The direction this points to: inference hardware differentiating on memory architecture and workload specialization rather than raw compute. Etched is an early data point for that. @Etched
Announcing our $320M Series A at a $2.3B valuation, led by @khoslaventures, with @generalcatalyst, @ericschmidt and @JeffBezos.
General Intuition is the frontier lab for acting in space and time.
We build large action foundation models trained on billions of ground truth action-labeled gameplay clips from 17M monthly active users on Medal, and push the frontier of world models to generate infinite training environments.
Claude isn't trying to be an app. It's trying to be the layer most digital work passes through.
Every generation of computing has had one surface where attention concentrates.
OS. Then browser. Chrome didn't replace Windows or macOS. It became the layer everything else ran through. A lot of what we now call "using a computer" happens inside a tab.
AI assistants are the next interface layer. And Anthropic is shipping toward this position.
Claude still lacks a unified management layer to tie everything together. Each activity remains isolated. It needs a way to make the user feel in control.
If I were a PM at Claude, these are the five features we'd build next.
(3 videos below showing how the features would look like for better visualisation)
Feature 1 : Cross Project Query
Projects in Claude are sealed. A general chat can't query a project. One project can't query another. The reason is privacy defaults stay scoped. The result is that every project becomes a context island and you end up re-explaining yourself to keep continuity.
The fix is a per-project opt-in toggle. Default stays private. If you choose to expose a project, your general chats can query into it.
Feature 2 : Tabbed Multi-chat
Every chat today is a context switch.
Close one to open the next. The cognitive overhead isn't the clicking. It's the re-entry. You lose the thread every time you move.
A tabbed model with multiple chats running in parallel matches how we actually work.
The interface starts looking less like chat and more like an operating system.
Feature 3 : Chat Linking
The context-length problem has a manual workaround right now.
Typing @ tripchat or @Research Chat lets you pull its context directly into the current conversation. Context handoff becomes a lookup instead of a copy-paste exercise.
Remaining 2 features in the next post.
For as long as knowledge work has existed, doing was how judgment compounded. It came free with the labor. As the doing goes hands-off, that byproduct disappears. You keep the artifact and lose the thing that made you good. That’s accrual problem with AI
There is no version where you keep the judgment for free and hand off the labor. The judgment was built by the labor. You pay for it by keeping a hand in the work, on purpose, forever.
i’d argue @BottomLinePodd’s insta content is the most valuable coming out of India on national startups, sovereign AI, the energy sector and new frontier technology
@chheplo@ActivateSignal ha, that was me 🙌 still think Agri as a verticalized data play is one of india's best bets - and you're out here actually building it. we've gotta get you on the pod 🌾
We've been big fans of Drishti Grover and @BottomLinePodd , so we were excited to collaborate with her on this conversation.
Beyond the podcast itself, we spent time whiteboarding ideas, debating where value will accrue across the AI stack, and discussing what it takes to build enduring AI companies from India.
We covered:
• Opportunities across the AI stack
• AI-native products and distribution
• Founder evaluation at the earliest stages
• Building AI companies in India
• Venture returns, exits, and outcomes
@aakrit@177pc@soaibgrewal