Scott Swingle retweeted
Meet the new Stitch, your vibe design partner.
Here are 5 major upgrades to help you create, iterate and collaborate:
🎨 AI-Native Canvas
🧠 Smarter Design Agent
🎙️ Voice
⚡️ Instant Prototypes
📐 Design Systems and DESIGN.md
Rolling out now. Details and product walkthrough video in 🧵
that’s why I’m preparing my toddlers to be athletes
Claude doesn't know what time it is or when previous messages were, which makes it less useful for ongoing chats (i.e. it doesn't know if it's been 3 days or 3 minutes since that last message). It sometimes even just guesses a wrong time?
@alexalbert__ please add message times!
Scott Swingle retweeted
I have never been more certain that if AI development stopped today, we would still have massive & rolling disruption across society & the economy for the next ten years as people figured out how to harness what models can already do. And the end of AI progress seems unlikely.
Scott Swingle retweeted
I think you identified most of the mistakes. I think the blind spot was really undervaluing design, marketing, and sales; common pitfalls for engineering heavy teams.
You guys out-engineered the field massively, but the UX held back the adoption and sales process.
I think context windows will be enough
a big chunk of context will be snippets of mistakes made previously by agents in similar situations. models will vastly improve at generalizing from these
and training models with consistent/"native" memory harness/tools
big unresolved question right now: will memories / continual learning be in weights or context window (via harness / memory tools) or both
Scott Swingle retweeted
Sad we couldn't make the business work. Thought I'd share some of my favorite ideas of ours on agent design
🧵
We're shutting down Mentat.
Thank you to our users! Many of you tell us it's your favorite coding agent. We just didn't reach sustainable growth.
I'm proud of our exceptional team: @ja3k_, @granawkins, @plutobyte. Reach out if you want to talk!
We're shutting down Mentat.
Thank you to our users! Many of you tell us it's your favorite coding agent. We just didn't reach sustainable growth.
I'm proud of our exceptional team: @ja3k_, @granawkins, @plutobyte. Reach out if you want to talk!
a jump in growth rate from 0.2% to 2% sounds suspicious, but:
imagine the economy HAS grown 2% year for the last 2000 years. Then 2000 years ago global GDP would have been just $0.0008 total (not per capita!)
Haiku 4.5 does well on LoCoDiff: better than GPT-5, about the same as Gemini 2.5 Pro
Introducing Claude Haiku 4.5: our latest small model.
Five months ago, Claude Sonnet 4 was state-of-the-art. Today, Haiku 4.5 matches its coding performance at one-third the cost and more than twice the speed.
ALT Chart showing Claude Sonnet 4.5 leads software engineering performance on SWE-bench Verified with 77.2% accuracy, followed by Haiku 4.5 at 73.3%.
why LoCoDiff is the best long context benchmark:
xitter.cf/bio_bootloader/status/…
