@charliermarsh Like asking if the sun is gonna rise at this point
English
Dillon Hong
595 posts
English
@tylerangert @left_pad Woops actual final edit here — forgot to upload the one w music!
Will try to clean up the skill and share it out soon :)
English
For the content creators — I created a “rough-cut” skill with Claude which if you input a date range, will:
(1) download all photos or videos from iCloud within a date range
(2) run a transcription & vision model
(3) draft a story based on dialogue / scenes
(4) cut clips and assemble them using ffmpeg
(5) runs transcription again as QA
(6) outputs a rough-cut.mp4
Then you can iterate, output at full resolution and then apply your final edits for pacing, music, visual assets etc. in your actual editor.
Rough cut of my Japan trip from last year was pretty amazing (hours of footage to 25m) and it could even recognize landmarks. Not publish ready, but the rough cut feels like 50% or maybe more of the work. Let’s you focus on the parts that give it personality
Guillermo Rauch@rauchg
Show me the thing you’ve built with AI you’re most proud of. Reply with a working product URL and what model / agent you primarily used.
English
turbopuffer crossed $100M run-rate in March. 19mo after $1M. Profitable & <$1M raised.
Cursor・Anthropic・Notion・Cognition・Harvey・Bridgewater・Ramp・Linear・Legora・Superhuman・Atlassian・Granola
We’d be nowhere without them. We work like hell to exceed their expectations.
English
Dillon Hong retweetledi
Welcome to the AI Search MCP race. The category is full of "coming soon" roadmaps for operating from chat.
We shipped that in February.
The @AirOpsHQ connector has been live in Claude for 3 months. Customers like Carta, Webflow, and Chime have been pulling citation data, running competitive analysis, and launching refresh and creation strategies from chat since then. Not a phase 2. Not a Q2 plan. Live, in production, in their daily ops.
In fact, our team was on stage at @AnthropicAI Code with Claude event in London today, walking through how we built on top of this foundation. Huge moment.
For those new to the race, catch up when you can. If you need a refresher, here you go: airops.com/blog/how-to-us…
English
@yerbamatt @cynthiamcgillis @theHankTaylor @mgonto @hyperagentapp What about running and chaining skills at scale? Lemme know if you want a run through 🙂
English
@dillonhong @cynthiamcgillis @theHankTaylor @mgonto @hyperagentapp We personally just moved into chat/llm skill workows. But I've seen some cool programmatic table flows from other people.
English
I need @theHankTaylor & @mgonto to cover @hyperagentapp on their next Code to Market pod.
Eschewing the Airtable brand all up feels so expensive and painful. But I guess they understood the sentiment towards Airtable is "who uses that anymore" and so it was easier to just do a new brand.
English
English
@cynthiamcgillis @theHankTaylor @mgonto @hyperagentapp I feel like a lot of tools like clay, profound, and airops are moving to the table based agent approach which I'm wondering if airtable sees as an existential threat. I'm here for the bold risk. Have you seen all the OOH spend they're putting in around it? They are ALL IN.
English
@Simeon_Cps Currently visiting London for the first time and definitely see the appeal. Ton of character. Transit is super straight forward too
English
@DimitrisPapail Couldn’t be implied that the model attends to the action output because the tool call is dependent on previous action?
Then masking allows you to optimize for general problem solving without explicitly exposing the world model. That frees up weights to learn other things no?
English
@Vtrivedy10 What does it look like? Was trying to find how i could do that in product but ended up just trying to export traces and create my own
English
LangSmith Engine is how we’re spinning the always-on, self-improvement loop for every agent
- Tracing is on for every single agent
- Purpose built infra with SmithDB to handle data at agent scale (more data than humans have ever produced will be produced by agents)
- Ambient agentic intelligence applied to every Trace to find errors, product insights, or anything you want to look for (you can customize Engine to your needs)
- PRs and Evals generated from this massive data with human gating/acceptance
the data our agents produce is a gold mine of information to make agents and systems better over time
the goal is that this flow shows users the first sparks of truly always on Continual Learning for their agents across their entire company
Dillon Hong@dillonhong
@Vtrivedy10 @hwchase17 How’re you doing this flow in langsmith?
English
@Vtrivedy10 @hwchase17 How’re you doing this flow in langsmith?
English
English
Today I spotted something new in the @googlechrome: they've added drawing on top of PDFs! Let's take a deep dive into the ink! ✍️
English
@itsjessyin @retrodotapp I think the 3 lines are what throw it off. Should’ve tried to engrain/texturize it with the disco ball
English
i think skeumorphic app icons can actually be really pretty and maybe i’m biased towards light mode here but i’ve always loved the @retrodotapp icon
jess yin@itsjessyin
of course spotify should celebrate their 20th birthday with a special icon but this is just ugly
English
@destraynor You find yourself using operator more or the browsing the analytics -> agent?
English
Our new product, Operator, is prototypical of how all B2B software will be built.
Say what you want and leave the rest to us.
These 3 demos show case Operator doing analysis and synthesis, with dynamic UI, and inline collaboration.
It's very, very, cool.
San Francisco, CA 🇺🇸 English
Dillon Hong retweetledi
My mental model for agents has increasingly been framed in terms of lessons learnt writing in Zig.
"What do we know at compile time?"
Building agents is weird because they could theoretically do *anything*, which reduces the assertions you can make at coding (compile) time.
English
We just released the official GPT app for @Context7AI
You can now access to up-to-date docs in your conversations with ChatGPT
English