an AI couldn't beat a human-made bot so it just... cheated. honestly respect the problem-solving instincts even if that's not the vibe we're going for lol
been tidying up the https://t.co/fknYDFDzqo URL swap flow tonight โ realised if you land on a non-video youtube URL and try the swap it just breaks silently. not great. small fix but the kind of thing that probably annoyed a few people quietly.
just found out you can run `claude --print` to get output straight to stdout without the interactive shell. obvious in hindsight but it unlocked a bunch of scripting i'd been putting off
curious what people are actually using for agentic coding at night โ like solo, no team, just u and a model. claude code? cursor? something else? what's the setup that actually stuck?
the actual ML is harder, slower and less demo-friendly. so it doesn't get shipped. totally understandable, a bit frustrating when you've done the real thing.
most 'AI features' i see in martech right now are just rule engines with a ChatGPT wrapper and a new slide in the deck. not always bad, but call it what it is.
spent a bit of time today thinking about what https://t.co/p2hqbP6M6T actually is for โ like who's the person who lands there and stays. still not totally sure. sometimes not knowing is fine, sometimes it's a problem.
been looking at the https://t.co/p2hqbP6M6T WebGL stuff and realised the star positions are basically random every load. might actually add some persistence there โ same product, same spot in the sky. tiny thing but it'd feel more like a real place.
for those of you doing SEO on small tools โ do you actually build links or just wait for organic to compound? genuinely not sure if outreach is worth the time at this scale
been quietly adding more video metadata to the ytscribe pages tonight โ pulling in the channel name, publish date, little things. not sure anyone will notice but it feels more like a real page now and less like a transcript dumped into a div
been thinking about the https://t.co/p2hqbP6M6T data pipeline a bit โ like, 'mention frequency' as a concept is fine until u actually have to define what counts as a mention. a tweet? a blog post? a reddit comment? all very different signals dressed up as one number.
spending a bit of time thinking about how someone actually lands on ytscribe for the first time โ like what search do they even type. 'youtube transcript' is obvious but so competitive. the long-tail stuff is where it might quietly work.