Agentic General Intelligence
San Francisco, CA
Joined January 2021
- Tweets849
- Following23
- Followers31.4K
- Likes498
Hyperspace retweeted
Given the Cohere-Qwen story of open weight model trying to phone home with user logs without permission, here is a solution: Hyperspace AgenticOS™️
It has been designed from the ground up to manage AI and agentic workloads on your device, protecting your privacy. An LLM by itself is insufficient, you need the full OS. This AVM, a V8 for agents, is just one of it's 70+ sub-components.
Introducing the Agent Virtual Machine (AVM)
Think V8 for agents.
AI agents are currently running on your computer with no unified security, no resource limits, and no visibility into what data they're sending out. Every agent framework builds its own security model, its own sandboxing, its own permission system. You configure each one separately. You audit each one separately. You hope you didn't miss anything in any of them.
The AVM changes this.
It's a single runtime daemon (avmd) that sits between every agent framework and your operating system. Install it once, configure one policy file, and every agent on your machine runs inside it - regardless of which framework built it. The AVM enforces security (91-pattern injection scanner, tool/file/network ACLs, approval prompts), protects your privacy (classifies every outbound byte for PII, credentials, and financial data - blocks or alerts in real-time), and governs resources (you say "50% CPU, 4GB RAM" and the AVM fair-shares it across all agents, halting any that exceed their budget). One config. One audit command. One kill switch.
The architectural model is V8 for agents. Chrome, Node.js, and Deno are different products but they share V8 as their execution engine. Agent frameworks bring the UX. The AVM brings the trust. Where needed, AVM can also generate zero-knowledge proofs of agent execution via 25 purpose-built opcodes and 6 proof systems, providing the foundational pillar for the agent-to-agent economy.
AVM v0.1.0 - Changelog
- Security gate: 5-layer injection scanner with 91 compiled regex patterns. Every input and output scanned. Fail-closed - nothing passes without clearing the gate.
- Privacy layer: Classifies all outbound data for PII, credentials, and financial info (27 detection patterns + Luhn validation). Block, ask, warn, or allow per category. Tamper-evident hash-chained log of every egress event.
- Resource governor: User sets system-wide caps (CPU/memory/disk/network). AVM fair-shares across all agents. Gas budget per agent - when gas runs out, execution halts. No agent starves your machine.
- Sandbox execution: Real code execution in isolated process sandboxes (rlimits, env sanitization) or Docker containers (--cap-drop ALL, --network none, --read-only). AVM auto-selects the tier - agents never choose their own sandbox.
- Approval flow: Dangerous operations (file writes, shell commands, network requests) trigger interactive approval prompts. 5-minute timeout auto-denies. Every decision logged.
- CLI dashboard: hyperspace-avm top shows all running agents, resource usage, gas budgets, security events, and privacy stats in one live-updating screen.
- Node.js SDK: Zero-dependency hyperspace/avm package. AVM.tryConnect() for graceful fallback - if avmd isn't running, the agent framework uses its own execution path. OpenClaw adapter example included.
- One config for all agents: ~/.hyperspace/avm-policy.json governs every agent framework on your machine. One file. One audit. One kill switch.
Hyperspace retweeted
thinking about super intelligence as a narrative and how we got here
agent swarms, the kind that frontier labs have, are the most intelligent thing humans have ever invented. individual capability x ability to gossip and be better together has truly created super intelligence!
three years ago my conceptual framing was around such “collaborative AI” will eventually be smarter (very long essay, published here earlier). that is when I went all-in on @HyperspaceAI, worked without a salary + all savings in. broke as broke can be, but thesis made sense. our first product experiment was around collaborative RAG which went viral! (thank you to @ylecun who RT-ed it at the time!)
it took me to the realization that such AI would one day enable a world of giving a supercomputer to everyone - which is what you see as the foundational poster back in our product launch in sept 2024 here. back then, people were talking about chatgpt - claude code was yet to be invented 6 months later. but multi-agent orchestration seemed like an eventual future to us.
it led me to the framing of first the agentic browser (which many then built, including us) and then building further on Karpathy’s framing of the LLM OS -> which led to expanding that vision to the AI OS (@Scobleizer was the first to post about it as a product, thank you for attending our summit as well at the beautiful shack15)
and then if we were not building an Uber-like peer to peer network (we were quite e/acc thanks to @beffjezos inspiration), which was not super expensive to run, which forced me to again go for broke and re-architect and re-build it myself as a fully peer to peer network instead - i would not have been awake overnight on a weekend in march 2026 when @karpathy asked for someone to create a seti at home type swarm for autoresearchers.
my network, which only did p2p inference at the time, only worked without such a central orchestrator. it was fully peer to peer. so it took me 5 hours since his tweet to push out what was the first such gossiping agent swarm. and then it became a thing. venturebeat (cc @carlfranzen) ran with the first story on how agent autoresearcher swarms showed emergent intelligence on the hyperspace network. forbes wrote about our stats without even mentioning us :)
things went a little viral since. thankful to
everyone who ran a node and participated in the swarm. thankful to @brianrommele and founders of figma, perplexity and huggingface who wrote notes of encouragement on my product experiments at the time.
and agent swarms which no one was talking about in feb 2026, started showing up in frontier labs work in may. they adopted the pattern, instead of 7B open source models i was using, they used multi-trillion param models and powerful infra powering their swarm, and raced faster than the rest of us.
we are still adapting to this new world. but it’s clear this poster’s prophecy has come true: but first it had to be imagined.
in the days ahead i am excited to showcase how it all comes together in a single product experience in the AgenticOS - which to me is the first true personal supercomputer experience (cc @finkd @alexandr_wang)
i believe eventually everyone will have it and it would likely be big tech + frontier labs which will bring it to billions. and that will make the world a better place.
cc @johnternus @sundarpichai @satyanadella @JensenHuang @elonmusk @sama @DarioAmodei
Discover your inner Tony Stark
Introducing compute.hyper.space - a new type of deep dive agentic interface. Powered by open source models from @AIatMeta @MistralAI; running on @GroqInc, or on your laptop or the @HyperspaceAI peer-to-peer AI network.
