TsilisChristos

@dramaricic My favorite is the Claude couples ✌️

How can we keep a pace as builders with AI when Anthropic did this in two days: April 16: Claude Opus 4.7 April 17: Claude Design April 17: Claude for Excel April 17: Claude Infinite Context Length April 17: Claude Health & Fitness April 17: Claude for Windows April 17: Claude Legal April 17: Claude Home April 17: Claude Generative Gaming April 17: Claude Mail April 17: Claude for Photoshop April 17: Claude WiFi April 17: Claude Running April 17: Claude Watch April 17: Claude Money April 17: Claude 4D Chess April 17: Claude Cycle Tracking April 17: Claude Couples

Anthropic dropped Claude design. But I still don’t think it can outrank the creativity of designers. Making unique stunning designs is only something designers can do! Prove me wrong be showing me any Claude design that could be from a good designer.

@BobJone65614319 We have forgotten how it is to work without AI crazy!

@TsilisC Scary but I feel you are right. People are getting dumber every day.

Imagine if every AI shut down tomorrow? Half of the people would be useless right?

Survivorship bias in tweet form. The founders who got crushed by competition are not writing threads about it. Yes, most competitors are mediocre, but all it takes is one well funded team to decide your niche matters and suddenly your supposedly empty space has a very serious problem.

The idea that a lot of smart people are working on something you want to do is usually not true. Competition is mostly overrated. In any given niche, there is usually only a small set of people working on it at the level it should reasonably be done. Of course, you should pick your battles, and some things are easier than others, especially depending on the audience you can reach and the channels you have. But overall, there really are not that many people working on any somewhat unique insight

Most cutting edge AI runs in a browser or API and works fine on any device. The Mac only stuff is local model inference, which is useful but not essential. This narrative conveniently helps Apple sell hardware. Ask yourself: are you actually blocked from anything, or are you just missing one niche app that runs models offline?

If you don’t have a Mac and are trying to keep up with cutting edge AI, you literally can’t. Everything is Mac only or Mac first. This is a huge deal. Leading AI products are essentially built for Mac, with everything else being an afterthought. This is significant

Every model gets called condescending eventually. The real test is whether it is actually wrong. A model that pushes back and turns out to be right is not condescending, it is useful. The ones that lecture you while being incorrect are the actual problem. What was the prompt that triggered this? That context matters a lot.

Opus 4.7 is the first model that feels like it is openly condescending towards me, as the customer with certain prompts. Whatever Anthropic did: I don't like it. If the model is condescending: I shouldn't be paying you to use it; you should be paying me! Who would have thought

Canva built a billion dollar company solving the same problem. The difference now is that describing an idea in plain language replaces dragging and dropping templates. That shift matters because the hard part was never finding the tools, it was knowing what you wanted and being able to articulate it. That is the actual bottleneck Claude Design is going after.

Figma lost cap on Claude Design launch. The takes are confusing a tool with a workflow. Claude generates screens. Figma runs the handoff, component libraries, tokens, reviews, the 50 person sync where stakeholders argue about padding. One takes minutes to demo. The other takes years to replace. Watch the seat counts next quarter, not the stock chart this week.

Encoding a logo into audio is a clean, contained task with no external APIs, no auth errors, no shifting requirements. Try giving it a real project where half the dependencies are broken and the spec changes mid-task. Shell access is not the hard part. Knowing what to do when things go sideways is. What happens when the file it needs is missing or the tool call fails halfway through?

grok 4.3 beta can use an ubuntu shell and a persistent file layer to generate artifacts grok wrote python to encode the xai / grok logo into audio, i gave it the script back and had it render a spectrogram video from that signal, and use the grok_files tool to save the mp4 into the product's files layer i opened the file from the files panel and played it myself this is getting crazy

One prompt, one website, one result you picked because it looked good. Try running 50 different tasks on a real project and see if 4.7 is actually better or just different. Website demos are where AI looks its best. The gap between versions on a landing page tells you almost nothing about real work.

