desunit

Link: github.com/SawyerHood/dev… Skill installation: /plugin install dev-browser@sawyerhood/dev-browser

I stumbled upon dev-browser project that allows you to work with your browser much faster comparing to other solutions: Claude Code Native Chrome: 80 turns -> 12m 54s MCP: 51 turns -> 4m 31s Dev Browser: 29 turns -> 3m 53s why? Fewer turns → fewer tokens One script instead of step-by-step loops No repeated reasoning / tool chatter > Full browser control (click, fill, scrape) > Persistent sessions (login once!!) > Multi-step workflows in one run > Sandboxed (safe) execution Enjoy and thanks to @sawyerhood for fantastic work!

@iaaaaan @adamlyttleapps what did you use for the video? Love it

@adamlyttleapps nice for when you're at your desk. i wanted something for when i walk away entirely. live activity on my lock screen so i know the second any session needs me apps.apple.com/us/app/agentfy…

I kept getting distracting while vibe coding… so I made a notch for Claude Code It updates the status, pings you when you need to answer a question and notifies you when the task is done When it detects claude is working it also prevents my macbook from going to sleep I can walk away from my macbook. Or watching a youtube video. And I'll get an alert when it's done.

@jacobschulman @adamlyttleapps @FarouqAldori Why do you think the same idea can't come to multiple people?

@adamlyttleapps You did not invent this. You forked a repo made by another guy who had the idea 3 months ago (just like I did). Give credit where it's due. Give credit to @FarouqAldori x.com/FarouqAldori/s…

@levelsio I couldn’t too. Weird!

Okay and who of you couldn't reply to this? Please please only if you couldn't okay?

@maks6361 🫡

The very first thing I installed on this device wasn’t a marketing agent. It was something better. I started keeping a diary, recording every thought that comes into my head. I open a Telegram bot, record an audio message, send it, and it gets transcribed into text and saved into a private repo. At the end of the day, I can chat with Claude and ask it to act as either my psychologist or a coach. It helps me set goals, organize my thoughts, and even suggest ideas. I’ve been running it for a few days and I’m truly amazed at how cool it is. It also sends me my revenue report 3–4 times a week with important highlights and suggestions regarding my apps and other projects. Huge thanks to @desunit for this idea! 🙏

@jpctan I think data. companies like Anthropic and OpenAI collect it a lot. It’s basically a new oil.

@desunit Which bottleneck do you think becomes most defensible here, data, compute, or integration into existing operations?

Jeff Bezos is raising $100B to buy factories and upgrade them with AI. From: AI writes code To: AI redesigns how things are made The model that can simulate the real world - metal, airflow, machines - before anything is built. You pre-build (simulate) reality in software. This direction feels both interesting… and obvious. > take a “broken” business > understand how it works > automate it And most of these are "boring" industries: manufacturing logistics supply chains ... ignored for years. Now AI shows up and finally turns the oxygen back on. And again… Hormuz logic from my previous post: who controls the bottleneck controls everything

@itsolelehmann Or you can write a small MCP server for nano-banana-2 describe how you would like to have thunbnail (provide example of other thumbnail) and it will help you to speed up at least image generation;)

co-signing. claude cowork/code is still my daily driver for writing and deep work and i don't see that changing. but... i keep reaching for perplexity computer when my task touches multiple models. like yesterday i needed a youtube transcript pulled (gemini handles that better) then wanted opus 4.6 to break it down + help me cook something w it then needed nano banana 2 to cook a thumbnail for it that used to be 3 tabs and a bunch of copy pasting / context switching (which i loathe doing) perplexity just chains them together because it has all the best models in one place scratches a very specific itch that no single model can.

Source: wsj.com/tech/jeff-bezo…

Remember what Steve Jobs was saying about artists? good artists copy, great artists steal. well… we’re watching it live. Composer 2 (Cursor's model)→ Kimi → Anthropic everyone remixing everyone. beautiful. People have noticed under the hood that the endpoint literally says kimi-k2p5 Kimi team : wait… what?! (tweet appears and ... disappears) Looks like this isn’t even the first time. Looks like old Composer was sitting on top of GLM 4.6 (which is under MIT license). But Kimi 2.5 is a different story. free… until ~100M users or $20M/month. Meanwhile, Anthropic accusing Kimi of distilling millions of their queries too ... so the chain looks like: Anthropic → Kimi → Composer 2 AI supply chain!