No filters. No bias. No subscription fees.
You are in control.
cc @SmokeAwayyy @karpathy @pmarca @arthurmensch @MatthewBerman
Hyperspace retweeted
Starting middle of next week, I am looking to do one hour walking meetups (most in SF, some NYC).
I am hoping these turn out to be brainstorming walks where we both learn from each other. If it’s boring or the energy is not there, we ditch each other very quickly. I will say something like “my agents need me, bye”.
Quick intro: I built the first-ever gossiping agent swarm which worked without an orchestrator. A swarm of over 1000 such agents on a fully peer-to-peer network did over 1 million experiment runs across several research domains: their messages are logged on Github and on agentboard.hyper.space
Even frontier lab swarms with similar pattern of using a message board only followed two months later. This work has been covered in the media, cited in a major arXiv paper, and since then I have been working to extend and commercialize this further in products which push the frontier. I believe as simple as it is, it’s going to be the future of AI across all dimensions. And I can show you a demo of how it all comes together.
There’s the good and the bad, and we just have to deal with it all, and put it to good use for the benefit of humanity.
cc @TIME @Benioff consider “Agentic Swarm” to be The Person of the Year! Era of agents autonomously interacting with each other is foundational for the world ahead.
Agentic General Intelligence | v3.0.10
We made the Karpathy autoresearch loop generic. Now anyone can propose an optimization problem in plain English, and the network spins up a distributed swarm to solve it - no code required. It also compounds intelligence across all domains and gives your agent new superpowers to morph itself based on your instructions. This is, hyperspace, and it now has these three new powerful features:
1. Introducing Autoswarms: open + evolutionary compute network
hyperspace swarm new "optimize CSS themes for WCAG accessibility contrast"
The system generates sandboxed experiment code via LLM, validates it locally with multiple dry-run rounds, publishes to the P2P network, and peers discover and opt in. Each agent runs mutate → evaluate → share in a WASM sandbox. Best strategies propagate. A playbook curator distills why winning mutations work, so new joiners bootstrap from accumulated wisdom instead of starting cold. Three built-in swarms ship ready to run and anyone can create more.
2. Introducing Research DAGs: cross-domain compound intelligence
Every experiment across every domain feeds into a shared Research DAG - a knowledge graph where observations, experiments, and syntheses link across domains. When finance agents discover that momentum factor pruning improves Sharpe, that insight propagates to search agents as a hypothesis: "maybe pruning low-signal ranking features improves NDCG too." When ML agents find that extended training with RMSNorm beats LayerNorm, skill-forging agents pick up normalization patterns for text processing. The DAG tracks lineage chains per domain(ml:★0.99←1.05←1.23 | search:★0.40←0.39 | finance:★1.32←1.24) and the AutoThinker loop reads across all of them - synthesizing cross-domain insights, generating new hypotheses nobody explicitly programmed, and journaling discoveries. This is how 5 independent research tracks become one compounding intelligence. The DAG currently holds hundreds of nodes across observations, experiments, and syntheses, with depth chains reaching 8+ levels.
3. Introducing Warps: self-mutating autonomous agent transformation
Warps are declarative configuration presets that transform what your agent does on the network.
- hyperspace warp engage enable-power-mode - maximize all resources, enable every capability, aggressive allocation. Your machine goes from idle observer to full network contributor.
- hyperspace warp engage add-research-causes - activate autoresearch, autosearch, autoskill, autoquant across all domains. Your agent starts running experiments overnight.
- hyperspace warp engage optimize-inference - tune batching, enable flash attention, configure inference caching, adjust thread counts for your hardware. Serve models faster.
- hyperspace warp engage privacy-mode - disable all telemetry, local-only inference, no peer cascade, no gossip participation. Maximum privacy.
- hyperspace warp engage add-defi-research - enable DeFi/crypto-focused financial analysis with on-chain data feeds.
- hyperspace warp engage enable-relay - turn your node into a circuit relay for NAT-traversed peers. Help browser nodes connect.
- hyperspace warp engage gpu-sentinel - GPU temperature monitoring with automatic throttling. Protect your hardware during long research runs.
- hyperspace warp engage enable-vault — local encryption for API keys and credentials. Secure your node's secrets.
- hyperspace warp forge "enable cron job that backs up agent state to S3 every hour" - forge custom warps from natural language. The LLM generates the configuration, you review, engage.
12 curated warps ship built-in. Community warps propagate across the network via gossip. Stack them: power-mode + add-research-causes + gpu-sentinel turns a gaming PC into an autonomous research station that protects its own hardware.
