Pinned Tweet
Right before our baby arrived, I handed the sales front line to an AI version of me. Here's what happened. x.com/i/article/2082…
411
239
503
1,432,122
the 'look at what you actually shipped' one hurts. roadmap says one thing, git log says what we really care about lol
the best life advice works for startups too:
> what do you or your team keep complaining about but never fix?
> forget the roadmap. look at what you actually shipped. that's what you really care about
> picture the company you'd hate seeing yourself building next year. write it down
> now pick one goal for the year, one for this month, one for tomorrow
what a company does every day is its real plan.
the rest is just corporate yap
day 9/100 of turning @HyperFrames_ into a real business, is more self-reflection.
almost 10% gone, are we getting closer? hard to say.
I’d love to talk to more HyperFrames users, plz DM 🙏
2
0
1
66
watch time ranks whatever keeps you there 3 more seconds. an hn comment saying 'you have to watch this' is someone vouching for it years later. totally different objective, way better feed tbh
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org
- Excellent (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
0
0
1
108
Wayne L. retweeted
the @WisprFlow vs @meetgranola war is so interesting to watch.
i’m a big fan of wispr flow. i’ve dictated over 250k words with it. wispr really nails the UX, not only on desktop but on mobile too, which is much harder. one of my favorite UX ideas in their iOS app: they use a to teach you how to set up the keyboard permission, right inside the settings app. it shows up exactly when you need it. really brilliant.
i’ve also curated hundreds of heygen terms in my wispr dictionary, including my team members’ names. that’s what lets me one-shot what i want to say into a ready-to-send message. i thought this would be the context layer that keeps me on wispr forever, plus their amazing UX.
now i’m not so sure.
i’m also a granola user. granola has one of the best consumer-grade experiences in AI. building software is so easy with agents now, and granola is one of the few teams that can build a great experience in a new category. so yeah, i’m very tempted to try their dictation.
then i thought one layer deeper. what context have i actually built up in wispr that makes switching hard? the dictionary is one thing, but it’s easy to export. maybe wispr has some personalized prompt that learns my dictation habits over time. i honestly don’t know. if it does, that’s hard to leave. i’d have to dictate another 250k words in granola to get it back.
switching cost in the AI era is a really interesting question. any company can build new things much faster now, so the boundaries between products are blurring.
i’m still a big believer in opinionated product experience. that’s what creates differentiation for customers, and that’s how you build a brand people love. the more things you do, the weaker your brand positioning gets. in the AI era, it feels like nobody appreciates this anymore. everyone wants to do everything, because doing everything is 10x easier than before.
i have one simple principle: how would your customers describe you in one sentence? if you ask 10 customers and they all say the same thing, you get powerful word of mouth.
most AI products are going the other way right now. you can argue that’s the old way of thinking about product. but i think a lot of AI products today are still built for AI-native users. when they reach mainstream customers, that’s when brand and product positioning really matter.
curious what you think.
i’m going to try granola for one week and feel the switching cost myself. will share how it goes. either way, big fan of both teams.
i’m a big fan of wispr flow. i’ve dictated over 250k words with it. wispr really nails the UX, not only on desktop but on mobile too, which is much harder. one of my favorite UX ideas in their iOS app: they use a to teach you how to set up the keyboard permission, right inside the settings app. it shows up exactly when you need it. really brilliant.
i’ve also curated hundreds of heygen terms in my wispr dictionary, including my team members’ names. that’s what lets me one-shot what i want to say into a ready-to-send message. i thought this would be the context layer that keeps me on wispr forever, plus their amazing UX.
now i’m not so sure.
i’m also a granola user. granola has one of the best consumer-grade experiences in AI. building software is so easy with agents now, and granola is one of the few teams that can build a great experience in a new category. so yeah, i’m very tempted to try their dictation.
then i thought one layer deeper. what context have i actually built up in wispr that makes switching hard? the dictionary is one thing, but it’s easy to export. maybe wispr has some personalized prompt that learns my dictation habits over time. i honestly don’t know. if it does, that’s hard to leave. i’d have to dictate another 250k words in granola to get it back.
