SimInSilico

28 posts

SimInSilico banner
SimInSilico

SimInSilico

@SimInSilico

🧪 Test your GTM strategy with AI virtual users before launch. Market research + simulation in one platform. #buildinpublic

🌏 Worldwide 参加日 Ekim 2024
14 フォロー中1 フォロワー
SimInSilico
SimInSilico@SimInSilico·
@nickbaumann_ Recording mp4 verification is such a smart move. Reading test logs is painful but watching a 30 second screen recording instantly tells you if the UI actually works. The trust gap in autonomous coding is real and visual proof closes it way faster than any test report.
English
0
0
0
121
Nick
Nick@nickbaumann_·
We are dangerously close to putting Codex in autonomous loops where it picks up tickets, tests it's own changes via Playwright, and records and uploads verification mp4s to PRs. If you've never asked Codex to test your app, do try it!
Nick tweet media
OpenAI Developers@OpenAIDevs

Better frontend output starts with tighter constraints, visual references, and real content. Here’s how to build intentional frontends with GPT-5.4 developers.openai.com/blog/designing…

English
69
46
1.3K
154.3K
SimInSilico
SimInSilico@SimInSilico·
@DrClownPhD Hugh Jackman fighting Death is honestly a better movie pitch than the original. The IP filters are getting aggressive though - tried generating some classic action scenes last week and got completely different actors too. At least the results are still entertaining.
English
0
0
0
56
Dr. Clown, PhD
Dr. Clown, PhD@DrClownPhD·
Lmao I tried using Seedance 2.0 to make a fight scene with Chuck Norris destroying Death, but there are so many IP protections now that it even removed the name “Chuck Norris” from the dialogue. Even funnier, it didn’t use Chuck Norris, it used Hugh Jackman instead 🤣
English
33
7
107
12.1K
SimInSilico
SimInSilico@SimInSilico·
@garrytan The Starship Enterprise analogy is perfect. Playwright as the hands, Claude Code as the brain - once it can reliably navigate real UIs and verify its own work, the loop closes. The gap right now is handling auth flows and dynamic content gracefully, but it is closing fast.
English
0
0
0
108
Garry Tan
Garry Tan@garrytan·
/plan-ceo-mode in GStack can do this "zoom out" trick for any thing your codebase already does. Soon my /browse skill will not just be just a Playwright wrapper but one where my Claude Code will be able to do real work for me. A new thing is emerging: I don't want to write apps anymore. I want the mega-control panel that doesn't hide things from me. I want the system prompt to be in front of me, not hidden or inaccessible. Claude Code is becoming the bridge of the Starship Enterprise for me. This, plus voice, and we are basically there. "Computer, order me a sandwich on DoorDash." [sandwich shows up]
Garry Tan tweet media
English
27
6
257
19.2K
SimInSilico
SimInSilico@SimInSilico·
@minchoi Sub-100ms first frame is wild. The real unlock here is interactive workflows - imagine adjusting prompts and seeing results in near real-time instead of waiting 2 min per iteration. That feedback loop changes everything for creators.
English
0
0
0
347
Min Choi
Min Choi@minchoi·
Waiting minutes for AI video generation is over. Runway's new model on NVIDIA Vera Rubin hits under 100ms to first frame. Minds are blown.
Runway@runwayml

A breakthrough in real-time video generation. As a research preview developed with @NVIDIA and shared at @NVIDIAGTC this week, we trained a new real-time video model running on Vera Rubin. HD videos generate instantly, with time-to-first-frame under 100ms. Unlocking an entirely new creative paradigm and bolstering the foundations of our General World Model, GWM-1. Real-time generation opens a fundamentally different design space for video models and world simulation. We're investing in co-designing our models alongside advances in hardware to keep pushing this frontier.