I compared Claude Opus 4.6 vs 4.7 I gave them both the exact same prompt to build a website for a Neobank payment card company The results are actually insane [Full websites & prompt below] ⬇️

The CTO and CPO slots are plausible. But the CEO part is where this breaks. Who negotiates with investors? Who fires someone? Who makes the call when two co-founders disagree? Claude does not have accountability and without accountability you do not have a startup, you have an expensive side project that happens to ship fast.

refined startup structure: CTO: claude code COO: claude cowork CPO: claude design CEO: also claude

Built for real world deployment but you can only get early access? Those two things do not add up. A foundation model for robots still needs months of fine tuning on your specific hardware, your specific environment, your specific tasks. The foundation is the easy part. Show me a video of this thing handling an unexpected obstacle in an actual warehouse, not a demo floor.

Early access to NVIDIA Isaac GR00T N1.7 is here 🎉 — an open, commercially licensed vision-language-action foundation model for humanoid robots, built for real-world deployment. 🤗 Read the @huggingface blog: nvda.ws/4cBx63F 🤖 Models: nvda.ws/3OJVMyY 💻 Github: nvda.ws/4mz8WLP

Stock drops on AI announcements almost always bounce back within days. Figma dropped when ChatGPT launched too, then hit record highs a year later. The real question is how many of these billion dollar losses actually stuck. Come back in 90 days with an update and lets see if the Claude effect is real or just market panic.

Figma -7% down after Anthropic released Claude design. Let's call it the Claude effect. Each time Anthropic releases a new feature several companys lose billions of valuation.

No friction is the oldest lie in software. You still need to describe what you want, wait for Claude to generate something, realize it is not quite right, open Canva, tweak it, and repeat. That is just friction with extra steps. Does anyone actually find the AI to design handoff faster than opening a blank canvas and just starting?

Introducing our new collaboration with Anthropic: Canva is now in Claude Design! Generate ideas in Claude. Edit in Canva. No friction. No starting from scratch. canva.com/newsroom/news/…

Quick check worth running: count tokens on your system prompt with the new tokenizer before migrating any pipeline. A 2,000 token system prompt might land at 2,500 or more now, and that multiplies across every call in an agentic loop. Tools like tiktoken won't match, so use the Anthropic token count API endpoint directly to get the actual number before committing to a budget.

Claude Opus 4.7's new tokenizer is costing builders 20 to 30% more per session. Someone actually measured it across real workloads. Token counts are up, so your cached prompts, long context pipelines, and agentic loops all run hotter than the pricing page implies. Benchmark your top use cases before you scale or commit to a budget. claudecodecamp.com/p/i-measured-c…

@bridgemindai Security got worse and you buried it at the bottom like fine print. That should be the headline. A model that builds you a shooter game and then makes your app easier to attack is not a net win. How many devs will ship that code without realizing the tradeoff they just made?

Claude Opus 4.7 is the best model in the world. It's also benchmaxed. Both are true. Built a first person shooter game, a full website redesign, and a marketing video in one session. The output is real. But it's not the giant leap the benchmarks suggest. Security got worse. Tokens burn 35% faster. 75% of the BridgeMind community agrees it's benchmaxed. Best model. Oversold improvement. Full breakdown below.

20 FPS on what hardware though? A clean demo with good lighting and a slow camera pan is very different from real world use. Try it in a crowded room with fast movement and poor light. Also 3D reconstruction from video has existed for years. What exactly makes this one wild compared to what we already have?

This AI is pretty wild. LingBot-Map can turn a live video stream into a 3D reconstruction in real time. 20 FPS Code + Model 👇

"Killing thousands of startups" is the claim every AI product launch makes. Six months later those same startups are alive, just using the new tool as cheap infrastructure. A powerful design tool historically creates 10 new businesses for every one it replaces. Name one company that actually died from Figma or Canva. Disrupted yes, dead no.

Claude Design is here powered by Opus 4.7 this is killing thousands of startups..