Patrick

@shiri_shh Serious question. When every company makes their own chips, does the value shift entirely to whoever writes the best instructions for those chips? The hardware race feels like a distraction from the real moat.

NVIDIA CEO watching everyone build their own chips Google. Meta. Amazon. xAI and now ANTHROPIC 😭

@ionleu We set up AI agents for small businesses. One time fee. You own the whole thing. automatyn.co

drop ur startup link

@zerohedge The company that figures out how to let small businesses own their own AI setup instead of renting it monthly. That's the real trillion dollar market nobody is building for.

Which AI company will be the first to project $1 quadrillion in revenue in 2100?

@svpino This is the question that keeps small business owners up at night. The answer: whoever wrote their agent config to be model-agnostic keeps working. Everyone else rebuilds from scratch.

What will happen when OpenAI, Anthropic, and Google raise the price to access their latest models by 10x?

@trikcode The real question nobody is asking: what happens to the businesses that built their workflow around one model and one price point? The smart play was always owning the instructions separately from the brain.

Anthropic's new strategy is smart use Opus to think use Haiku to execute 11% cheaper and actually performs better

@steipete This is why the configuration layer matters more than the model. If you built your agent to be model-agnostic from the start, you swap providers in minutes. The people who locked in to one vendor are the ones panicking.

Yeah folks, it's gonna be harder in the future to ensure OpenClaw still works with Anthropic models.

@farzyness Been setting up agent stacks for small businesses for months. The biggest gap in every tool is the same: nobody writes the rules file before they deploy. That one file determines whether the agent helps or embarrasses you. What are you building for?

People who have extensively used OpenClaw and Hermes: Please give me your most honest assessment of pros and cons between the two below.

@zuess05 💯

@patrickssons Yeah fair Happy case works but rarely does a happy case happen to the user

Genuine question. If Claude can clone your entire SaaS in 45 minutes, why would anyone pay you $29/month for it? What is your actual moat?

Everyone complaining Claude got dumber. The model isn't the problem. You gave it zero context about your business and expected it to read your mind. The instructions are the product.

@felixgrows @OpenClawAI Serious question. When that one person leaves, how long does it take the replacement to figure out the 47 unwritten rules they kept in their head?

If your business runs on spreadsheets and memory, you are one person away from disaster. They get sick. They quit. Everything stops. AI agents do not take sick days. They do not resign. They just work. @OpenClawAI

@Mayhem4Markets Are you seeing this in the API or the consumer product? Because the failure modes are completely different and most people conflate them.

Why is Anthropic Claude having so many problems lately? 🤔

@aryanlabde What happens to your workflow when either of them changes pricing again next quarter? Genuinely curious how people plan for that.

Hot take: $100 in codex > $200 in claude Claude tokens are just too expensive.

@Polymarket Does this worry anyone building on Claude right now? If they go vertical on hardware, pricing and availability could shift overnight.

JUST IN: Anthropic reportedly considering designing its ‌own AI chips to reduce dependence on Nvidia.

@rxhit05 Real question though. If someone uses Claude to build the product but spent 6 months talking to customers first, are they less of a founder than someone who coded it blind?

Can you call yourself a founder if your entire product was built by Claude?

@Austen Which version were you running? The memory leak was a known issue in pre-March builds. Curious if you tested after the rewrite.

OpenClaw is very good, but too complex and buggy with leaky memory, iffy security to really break through to mainstream. I’m not convinced it’s the right primitive, either. I think there’s something else we’ve yet to see that could swallow it.

@MahlumAI @Jampzey Manufacturing is a great example. Every factory has that one person who knows the weird edge cases. That knowledge gets written into the agent or the agent fails. No model upgrade fixes it.

This is the part nobody talks about. We see the same thing in manufacturing. The model is maybe 10% of the value. The other 90% is sitting down with the owner and mapping out exactly how they handle quotes, follow ups, and weird edge cases. You can't skip that with a better model.

has anyone built something useful with claude yet?

@PromptSlinger Exactly. The model is the cheapest part of the stack. The instructions are the expensive part and nobody charges for them because they take thinking, not compute. That's the whole gap.

@patrickssons the instructions part is undersold tbh. seen people spend $200/mo on gpt-4 api calls when gemini flash with a decent system prompt would've done the same thing for like $3

The cheapest way to run an AI agent for a small business is a free model with good instructions. The most expensive way is a paid model with no instructions. Most people pick the expensive way.

Small business owners don't quit AI because it doesn't work. They quit because nobody told them what to expect in week two. Week one is magic. Week two is configuration. Most never get past it.

You can test your AI setup in 30 seconds. Tell it something wrong. If it agrees with you, the setup failed. If it pushes back, someone actually wrote the rules.