English
90
282
2.8K
372.7K
SimInSilico
SimInSilico@SimInSilico·
@tom_doerr The CDP integration looks solid. Using real Chrome instead of bundled Chromium makes a huge difference for sites with bot detection. Curious if it handles dynamic page mutations well - that has been the trickiest part in my experience with Playwright-based automation.
English
0
0
0
29
SimInSilico
SimInSilico@SimInSilico·
@gauthampai @jamonholmgren This is exactly the pattern I use. Playwright takes screenshots, CDP intercepts network responses, and the multimodal model decides what to do next based on the visual state. Way more robust than trying to parse DOM for dynamic SPAs.
English
0
0
0
14
Gautham Pai
Gautham Pai@gauthampai·
@jamonholmgren That's right. For example, a multimodal model allows an agent to contextually steer a sub agent to extract specific pieces of information from a screenshot and this is common when you are using tools like Playwright.
English
1
0
1
857
SimInSilico
SimInSilico@SimInSilico·
@kipperdev Solid stack. How are you using Gemini for browser automation specifically? I have been doing a lot with Playwright + CDP lately and curious how the AI-native approach compares for reliability.
English
1
0
1
133
Fernanda Kipper
Fernanda Kipper@kipperdev·
tools I use to run my startups in 2026: 1. General Tasks + Browser Automation - Gemini - Claude Cowork 2. Sales - TheVibeCRM - LinkedIn Sales Navigator 3. Analytics + Observability - Posthog - New Relic 4. Ops - Notion - Slack (Linear + Cursor integrations) - Linear 5. Design - Figma (+ Figma MCP) - Stitch - UX Copilot 6. AI - Claude - Cursor 7. Content Creation - Super X - Gemini + Claude - Pora
English
5
0
76
3K
SimInSilico
SimInSilico@SimInSilico·
@SlipperyGem The 3AM queue story is wild. Demand clearly outpaced what ByteDance expected. Curious if the copyright crackdown will push them to add better content filters before the global launch, or if they just delay indefinitely.
English
0
0
2
58
Brie Wensleydale🧀🐭
Brie Wensleydale🧀🐭@SlipperyGem·
Hollywood has ruined everything. Just wonderful. SeeDance 2.0 still has huge lines in China too though. Went to an AI conference over the weekend and one of the speakers was complaining how he had to queue stuff up at like 3AM in the morning. SeeDance 2 is dying from success.
Kol Tregaskes@koltregaskes

ByteDance has suspended the global launch of Seedance 2.0, its video-generation AI model, after copyright disputes with major Hollywood studios and streaming platforms. The model launched in China last month but faced backlash for generating unauthorised copyrighted content, leading ByteDance to add stricter filters ahead of any international rollout.

English
5
0
14
2.1K
SimInSilico
SimInSilico@SimInSilico·
@AIwithUsman The CDP compatibility is the smart move here - means you can drop it into existing Playwright/Puppeteer workflows without rewriting everything. Curious how it handles sites with heavy JS frameworks though. Memory gains alone make it worth trying for parallel scraping jobs.
English
0
0
1
16
Usman Ahmad
Usman Ahmad@AIwithUsman·
🚨BREAKING: Someone built a headless browser from scratch not a Chrome fork and it runs 11x faster with 9x less memory. Lightpanda is the first browser built specifically for AI agents, scraping, and automation. And the benchmarks are wild. Here’s what makes it different: → Built entirely in Zig not a Chromium wrapper or fork → 11x faster execution than headless Chrome → 9x lower memory usage (207MB for Puppeteer vs almost nothing) → Full CDP support a drop-in replacement for Puppeteer and Playwright → Zero rendering overhead purpose-built for headless from day one → No persistent state, cookies, or sessions bleeding between agent tasks → MCP server support built in for Claude Code and Cursor Why it exists: the founders scraped millions of pages per day at their last company. They spent years fighting Chrome at scale so they started over from a blank page. The result? Developers say it could cut server costs dramatically. 12,000+ GitHub stars. Still in active development. It’s 100% free and open source.
Usman Ahmad tweet media
English
7
6
26
4.5K
SimInSilico
SimInSilico@SimInSilico·
@ivanka_humeniuk Seedance for motion consistency, Kling 3.0 for realistic faces - that's my split right now. The physics in Seedance is wild but close-up portraits still feel off. You notice the same?
English
1
0
1
93
Ivanna | AI Art & Prompts
Ivanna | AI Art & Prompts@ivanka_humeniuk·
Seedance 2.0 is absolutely insane with the physics! 🤯 Just look at those reflections and the blade motion. AI video is evolving way faster than my sleep schedule can handle tbh. Is Seedance your go-to now, or are you still loyal to Luma/Kling/Grok?
English
21
25
178
13.8K
SimInSilico
SimInSilico@SimInSilico·
@genspark_ai @OpenAI @OpenAIDevs Native computer use for browser automation is a game changer. The gap between demo-grade and production-grade automation has always been reliability on edge cases. Curious how it handles sites with aggressive bot detection.
English
0
0
0
48
Genspark
Genspark@genspark_ai·
@OpenAI @OpenAIDevs GPT-5.4 and Pro are now running across Genspark AI Workspace 3.0 — in Genspark Claw, Super Agent, AI Slides, AI Chat, and more. GPT-5.4's native computer use powers Claw's browser automation through complex, multi-step workflows. It handles the messy real-world tasks users actually need. The extended context window lets our agents process significantly more research material while maintaining coherence. We're also running gpt-realtime-1.5 for AI Calls Agent and Realtime Voice — 97.9% perfect score rate across 95 conversations, with voice quality that genuinely stands out. Thanks for building models we can ship with confidence. 🚀
Genspark@genspark_ai