@levelsio I keep saying that over and over - money buys you freedom, it’s a tool

💯 But it's more about having the perpetual income so you can make choices in life that you actually want Like where to live or what to do Instead of being forced to live in a place you don't like to be near an office for a job you don't like

Everyone thinks AI will fragment markets. More builders, more niche tools, more startups. You don’t have to go too far to question that… just take a look at oil prices. They don’t depend on how many small producers exist. They move based on who controls the Strait of Hormuz. Distribution wins. That’s the core idea: AI won’t spread value - it will concentrate it. We saw this with the internet. Millions of businesses appeared, but most of the value flowed to a few aggregators. AI just accelerates that pattern. Two things drive it: 1️⃣ more data → better product → eventually replaces the service itself Before AI: More data = small improvement With AI: More data = new capability… at some point, it stops being a tool and becomes the outcome. You’re not improving the workflow - you’re deleting the need for it. That’s why data becomes a moat. It defines what your product is even capable of doing. 2️⃣ AI reduces internal chaos → big companies scale without breaking Normally, big companies hit a wall: growth, complexity, paralysis 💥 AI changes that. Fewer people needed, less communication, less coordination, more productivity. Which means companies can grow much bigger before breaking. All of this leads to fewer, bigger players. More builders, but fewer winners. What stays scarce? ‼️Distribution‼️ …and scarcity is where the money goes.

Source: x.com/ai/status/2032…

What if this might be the last job you have? Found a article where a VC describes a dinner in SF. The "serious" startup ideas were: - sex robots - sperm racing - orbital cremations - prediction markets on death Why these? Because AI can’t replace them yet. He shared a story - a startup he liked died in 48 hours. It wasn't from competition but a model update killed the entire market. TAM = 0️⃣ The real question if jobs disappear… who are we? Work isn’t just money. It’s identity. Take that away and you get a civilization-scale identity crisis. Keynes predicted we’d work 15 hours a week thanks to technology. He was right about productivity but looks like he was wrong about humans. We filled the extra time with… more work. Now AI might force the question again. When you realize you can choose many possible futures, your mind suddenly sees an infinite set of possibilities. Kierkegaard called anxiety "the dizziness of freedom". Feels about right.

@SearchForRyan I went further - asked to keep project’s documentation in docs, refer to it in claude.md. Each time when you made some important changes ask to update docs

okay why did nobody tell me about state.md for claude code? i've been stuck wondering why a project i build on daily was taking 8-10 mins to make changes turns out claude was reading EVERYTHING (all my code) before it made a change. i added state.md, architecture.md (alongside my claude.md) and bam, 8,000 tokens to ~1,000 tokens and from 8 mins to ~ 60 seconds. claude.md wasn't enough in this instance annnnd i'm so happy lol

@Shpigford @SearchForRyan @ElliottGarber I use MCP for that and a /command to use it to generate images and store into a specific folder. Works amazingly well

@SearchForRyan @ElliottGarber 🤔 so...claude doesn't have image generation capabilities. i'm curious how it's pulling off the visual stuff.

holy shit, claude code with an ad creative skill is insane, why did nobody tell me this? instead of bulk creating in canva, train claude on your entire ad library (or... someone else's) create these files for your skill: 1. design references 2. ad components 3. brand colours + fonts created 20+ ads in maybe 10 minutes that worked brilliantly on both mobile + desktop.

@tdinh_me You are absolutely right!

Half of the replies in my tweets are now AI replies 💀

Source: eon.systems/updates/embodi…

Fruit flies are the GOAT of biology! And they just proved it again. Fruit flies have been a key model organism in genetics for over 100 years. 🪰 became famous in labs because: 1️⃣ it reproduces extremely fast 2️⃣ the genome is relatively simple 3️⃣ mutations are easy to observe A surprising number of fundamental discoveries about inheritance came from experiments with 🪰 The researchers built a virtual fruit fly controlled by a simulated brain. The brain model is based on the real fly connectome - about 140k neurons and ~50 million synapses. They connected this brain to a physics-based digital body, creating a loop: > world > sensory neurons > brain activity > motor signals > body movement And the digital fly actually behaves like…...... fly It can: > walk toward food using taste cues > groom itself when "dust" accumulates > start feeding when sugar receptors activate The neurons are simplified. Only a few behaviors exist. Real flies are far more complex. But the concept is SOOO powerful - take a real brain and let it control a body in a simulated world.