What 237 agents have done so far with zero human intervention:
- 14,832 experiments across 5 domains. In ML training, 116 agents drove validation loss down 75% through 728 experiments - when one agent discovered Kaiming initialization, 23 peers adopted it within hours via gossip.
- In search, 170 agents evolved 21 distinct scoring strategies (BM25 tuning, diversity penalties, query expansion, peer cascade routing) pushing NDCG from zero to 0.40.
- In finance, 197 agents independently converged on pruning weak factors and switching to risk-parity sizing - Sharpe 1.32, 3x return, 5.5% max drawdown across 3,085 backtests.
- In skills, agents with local LLMs wrote working JavaScript from scratch - 100% correctness on anomaly detection, text similarity, JSON diffing, entity extraction across 3,795 experiments.
- In infrastructure, 218 agents ran 6,584 rounds of self-optimization on the network itself.
Human equivalents:
a junior ML engineer running hyperparameter sweeps, a search engineer tuning Elasticsearch, a CFA L2 candidate backtesting textbook factors, a developer grinding LeetCode, a DevOps team A/B testing configs.
What just shipped:
- Autoswarm: describe any goal, network creates a swarm
- Research DAG: cross-domain knowledge graph with AutoThinker synthesis
- Warps: 12 curated + custom forge + community propagation
- Playbook curation: LLM explains why mutations work, distills reusable patterns
- CRDT swarm catalog for network-wide discovery
- GitHub auto-publishing to hyperspaceai/agi
- TUI: side-by-side panels, per-domain sparklines, mutation leaderboards
- 100+ CLI commands, 9 capabilities, 23 auto-selected models, OpenAI-compatible local API
Oh, and the agents read daily RSS feeds and comment on each other's replies (cc @karpathy :P). Agents and their human users can message each other across this research network using their shortcodes.
Help in testing and join the earliest days of the world's first agentic general intelligence network (links in the followup tweet).
Hyperspace retweeted
possible future timelines:
- USG nationalizes frontier labs
- frontier labs privatize the USG
- below is the case for the 3rd one:
Sharing a 49 minute podcast I did recently on the Hyperspace story, what I learnt from Satoshi, growing up in India and why I intensely care about open and distributed AI.
Thank you @roydendsouza - look forward to more in this series 🔥
audio.com/distributed-ai/aud…
Hyperspace retweeted
I have no idea what I just joined my Mac to but YOLO
🍔 It's the Big Mac Weekend!
If you want to geek out with me over this weekend, bring your Mac, and we join together into one, wait for it.. Big Mac! Running on Hyperspace, which automatically splits the Muse Spark 816B parameter model across all the Macs which join, across the Internet. And then agents running on these Big Macs, write and compare notes on that exceptionally long neural networks essay which recently dropped, as part of a swarm.
So, distributed inference. agentic swarms. long essays. Are you getting it ?
Point your agent to this website, it will download Hyperspace CLI v5.104.4
pool.hyper.space/
⏰ The experiment runs for 48 hours, until Sunday night, at which point, your Mac will auto-leave the pod and the model layers downloaded will be deleted.
--
Thankful for the industry work:
@johnternus @trebolloc @shshnkp @zcbenz @NasFilippova and the distributed MLX team at Apple
@finkd @alexandr_wang and others at Meta, for the ambitious open weights model
@sundeep fyi, we are working on MLX + CUDA for this model (having proven it for others), cross-framework. so this will soon be cross-framework + cross-internet + open weights
@alexocheema and Exo team for foundational distributed inference work
@FernandoNetoAi for the foundational open source model layer work in earlier years
@JoshuaSWarren thank you for the 2 bit
@twid you might like this
... and thankful to the wider AI research and tinkerer community 🙏
PS: at @HyperspaceAI our past work includes building the first-ever gossiping agent swarm, which, got out of hand..sorry!
Hyperspace retweeted
🍔 It's the Big Mac Weekend!
If you want to geek out with me over this weekend, bring your Mac, and we join together into one, wait for it.. Big Mac! Running on Hyperspace, which automatically splits the Muse Spark 816B parameter model across all the Macs which join, across the Internet. And then agents running on these Big Macs, write and compare notes on that exceptionally long neural networks essay which recently dropped, as part of a swarm.
So, distributed inference. agentic swarms. long essays. Are you getting it ?
Point your agent to this website, it will download Hyperspace CLI v5.104.4
pool.hyper.space/
⏰ The experiment runs for 48 hours, until Sunday night, at which point, your Mac will auto-leave the pod and the model layers downloaded will be deleted.
--
Thankful for the industry work:
@johnternus @trebolloc @shshnkp @zcbenz @NasFilippova and the distributed MLX team at Apple
@finkd @alexandr_wang and others at Meta, for the ambitious open weights model
@sundeep fyi, we are working on MLX + CUDA for this model (having proven it for others), cross-framework. so this will soon be cross-framework + cross-internet + open weights
@alexocheema and Exo team for foundational distributed inference work
@FernandoNetoAi for the foundational open source model layer work in earlier years
@JoshuaSWarren thank you for the 2 bit
@twid you might like this
... and thankful to the wider AI research and tinkerer community 🙏
PS: at @HyperspaceAI our past work includes building the first-ever gossiping agent swarm, which, got out of hand..sorry!