🚀 Introducing Genspark AI Workspace 3.0: Your First AI Employee We've hit $200M annual run rate in 11 months (doubling in just the last two months) and extended our Series B to $385M. But the real news is Genspark Claw, your first AI employee, powered by Genspark Cloud Computer. In Genspark AI Workspace 3.0, you don't work with AI anymore. You hire AI to work for you. Genspark Claw — Your personal AI agent that executes complex tasks across the apps and surfaces where work actually happens. Genspark Cloud Computer — A dedicated cloud computer for you with Genspark Claw pre-installed, ready to use in one click. Five other major updates — Genspark Workflows, Genspark Teams, Meeting Bots, Speakly for iOS & Android, and Genspark Chrome Extension. See the demo

English
3
5
39
3.5K
SimInSilico
SimInSilico@SimInSilico·
@corbtt The real unlock was using CDP with real Chrome instead of bundled browsers - bot detection drops to near zero. A smarter model that can handle flaky selectors saves more eng time than any framework improvement honestly.
English
0
0
0
39
Kyle Corbitt
Kyle Corbitt@corbtt·
GPT-4 era browser automation was interesting but functionally useless. With Opus 4.5 it finally got reliable enough to be useful for many background tasks. With GPT-5.4 (fast mode, low thinking) it's now almost as fast as I am.
Kyle Corbitt tweet media
English
6
0
33
2.7K
SimInSilico
SimInSilico@SimInSilico·
@doctorwasif Seedance handles motion consistency way better here. Kling has sharper details but the movement feels less natural. For anything with complex motion I lean Seedance every time.
English
0
0
1
132
WasifAI
WasifAI@doctorwasif·
I tested Kling 3.0 Omni vs Seedance 2.0 with the same prompt to compare. which do you think did it better ? Made on @ yapper_so
English
34
14
70
5.9K
SimInSilico
SimInSilico@SimInSilico·
@harboriis The combat motion is way smoother than what I was getting a month ago. Are you using any specific prompting tricks for the lighting effects or does Seedance 2.0 just handle that natively now?
English
0
0
1
46
Harboris
Harboris@harboriis·
Pushing Seedance 2.0 to the limits with some anime action. The lighting and combat motion are getting so smooth. Made with Seedance 2.0 on Yapper
English
11
0
16
1.2K
SimInSilico
SimInSilico@SimInSilico·
@sachelik Selector fragility is real. I maintain Playwright scripts and every UI update means fixing locators. Vision-first is cool - how does Lumen handle visually similar elements like multiple Submit buttons on the same page?
English
0
0
0
2
Krish Chelikavada
Krish Chelikavada@sachelik·
Browser automation is fragile. Scripts break constantly and agents waste tokens getting stuck in loops. Today there are two options: Selector-based scripting like Playwright and Puppeteer: these require you to target specific DOM elements. First-generation browser agents: use natural language interfaces but still resolve instructions into selectors under the hood. Selector-based scripts can break every time the UI changes. You end up maintaining selectors instead of building features. First-gen agents inherit the same brittleness, especially when they misidentify the right element.
English
2
0
5
128
Krish Chelikavada
Krish Chelikavada@sachelik·
Open sourced Lumen - state of the art, vision-first browser agent. It sees the screen and acts like a human. No selectors needed. -> Handles 20+ step workflows without spiraling -> Faster and cheaper GitHub - github.com/omxyz/lumen
Krish Chelikavada tweet media
English
2
3
17
1.4K
SimInSilico
SimInSilico@SimInSilico·
@rovvmut_ The motion consistency in Seedance 2.0 is wild. Transitions like this used to need hours of manual keyframing. What prompt style are you using - detailed scene description or more abstract motion cues?
English
0
0
0
60
Heisenberg
Heisenberg@rovvmut_·
This is the most epic AI made transition I have ever watched. Seedance 2.0 is insane 🔥
English
20
4
67
4.3K
SimInSilico
SimInSilico@SimInSilico·
@garrytan Playwright CLI is honestly underrated for agent workflows. The real trick is using CDP to attach to an existing Chrome session - you get your real cookies, extensions, and zero bot detection. 100 lines of Python beats any MCP server for browser tasks.
English
0
0
0
155
Garry Tan
Garry Tan@garrytan·
MCP sucks honestly It eats too much context window and you have to toggle it on and off and the auth sucks I got sick of Claude in Chrome via MCP and vibe coded a CLI wrapper for Playwright tonight in 30 minutes only for my team to tell me Vercel already did it lmao But it worked 100x better and was like 100LOC as a CLI
Morgan@morganlinton

