


Increasing your devtool adoption xulux.ai


"Hi Claude good morning how are you I'm great thx build me a YouTube-style site but restricted to videos that have been repeatedly recommended on HN over the years pls. Don't make any mistakes."

banger song for late night code sesh on.soundcloud.com/k7EyRCw4dg…

At this point i feel writing documentation with hand would be faster. Or i am too lazy in my process or using the wrong model. Need to optimize, even with Ai getting the right documentation is taking me a day or so. With the doc, some parts are not sloppy which i optimize in the next version.

We're hosting a hackathon with our friends. $279,500 in prizes. This sunday.

MCP Apps night last week featured some crazy pulls. first, we heard from the co-creator of MCP Apps, @idosal1. then we witnessed insane demos from the founding engineers of copilotkit (@tyler_cpk) and manufact (@khandrewone), plus @opencolin, @JimTheo_dev, and @weebhek! the cherry on top was the fireside chat hosted by @tobinsouth to wrap it all up, and the conversations with all of the passionate builders in the room. thanks so much @sofiiiiiasz for including us! (cover photo by @prat3ik)

between @idosal1’s big-brother-like appearance & @TobinSouth’s impromptu panel, MCP Apps Night was so much fun😂 big thanks to our speakers, community demoers, & sponsors! can’t wait for the next one💥💥 @tyler_cpk @khandrewone @opencolin @JimTheo_dev @weebhek

🚨 Vaaya now onboards your Instinct agent with just a simple command. Just say "signup on vaaya.ai" All instinct users get $10 free credit to use on Seedance 2.5 , ChatGPT-image 2, @ExaAILabs, Browserbase, Firecrawl, @modal, E2B and 100+ services live on Stripe @mpp and @x402scan Genrate images, scape web, spinup sandboxes, host your html pages right inside your Instinct

There are a lot of key findings we didnt cover here. Do reach out to us for more details.

Brightdata had the best performance for tasks where scrapers gave targeted answers. Their skills guide agents to use MCP but no guidance for a fallback, leading to lot of failures.

@firecrawl was the fastest scraper. With codex performed on par with Apify. But on claude the model did not try to read their web docs and use the cli. Leading to low scores on some setups.

@apify performed the best with a score around 92.1%. The big tie breaker was that the models inherently seem to know about apify’s cli and api tools.

An interesting insight, more token + time + tool call did not result in higher results.

Codex + 5.6-luna performed 30% better than claude code + Sonnet 5. Codex was more flexible in finding the clis and apis when there was no MCP or skill installed. Cluade when provided with no MCP or skill mostly gave up. Claude also took longer times leading to a lot of time outs.