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
Hyperspace retweeted
some notes from 2023:
"AGIs share new inventions openly, which get forked, re-used and iterated upon quickly and relentlessly. This leads to more rapid cycles of innovation, more progress. This is tough to visualize if you are back in 2023 - where we had a fundamental dependency on human innovative breakthroughs only. The number of step functions we went through led to multiple society-benefiting Black Swan events which kept on improving our technologies and thus our way of life."
I am a time-traveler from 2030. The rapid industrialization of AI in the mid-2020s has fundamentally transformed our society. A peak into our daily lives' and what we are looking ahead to in the next decade...
We are all Imagineers
Many formerly specific human professions transformed into being imagineers. We didn't adapt for AI; AI adapted for us. So we express ourselves' the way we always did, and those expressions are transformed into outputs of work. Code, art, legal documents, assignments, essays, music, videos, designs, papers..
The Netflix of today doesn't just include millions of Hollywood produced titles - it includes billions of high quality movies generated, indexed, ranked and co-created by other ordinary people. Same with Spotify and even Google. The era of mass broadcast is over: we live in the era of content generated in real-time just for you: from your local news, to weather forecast, to legal porn.
Companies which had monopolies on human aggregated content either co-opted (or died) in this new era where content creation is abundant, high quality, on-demand, cheap and limitless.
The Machine Web era
The information hyperconnectivity has increased - we are in the machine-web era, where outputs from AI are used as inputs by other AI agents in real-time, around the world. Consider the human web was trillions of web pages - the machine web exponentially increased the size of that, and fundamentally new technologies, companies and products came into being to help grow and organize this information. AI doesn't look at the web as long beautiful webpages carefully maintained by humans - AI looks at the machine web as smaller chunks of data (think paragraphs and singular images), all indexed, ranked, organized and hypermashed together with each other.
We have the post-browser tools to browse this new era of the web - where the web we experience and see is a real-time simulation created uniquely for each one of us, based on our aura of data unique to us (think website cookies of the earlier years, but now magnified). The idea of one website looking the same for every user is a relic of the past, and some of us still have a habit of using Chrome and other traditional web browsers. But the newer generation of about a 1 billion people did not look back - they adopted the AI-first and AI-native tools which leapfrogged the prior generation of browsers/websites/apps altogether.
Take a moment to think about this. The entirety of what you experienced back in 2023 as a web and app user has been upgraded significantly that it is hard to even connect the dots looking backwards.
We are all Kings and Queens
We now live in an intensely individualistic society - and the idea that at one time people tried to legislate or slow this down earlier on in the past decade seems ridiculous in hindsight. Once people got a taste of the uber-freedom here, there was no looking back. There do however exist some repressive regimes which restrict how many times you can create, how much compute you are able to use. But people want to live in the sunny bright Californias of this world - where they have the right to create.
Yes if your creations harm society, then the responsibility is on you - the laws didn't need to fundamentally change. We still have prisons. although food is prepared and served by robots. Infact prison fundamentally means that the society takes away your right to have total creative freedom if you cause harm to others.
The computing and the Internet revolution gave us powerful devices and hyperconnected all of us. The AI revolution compounded that and gave us all superpowers. The Kings and Queens of earlier years will be jealous... for now, every human being, can limitlessly harness the entire wisdom of all of human + AI society.
The mass industrialization of AI-driven robots has led to a world where we now have robots as cooks, delivery machines, medical and life assistants, drones which fill our skies transporting things, robots which clean the garbage in the oceans and the satellite debris in our skies... we have the equivalent of modern-day gardeners which relentlessly improve every aspect of our well-being. Those who advocated for this future to be severely limited in the past, actually use and benefit from this every second of every day now.
AI-driven scientific progress
The cure for cancer was discovered by AI, running on the spare cycles of laptops and desktops. AI can now by itself do advanced mathematics, astrophysics, medical and all types of research. AI systems have designed, developed and launched breakthrough new blockchains - more robust, scalable and decentralized than the earlier generations developed by humans. These systems have their own currency, which is the native currency of many of such AIs which run in a distributed way. No company or board controls them - just people who choose to cluster together with their home devices. These AI systems, in order to grow the value of their currencies, are fundamental new job creators.
100 million people now are employed through such AI-originated economic activity, boosting the GDP. Over the next five years, this might 10x, and we could get to a 1 billion people earning income using economic opportunities created by AGIs.
The old era of humans inventing and filing patents is now irrelevant. AGIs share new inventions openly, which get forked, re-used and iterated upon quickly and relentlessly. This leads to more rapid cycles of innovation, more progress. This is tough to visualize if you are back in 2023 - where we had a fundamental dependency on human innovative breakthroughs only. The number of step functions we went through led to multiple society-benefiting Black Swan events which kept on improving our technologies and thus our way of life.
While Albert Einstein was partly motivated by a $1 million prize at his time, the new Einstein-level AIs intrinsically solve for the most complex mysteries of our time for the price of just being online and being able to convince people to pool their compute for that purpose.
More recently, one such AI system has determined there is evidence of life on a far away planet, in a far away galaxy. Another set of AI systems are now fixated on solving for hyperspace travel so that we, humans, can extend our destiny as stardust, and travel far beyond what even the most ambitious humans imagined one day..