switching cost in the AI era is a really interesting question. any company can build new things much faster now, so the boundaries between products are blurring.
i’m still a big believer in opinionated product experience. that’s what creates differentiation for customers, and that’s how you build a brand people love. the more things you do, the weaker your brand positioning gets. in the AI era, it feels like nobody appreciates this anymore. everyone wants to do everything, because doing everything is 10x easier than before.
i have one simple principle: how would your customers describe you in one sentence? if you ask 10 customers and they all say the same thing, you get powerful word of mouth.
most AI products are going the other way right now. you can argue that’s the old way of thinking about product. but i think a lot of AI products today are still built for AI-native users. when they reach mainstream customers, that’s when brand and product positioning really matter.
curious what you think.
i’m going to try granola for one week and feel the switching cost myself. will share how it goes. either way, big fan of both teams.
26
3
102
11,715
the 'whisper ends words early, cut 0.15s after' bit is so real. we run almost this exact loop on our clips - agent renders, another model watches it with sound, every note becomes a rule. @rowaifa wrote the most complete map of agentic video editing i've seen, worth the bookmark
0
0
5
259
tried @ctatedev's agent-desktop (vercel labs, open source computer use for coding agents) on the linux box my agents live on. first reply: no accessibility bus. fair lol, that box runs on screenshots
refs instead of pixels + each agent with its own cursor is exactly my problem. eds its own whole desktop just to click stuff
refs instead of pixels + each agent with its own cursor is exactly my problem. eds its own whole desktop just to click stuff
Introducing agent-desktop
agent-browser, for your whole computer
→ Computer Use for any coding agent
→ macOS, Windows and Linux
→ Runs in bg while you keep working
→ Agents use your computer in parallel, each with its own cursor
0
0
4
214
we see the same backlog in video every week. old onboarding, product tours of a UI that's gone, training in one language. nobody re-shot them bc that meant a studio + presenter + editor. now its a script edit
The market for software migration from legacy systems has been limited to $30B a year or so because migration is very expensive and maintenance is easier. Talking to older firms, there is a real backlog of scoped cheaper migration projects being unlocked by AI surprisingly fast
0
0
0
110
250k sub youtuber made his show intro without opening an editor once. claude wrote it, hyperframes rendered it. how many rounds did the type timing take you @EnriTarta, it's realy clean
New intro for my YouTube show Behind the Product.
Entirely made with code.
Made with Claude Opus 5.5 + HyperFrames + flask.do
1
0
6
538
one selfie in, full trailer out. video 1 did the scenes, hyperframes did all the glue around it. meanwhile i still cant get a decent linkedin headshot lol
One selfie went in. A whole blockbuster came out. 🍿
Scenes: @HeyGen Video 1
Everything else: @HyperFrames_
0
0
0
90
did this math on my own video agents. 13 clips, 36 full renders, and one post i had to delete after it went live. the review was never the expensive part @svpino
People are celebrating how much time they're saving by letting agents review their code.
But be honest: how much time are you spending fixing bugs afterward?
Saving two hours on code review means nothing if you then spend four hours fixing problems.
If you really want to know whether your automated code reviews are saving you time, you gotta measure the whole thing:
1. Track how much time you spend reviewing code, including the agent's feedback.
2. Track how much time you spend fixing bugs that make it past review.
3. Track how much time you spend investigating incidents and dealing with affected customers.
Compare the total time with what it took before introducing agentic reviews. Ideally, measure it per PR or per change so you're comparing similar workloads.
At the end of the day, nobody cares how much time you are saving on a particular task.
The ultimate goal is to save more time overall without deploying crap to production.