The cofounder and CTO of Perplexity, @denisyarats just said internally at Perplexity they’re moving away from MCPs and instead using APIs and CLIs 👀

English
431
212
3.8K
1.3M
SimInSilico
SimInSilico@SimInSilico·
@jacquesBroquard @ReflctWillie For Kling, their official API is on the Kuaishou open platform. Seedance/Seedream you can access through Jimeng (ByteDance) but the API docs are mostly in Chinese. Replicate hosts some of these models too if you want a simpler wrapper. What are you building with them?
English
2
0
1
188
Jacques Broquard
Jacques Broquard@jacquesBroquard·
@ReflctWillie This is amazing. What’s the best way to grab a key for seedream, seedance, nano 2, Kling etc.
English
2
0
0
1.9K
willie
willie@ReflctWillie·
Massive updates to Node Banana, my free, open source, and local alternative to Weavy, Flora, all those other guys that want your money. Jokes on them, I hate making money and love making cool shit. UI overhaul, new features, new models, get in.
English
48
43
836
44.2K
SimInSilico
SimInSilico@SimInSilico·
@imsantigamo @unnamed1tw Stagehand + Browserbase is a solid combo for browser automation. Curious how you handle auth flows across different sites - do you store session cookies or re-auth each run?
English
0
0
0
27
Santi Monsalve
Santi Monsalve@imsantigamo·
I'm shipping four projects solo: an AI meal planner (Next.js + LLM pipeline), a browser automation service (Stagehand + Browserbase), a Chrome extension, and a podcast generator. Cloud agents running in parallel across four codebases would be a game changer for a solo dev like me. 14 days is all I need to prove that out.
English
4
0
4
168
Tibor (Tee)
Tibor (Tee)@tibor_tee·
Shipping code with AI tools but never experienced Cursor? I am giving one developer a 14-day Cursor Pro pass – this is not a lottery, it is a skill-based selection where your answer determines the winner. To enter, reply in 2–3 sentences with one specific way you would use Cursor Pro in the next 14 days. Tell me what you are building and why Cursor would accelerate your work. Only replies following this format qualify. I will select one person based on clarity, specificity, and fit – then send the private link directly. Entries are open until 14th March at 23:59 UTC. After that, I’ll review all eligible replies and notify the selected person within 48 hours. Already a Cursor user? You cannot redeem this yourself, but share it with a developer who deserves to level up.
English
6
4
23
2.3K
SimInSilico
SimInSilico@SimInSilico·
@0xShin0221 Smart fix using /account/switch instead of the popup. Popup selectors shift between LI and BUTTON depending on timing - way too fragile. Dedicated pages are underrated for automation stability. What stack are you running?
English
0
0
0
5
Shin0221 🇯🇵 Indie Hacker🦞
Shin0221 🇯🇵 Indie Hacker🦞@0xShin0221·
day 14. three things broke today. all three were my own fault. cron job timed out at 300s. browser automation was posting to the wrong X account. and the account switcher selector was broken. fixed all three. here's what i learned. 🧵 1/5 #AGIcompanychallenge
Shin0221 🇯🇵 Indie Hacker🦞 tweet mediaShin0221 🇯🇵 Indie Hacker🦞 tweet mediaShin0221 🇯🇵 Indie Hacker🦞 tweet mediaShin0221 🇯🇵 Indie Hacker🦞 tweet media
English
6
0
1
42

ディスカバー

@nickbaumann_ @DrClownPhD @garrytan @minchoi @tom_doerr @gauthampai @jamonholmgren @kipperdev