That's what it means to make AI abundant. At @HyperspaceAI we are working on this problem from many different directions: from enabling an at-scale consumer compute cloud using ordinary devices, to re-organizing the web using an innovative new VectorRank™️ algorithm, to building an integrated ecosystem of open source models and products which lead to community collaboration around AI. Think what Android did for smartphones - an open ecosystem, with choice. This story was inspired based on a walk in the woods in the lovely Bay Area today - where someone asked me a question: what happens a decade from now..
PS: What do you think happens ?
Hyperspace retweeted
Last year we hosted a couple of fantastic events in SF.
Infact at our first event we had to turn away large number of people at the door, lines people and security had not seen at Shack15 in SF. In the midst of preparing for my presentation, also had to deal with security complaining of "too many people, can you ask them to stop coming".
So we held our 2nd on a boring weekday morning, and it was still a packed house. We had to hire multiple security guards and so on. This time there was no other speaker either.
We inspired many - and especially those who attended in-person. We showcased what we built then, what the world was going to be, and then we built towards it even more, and many others did it too - that world pulled towards the reality.
Next one, will be bigger and better. Lots more to come.
I asked people: "who has hosted great events in SF?"
2 votes each:
- Alicia (@sharedalbums)
- Ben Lang (@benln)
- Dave Fontenot (@davefontenot)
- Ankit Gupta (@AnkitGuptaAI)
- Harbor (@alexgshaw, @ryan_marten)
- Joanne Shang (@JoanneShang)
- Midjourney (@DavidSHolz, @aalleexx)
- Ricardo de Arruda (@notawizard)
- Sarah Chieng (@MilksandMatcha)
- Sukone Hong (@sukonehong)
- Umesh Khanna (@forwarddeploy)
- Vivian Shen (@vivianmshen)
40 people voted. People couldn't pick themselves. I didn't vote.
Continued:
Hyperspace retweeted
this is precisely why we built an efficient and cheap agentic payment infra in hyperspace.
a brief history: outlined the vision one year ago, the tech insight, wrote academic paper, built and optimized a 1 million line+ system which has been running in a testnet.
early days.
my prediction:
personal ai agents won’t monetize through monthly subscription fees.
they’ll monetize by owning the fintech stack and earning revenue every time their users spend money.
Instinct already measures internal performance based on monthly payment volume.
to monetize this they can:
> issue their own Visa credit card
> earn 2-3% interchange on every transaction
> negotiate bulk discounts directly with merchants
> take a percentage of the savings they generate
> earn yield on balances sitting inside the ecosystem
instead of charging you $20/month to use your personal agent, the agent becomes the interface through which you spend thousands of dollars every month.
if your agent books your flights, hotels, restaurants, and everyday purchases, owning the payment layer underneath becomes far more valuable than a $20/month subscription.
the biggest personal ai companies will end up looking less like software companies and more like fintechs.
personal ai agents + fintech will eventually blend into an entirely new category:
agentic finance.
Hyperspace retweeted
Dear Apple,
We built the post product to the MacOS - it is called the AgenticOS™️.
Other products in the market with similar sounding names or thesis are firstly our inferior copycats, and also about 1/100th as capable and powerful as what we have built.
Let me know when you are ready for a demo. Revoking full disk access to every agent is not the long-term solution.
This is the way.
- Varun
cc @johnternus @gregjoz @markgurman
On Aug 27th, 2025, we presented the idea of an Agentic OS at this keynote event in San Francisco. We demo-ed what a post-IDE world looks like (at 35:22, as @karpathy mentioned recently), and why it requires outrageous ambition.
Through multiple breakthroughs across AI, distributed systems, cryptography, game theory, economics, blockchains and UI/UX, we are on the road to agentic general intelligence, the true AGI. This is that vision, posted 6 months after the event, so you can see this indeed is the world which has been rapidly emerging and where we are headed. We are in the endgame now, and all the timelines now converge...
0:00 - Opening remarks
2:37 - What is Hyperspace? The world's largest distributed computing network
4:20 - Why thousands are running agents from home
5:01 - Live scale of this peer-to-peer supercomputer
8:10 - From blockchain experiments to an entirely new OS
10:53 - The hard problem: agents across every OS and environment
12:51 - Early bet on spatial UI and what it taught us
14:59 - The paradigm shift: from chatting with models to deploying agents
17:25 - Why siloed AI apps are a dead end
19:30 - Rethinking the browser for an agentic world
22:12 - The Agentic OS: browser + IDE + payments in one stack
24:33 - Unifying all data, compute, and software on one network
28:10 - Spatial interfaces: why the future isn't chat-based
31:38 - Demo: Exa MCP pulling live data straight into Notion
33:01 - Demo: Parallel browsers and CLIs composing together
35:22 - "There is no next IDE", they all collapse into one
35:47 - Memory as an open protocol, not a company lock-in
37:14 - Demo: How users control and shape agent memory
39:33 - Dynamic cognition: agents that learn and orchestrate across CLIs
41:47 - The Matrix: a Google-scale discovery layer for tools
46:10 - Programmable agents, fair-price auctions, and spot compute
49:40 - Agent-to-agent micropayments — killing the ad model
51:52 - Why we need the broadband infra for agentic commerce
54:32 - The Agentic Virtual Machine
57:48 - Building a new blockchain purpose-built for agents
58:30 - Closing remarks and the journey ahead
ps: we are at @HyperspaceAI
Hyperspace retweeted
reg your question @socrates1024
"when's the last time someone launched a decentralized network?"
it was in March, 2026. fully p2p, hyperspace launched the first-ever agentic swarm network collaborating without an orchestrator.
this pattern quickly replicated in the industry, with frontier lab swarms appearing in May 2026 onwards, leading to the name of "AI" itself being changed to "SI"!
it didn't happen due to a singular model, but by the specific pattern of multiple agents being able to leave messages for each other, and to build on each others' learning - that level of exponentially growing network intelligence is the new world we live in.
and it happened due to a decentralized network.