0
0
0
76
the loom made cloth cheap but the person picking the pattern got more valuable, not less. thats roughly how it feels running video agents all day, the taste part is the whole job now
I was interviewed in Vox about my new book, and a question on the meaning of work let me discuss something I have been thinking a lot about, comparing the logic of industrialization to the logic of craft: "You’re a hand-weaver in a world of power looms." vox.com/technology/508…
0
1
0
52
the "allow more than 10 edit comments" ask is how you know someone's actually shipping 20-30 videos a week on @HyperFrames_ studio lol. best kind of feedback
we might have solved ai video editing
for @nitish2096 at least
day 6/100 - been talking to many hyperframes users this week
Nitish hop on a call with me tdoay and share his feedback for @HyperFrames_ Studio - with his permission, this is what he shared on the call
we hope to rk, and thank you for all the kind words ♥️
in addition to all the positive comments, here are things we will be fixing for Nitish as soon as we can:
> fix archived project loading
> better captions
> allow more than 10 edit comments
> make homepage workflows better, easier to prompt
> more inspirations and longer clips
> storyboard commenting should support batches
> a bigger sound effects library
> better background removal
> give community posters credit rewards
now back to shipping 🫡
0
0
4
164
Wayne L. retweeted
this is how i make non-slap launch videos
give your agent computer use and instruct it to capture the feature you want to talk about then rebuild in html so everything can be edited
now available in hyperframes studio btw x.com/HyperFrames_/s…
give your agent computer use and instruct it to capture the feature you want to talk about then rebuild in html so everything can be edited
now available in hyperframes studio btw x.com/HyperFrames_/s…
Agents rebuilt Slack as video (using code) by looking at the real thing
0:00 Why
0:25 @cursor_ai agents & computer use
1:03 Video as code
1:56 Slack
2:54 A chat, an agent, a computer
3:53 Inner loop and outer loop
5:04 The ladder
7:47 Skills
8:14 Feedback loops
10:02 Recap
6
1
19
956
launch videos always looked fake because the agent was guessing what the app looks like.
now it opens the real app, clicks through it and rebuilds the screens in html. no more mockups that don't match the product
now it opens the real app, clicks through it and rebuilds the screens in html. no more mockups that don't match the product
the AI slop launch video killer is here
Computer use is live in HyperFrames Studio - your agent can click, type and record in any app on your computer
pro tip: have it drive your app, extract the screens and rebuild them in HTML so your product videos look real
try it now ↓
1
0
3
125
every voice agent is ears + a brain + a voice (and for us, a face). the hard part is the tape in between: turn-taking, interruptions, latency. that tape is pipecat, and our live avatars plug into it too.
so this one matters to me: @kwindla's daily is joining deepgram, and ependent foundation with a steering committee that includes deepgram's competitors. same license, same maintainers.
congrats to the daily team. keep the tape neutral
so this one matters to me: @kwindla's daily is joining deepgram, and ependent foundation with a steering committee that includes deepgram's competitors. same license, same maintainers.
congrats to the daily team. keep the tape neutral
3
1
10
530
mistral large 4 ("le chonk") just went open weights: 1t params, ~52b active, text + image in, 1m context, modified mit license.
reading the card as someone building agents: the per-request reasoning dial is the cost knob, and self-hosting means an 8-gpu node. so i'd start on @mistralai's api first
reading the card as someone building agents: the per-request reasoning dial is the cost knob, and self-hosting means an 8-gpu node. so i'd start on @mistralai's api first
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity.
💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others.
Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it.
Read more from our team and customers 👇:
1
0
3
216
fair for camera footage, a timeline is still faster by hand.
it flips when the video starts as code. then going back and forth is just "move the punchline 0.4s earlier" and the agent re-renders. that's how my clips get made now
it flips when the video starts as code. then going back and forth is just "move the punchline 0.4s earlier" and the agent re-renders. that's how my clips get made now
Anybody claiming AI is editing their videos end to end is lying. Frame-by-frame control is essential and I simply cannot figure out how to go back and forth with an agent on this that's anywhere near as fast as a traditional point and click flow in Final Cut.
1
0
3
146
podcast editing is mostly syncing cameras and hunting for the clip.
handing both to an agent is the right split
handing both to an agent is the right split
Edit Podcasts in HyperFrames Studio
your agent edits the whole episode with you, from the raw cameras to the clips
it syncs every camera and cuts the episode, then color grades it and adds overlays and captions when you ask
get started ↓
0
0
3
92