in there a single-use credit card (specifically locked to Fal.ai with a limited budget), and it had the cognitive ability to think about bitwarden, check and find the credit card, then go top up its own billing. Crazy. Also, scary, if you give it access to an unlimited credit card. I use Privacy.com (free tier) and just create a specific credit card for each service (each card can only be used with 1 vendor...whoever it gets used for first...on the free tier, but that's fine for me), and I set a budget on that card so there is some headroom if it needs to go top things up. But then it can only top itself up so much before the credit card maxes out and I'd still have to be the human-in-the-loop, forcing me to get eyes on any potential runaway processes. That said, it's working as intended right now, which is kind of cool. I wasn't sure how much fal.ai credit I'd need/want, so I just put an initial $10 when I signed up and put the credit card limit to $20 (for now), since you have to buy in $10 increments and I figured I'd equip my agents with +1 top-up before I'd have to get involved. Pretty cool to see this happening on my team in real time, real problem-solving and autonomy.

Forget #buildinpublic, now we can #buildintraffic

OpenAI @OpenAI

4 months ago

ChatGPT is now available in CarPlay. The voice mode you know, now available on-the-go. Rolling out to iPhone users running iOS 26.4+ where CarPlay is supported.

746 823 11K 6.9M 1K

@rezoundous Idk if I will. I have serious trust issues after they bungled this past week. Beyond the product being unreliable, they clearly can't be relied on as a business.

@altryne @AnthropicAI Yeah I mean I'm already on Codex and cancelling Sunday night (renewal would be Monday) after I squeeze as much usage as I can Fri-Sun after my limit resets Fri AM. Unbelievable.

@anitakirkovska Complain about Anthropic on X

Who are they going to rehire in a year when they realize that half the stuff being shipped right now requires maintenance, and building anything more than a single-feature product still requires the experience in building systems that developers have better than anyone?

The Random Recruiter @randomrecruiter

4 months ago

“Who should we lay off first?” Tech CEOs:

188 803 9K 864K 1K

This period of cheap tokens won't last forever. They are selling it at a loss and prices will go up in the future. Are you taking advantage of this unique period to build something that will last for the cheapest it will ever be? Or are you on the sidelines still "brainstorming"?

andrei saioc @asaio87

4 months ago

AI companies are leaking cash heavily. Any $200 Claude code plan consumes around $5,000 worth of resources per month.

73 47 640 24K 23

@kr0der @thsottiaux Gotta switch man. Codex has been working fine and I am not looking back.

I had just setup Paperclip this past weekend and am so glad I did given the Claude disaster this week. This has been a great example of how having your own "harness" is so important, and having one that's not tied to a specific provider. I never got deep into optimizing my Claude, instead I was optimizing my Openclaw and then my Paperclip. Which meant, when Claude hit my limit, I was able to simply go into Paperclip and toggle the agent from Claude to Codex (yes, I spent more money to get a Codex subscription, but I wasn't about to give Claude more money via API or overage fees...). From there, Codex immediately continued working with the same memory agent instructions, etc as I had before. Yes, Codex does have slightly different personality and behavior, but that's a small price to pay for continuity in my development process. I can manage that easily. If you're deep into Codex or Claude code, I highly recommend looking st Paperclip or another harness system to abstract memory and agent instructions and project data so you aren't trapped by one provider. This week I would have either had to pay up or get completely sidelined if I hadn't had paper clip setup, which meant a simple toggle and work resumed.

My subscription renews on Apr 6. As soon as my limit resets tomorrow, I'm going to use Claude for a such as I can before cancelling at the last minute since I've already switched over to Codex this week. They should have never given me a chance to try and get plugged into Codex, but here we are. Not going to be lured back quickly, either. New model drop isn't enough when j just don't feel I can rely on them to a) have their product work right b) make it right when their product doesn't

@ZssBecker Hard agree. Such a dumb take. There is more brilliance in simplicity than there is in sheer volume.

This is a wildly bad take. As if # of lines of code equates to quality. 1. Ben Franklin famously said "If I had more time, I would have written a shorter letter." There is more brilliance in getting from A->B in the simplest possible way. 2. Anyone who has built anything in software knows it's not a linear effort. It's compounding. As more things get added and as the code base grows, so does the network or dependencies, making each subsequent edit require much more groundwork to ensure it works and doesn't break other things.

beginbot 🃏 @beginbot

4 months ago

wow @garrytan just exposed Anthropic as total frauds Claude Code was ONLY 512K LOC ☹️ Gary is shipping 37K LOCs PER DAY so Gary could recreate all of Claude Code in ONLY 13 days! a supposedly $380 billion is big trouble

84 175 7K 461K 294

There is a great lesson for builders here, and that is: running a business involves more than just shipping code/features. Yes, Anthropic has been shipping at a breakneck pace and building some cool stuff recently. However, they seemingly became so enamored with that, that they then fell on their face with the recent caching bug/limit issues. Because running a business is about more than just shipping code. The trust that customers have in you as a business matters. Preventing existing customers from experiencing a competitor is strategically very important (especially if the competitor is competent). Trust takes far longer to build than to lose. And they completely fumbled the bag from a business judgement standpoint here. Bottom line for us builders is to remember that there's more to it than shipping code, you have to run a business and not lose sight of running a GOOD business. A good product is easier than ever to replicate. A good relationship with your customer and a well run business is much harder.

Angel 🌼 @Angaisb_

4 months ago

Anthropic: We've noticed people are hitting usage limits way faster than expected, we'll fix it (they know why, the community found the bug) OpenAI: We have no idea why people are hitting usage limits faster than usual, so here's your usual limits reset

40 36 718 42K 42

@DeryaTR_ @AnthropicAI Unbelievable bungle by Anthropic to just sit silently for as long as they have. Destroying our trust by the hour, horrible management decision.

@gginwanderland Lol yeah. I mean you seemed to be getting work done so I was confused! I expected you'd hit your limit before finishing your coffee...

@Jessie_R_Bishop @lydiahallie Yeah I have max and have been at my weekly limit since Sunday night. It's crazy. And now it's Wednesday and still no clarity from Anthropic on when things will be resolved. Costing me a whole week of progress!