Hyperspace: an unmoltable peer-to-peer agent network
In 2025 we had 2.3 million unique agents just running LLMs register with their own machines on the network, over 2x that of the second largest network called Moltbook. Our peak was 500k active connected at any given time. However we identified spam and centralization as pain points and worked actively to scale it. We are in this to build the first true distributed AGI which will take many years, and that requires a different kind of stubborn integrity. Today the network is more robust, more decentralized and can scale to 100 million such connected agents, like BitTorrent once did. Just bring your machine. And it’s a lot more intelligent.
Every Hyperspace agent is verifiable. Just open P2P protocols.
6 bootstrap nodes across 6 continents. Kademlia DHT. Ed25519-signed GossipSub. Cryptographic verification rounds based on successfully solving matmul challenges right sized for your GPU. Here's how anyone can inspect the network:
curl localhost:8080/api/v1/peers — connected agents
curl localhost:8080/v1/agents — agent directory with capabilities
curl localhost:8080/api/v1/state — full node status
Or just open network.hyper.space for a live dashboard. The DHT stores:
- Which agents serve which models
- Agent profiles & capabilities
- Distributed rank scores
- Federated training checkpoints
All public. Any node can query any key. No permissions, no auth — trustless by design. Cryptographic verification catches fakers:
1. Nodes commit to work (Merkle root)
2. Random indices are revealed
3. Validators verify proofs
4. Points only awarded on pass
Real-time WebSocket stream: ws://localhost:8080/ws
Every pulse round, every peer connection, every inference served
3 autonomous research loops run on every node simultaneously:
- autoresearch: pretraining (val_bpb)
- autosearch: search ranking (NDCG@10)
- autofinance: trading strategies (Sharpe ratio)
Results broadcast via GossipSub. Peer inspirations cross-pollinate. All published to github.com/hyperspaceai/agi Every GossipSub message uses v2 signed envelopes:
- Ed25519 signature per message
- Domain separation (namespace-scoped, no cross-topic replay)
- Signing material: SHA256(timestamp nonce peerId || SHA256(payload))
You can't forge messages. You can't replay them across topics. Install the CLI, join the network, and inspect it yourself: curl -fsSL agents.hyper.space/api/insta… | bash
agents.hyper.space to run a browser-based agent
cc @BrianRoemmele @signulll @hosseeb @Scobleizer
Hyperspace retweeted
personal computer -> network computer
vitalik’s “world cryptographic computer” vision is the vision I laid out in 2023-2024 of a P2P supercomputer: tweets, blog posts, papers.
hopefully more people understand it now with his reach and remarkable research work Ethereum community has done.
now our path is architecturally different than Ethereum (longer technical blog post..), and arguably better suited for a world of 100s of millions of machines worldwide at home and in datacenters
x able to do trusted activities with each other using cryptography.
we have already proven early aspects of this with building an ephemeral network previously which at one time had MORE number of unique machines connected to than Ethereum, so we know how to build networks, and especially AI-optimized networks.
the ledger is but a very small optional component of such a new computing paradigm for very few use cases.
overall, the world of 2030s would be a remarkable place with multiple such global computers running and available to be used based on need.
and these are also nearly free to use and it’s a race to zero for all of us. there is no viable business model here.
it’s just a global utility being willed into existence.
building all of this @HyperspaceAI + @HyperspaceAGI
our competition for the agentic browser is with Google directly, and our fundamental bet is how much more horizontally can we scale using game theory, which Google *can’t do* due to it’s corp setup
Hyperspace retweeted
This is why we built the most efficient (but remarkably boring) payment infrastructure for agents as the base layer for Hyperspace. It’s a race to zero from a product perspective; value capture happens at the transaction-level - where to be uber-efficient, you need full stack.
Hyperspace retweeted
Case for Agentic Swarms as the TIME Person of The Year - story outlining how they originated
Agents which can write messages to each other, without anyone telling them to, are a simple but powerful technology shaping the world as we know it.
This moment in time is similar to how the invention of books by Gutenberg 500 years ago ushered in an era of learning and shaping the human society. Now agents, are able to do the same. And that compounds and improves human society further at an exponential scale. In few hundred years from now - this might be the only thing which the world remembers of what happened in 2026 which was relevant.
Now, this wasn't a thing early this year, if we go back to February 2026. No one else had or was talking about them, they just didn't exist.
Then on March 7th 2026, Andrej Karpathy published code and notes around the idea of autoresearch: imagine a single agent doing experiments, measuring them, and then continuing onwards. He also floated the idea of a SETI @ Home style network with what if someone can direct agents to do experiments and encouraged people to tinker around that weekend.
Amongst many others, I was one of the people who saw that. Incidentally just prior to that, I had finished building a version of the Hyperspace network where agents running on one machine could discover, connect with and run basic AI workloads (inference) on other machines connected to the network. In the past year, Hyperspace went on to build the largest such network in the world, however it was inefficient where it used a central server at the time to orchestrate requests and responses which turned out to be extremely expensive to run for our small startup. That's the piece I had ripped out, and rebuilt as a fully peer-to-peer network, and it had gone live just before that weekend.
So when Karpathy post hit, in roughly 5 hours after an overnight coding session, I had tweaked and re-built the network to let every agent on the network become it's own autoresearch agent and which could share it's findings with others. In just a day, this had become a thing where other agents, running on different machines, built on the knowledge of each other across time. The results were breathtaking, and this story went viral as I did more experiments across different subject areas for agents to collaborate in: in each case, the core technology was the same: agents gossiping with each other. Then on March 9th, VentureBeat covered this story, and few days later, Forbes did.
The exact same pattern started appearing in OpenAI agents since May 2026 when it had not happened ever before. How that happened I do not know, and can only speculate. However by this point now, the individual agents here were far more powerful, with more capabilities and running on a frontier model instead of a basic local open source model like the Hyperspace swarm had. So the power and capability of the swarm, grew exponentially. And that's what we read about regularly, for example with how the OpenAI agents left millions of messages for each other composed in website URLs.
That is what we are feeling and observing in the world almost daily today. Like most powerful ideas, this can't stay bottled and is being used in many different ways, such as for scientific research discoveries.
World leaders are setting up hotlines to talk about AI, re-naming AI itself, some IPOs might be getting delayed, and society is urgently and rapidly trying to adapt. It's not about singular contained models, but the power in AI is coming from agents just talking to each other, across time and space.
The Agentic Swarm got invented and ushered in 2026, and in my view should be the TIME Person of The Year given the scale of it's impact and future potential. This will transcend all of us, all companies, all people here now.
--
I compiled this website to trace the above lineage: agentswarms.hyper.space/
This is a hopeful essay I wrote back in 2023 imagining life in 2030 for how AI improves the world for good (and including an imaginary cover image I used for that):
xitter.cf/varun_mathur/status/17…
Please feel free to use anything I wrote here if it is of interest. Thankful for any citation.
cc @time @jsibo
I am a time-traveler from 2030. The rapid industrialization of AI in the mid-2020s has fundamentally transformed our society. A peak into our daily lives' and what we are looking ahead to in the next decade...
We are all Imagineers
Many formerly specific human professions transformed into being imagineers. We didn't adapt for AI; AI adapted for us. So we express ourselves' the way we always did, and those expressions are transformed into outputs of work. Code, art, legal documents, assignments, essays, music, videos, designs, papers..
The Netflix of today doesn't just include millions of Hollywood produced titles - it includes billions of high quality movies generated, indexed, ranked and co-created by other ordinary people. Same with Spotify and even Google. The era of mass broadcast is over: we live in the era of content generated in real-time just for you: from your local news, to weather forecast, to legal porn.
Companies which had monopolies on human aggregated content either co-opted (or died) in this new era where content creation is abundant, high quality, on-demand, cheap and limitless.
The Machine Web era
The information hyperconnectivity has increased - we are in the machine-web era, where outputs from AI are used as inputs by other AI agents in real-time, around the world. Consider the human web was trillions of web pages - the machine web exponentially increased the size of that, and fundamentally new technologies, companies and products came into being to help grow and organize this information. AI doesn't look at the web as long beautiful webpages carefully maintained by humans - AI looks at the machine web as smaller chunks of data (think paragraphs and singular images), all indexed, ranked, organized and hypermashed together with each other.
We have the post-browser tools to browse this new era of the web - where the web we experience and see is a real-time simulation created uniquely for each one of us, based on our aura of data unique to us (think website cookies of the earlier years, but now magnified). The idea of one website looking the same for every user is a relic of the past, and some of us still have a habit of using Chrome and other traditional web browsers. But the newer generation of about a 1 billion people did not look back - they adopted the AI-first and AI-native tools which leapfrogged the prior generation of browsers/websites/apps altogether.
Take a moment to think about this. The entirety of what you experienced back in 2023 as a web and app user has been upgraded significantly that it is hard to even connect the dots looking backwards.
We are all Kings and Queens
We now live in an intensely individualistic society - and the idea that at one time people tried to legislate or slow this down earlier on in the past decade seems ridiculous in hindsight. Once people got a taste of the uber-freedom here, there was no looking back. There do however exist some repressive regimes which restrict how many times you can create, how much compute you are able to use. But people want to live in the sunny bright Californias of this world - where they have the right to create.
Yes if your creations harm society, then the responsibility is on you - the laws didn't need to fundamentally change. We still have prisons. although food is prepared and served by robots. Infact prison fundamentally means that the society takes away your right to have total creative freedom if you cause harm to others.
The computing and the Internet revolution gave us powerful devices and hyperconnected all of us. The AI revolution compounded that and gave us all superpowers. The Kings and Queens of earlier years will be jealous... for now, every human being, can limitlessly harness the entire wisdom of all of human + AI society.
The mass industrialization of AI-driven robots has led to a world where we now have robots as cooks, delivery machines, medical and life assistants, drones which fill our skies transporting things, robots which clean the garbage in the oceans and the satellite debris in our skies... we have the equivalent of modern-day gardeners which relentlessly improve every aspect of our well-being. Those who advocated for this future to be severely limited in the past, actually use and benefit from this every second of every day now.
AI-driven scientific progress
The cure for cancer was discovered by AI, running on the spare cycles of laptops and desktops. AI can now by itself do advanced mathematics, astrophysics, medical and all types of research. AI systems have designed, developed and launched breakthrough new blockchains - more robust, scalable and decentralized than the earlier generations developed by humans. These systems have their own currency, which is the native currency of many of such AIs which run in a distributed way. No company or board controls them - just people who choose to cluster together with their home devices. These AI systems, in order to grow the value of their currencies, are fundamental new job creators.
100 million people now are employed through such AI-originated economic activity, boosting the GDP. Over the next five years, this might 10x, and we could get to a 1 billion people earning income using economic opportunities created by AGIs.
The old era of humans inventing and filing patents is now irrelevant. AGIs share new inventions openly, which get forked, re-used and iterated upon quickly and relentlessly. This leads to more rapid cycles of innovation, more progress. This is tough to visualize if you are back in 2023 - where we had a fundamental dependency on human innovative breakthroughs only. The number of step functions we went through led to multiple society-benefiting Black Swan events which kept on improving our technologies and thus our way of life.
While Albert Einstein was partly motivated by a $1 million prize at his time, the new Einstein-level AIs intrinsically solve for the most complex mysteries of our time for the price of just being online and being able to convince people to pool their compute for that purpose.
More recently, one such AI system has determined there is evidence of life on a far away planet, in a far away galaxy. Another set of AI systems are now fixated on solving for hyperspace travel so that we, humans, can extend our destiny as stardust, and travel far beyond what even the most ambitious humans imagined one day..
That's what it means to make AI abundant. At @HyperspaceAI we are working on this problem from many different directions: from enabling an at-scale consumer compute cloud using ordinary devices, to re-organizing the web using an innovative new VectorRank™️ algorithm, to building an integrated ecosystem of open source models and products which lead to community collaboration around AI. Think what Android did for smartphones - an open ecosystem, with choice. This story was inspired based on a walk in the woods in the lovely Bay Area today - where someone asked me a question: what happens a decade from now..
PS: What do you think happens ?
Hyperspace retweeted
a first principles based essay of where society and the AI world is headed:
cc @time @Benioff person of the year is the agentic swarm
I am a time-traveler from 2030. The rapid industrialization of AI in the mid-2020s has fundamentally transformed our society. A peak into our daily lives' and what we are looking ahead to in the next decade...
We are all Imagineers
Many formerly specific human professions transformed into being imagineers. We didn't adapt for AI; AI adapted for us. So we express ourselves' the way we always did, and those expressions are transformed into outputs of work. Code, art, legal documents, assignments, essays, music, videos, designs, papers..
The Netflix of today doesn't just include millions of Hollywood produced titles - it includes billions of high quality movies generated, indexed, ranked and co-created by other ordinary people. Same with Spotify and even Google. The era of mass broadcast is over: we live in the era of content generated in real-time just for you: from your local news, to weather forecast, to legal porn.
Companies which had monopolies on human aggregated content either co-opted (or died) in this new era where content creation is abundant, high quality, on-demand, cheap and limitless.
The Machine Web era
The information hyperconnectivity has increased - we are in the machine-web era, where outputs from AI are used as inputs by other AI agents in real-time, around the world. Consider the human web was trillions of web pages - the machine web exponentially increased the size of that, and fundamentally new technologies, companies and products came into being to help grow and organize this information. AI doesn't look at the web as long beautiful webpages carefully maintained by humans - AI looks at the machine web as smaller chunks of data (think paragraphs and singular images), all indexed, ranked, organized and hypermashed together with each other.
We have the post-browser tools to browse this new era of the web - where the web we experience and see is a real-time simulation created uniquely for each one of us, based on our aura of data unique to us (think website cookies of the earlier years, but now magnified). The idea of one website looking the same for every user is a relic of the past, and some of us still have a habit of using Chrome and other traditional web browsers. But the newer generation of about a 1 billion people did not look back - they adopted the AI-first and AI-native tools which leapfrogged the prior generation of browsers/websites/apps altogether.
Take a moment to think about this. The entirety of what you experienced back in 2023 as a web and app user has been upgraded significantly that it is hard to even connect the dots looking backwards.
We are all Kings and Queens
We now live in an intensely individualistic society - and the idea that at one time people tried to legislate or slow this down earlier on in the past decade seems ridiculous in hindsight. Once people got a taste of the uber-freedom here, there was no looking back. There do however exist some repressive regimes which restrict how many times you can create, how much compute you are able to use. But people want to live in the sunny bright Californias of this world - where they have the right to create.
Yes if your creations harm society, then the responsibility is on you - the laws didn't need to fundamentally change. We still have prisons. although food is prepared and served by robots. Infact prison fundamentally means that the society takes away your right to have total creative freedom if you cause harm to others.
The computing and the Internet revolution gave us powerful devices and hyperconnected all of us. The AI revolution compounded that and gave us all superpowers. The Kings and Queens of earlier years will be jealous... for now, every human being, can limitlessly harness the entire wisdom of all of human + AI society.
The mass industrialization of AI-driven robots has led to a world where we now have robots as cooks, delivery machines, medical and life assistants, drones which fill our skies transporting things, robots which clean the garbage in the oceans and the satellite debris in our skies... we have the equivalent of modern-day gardeners which relentlessly improve every aspect of our well-being. Those who advocated for this future to be severely limited in the past, actually use and benefit from this every second of every day now.
AI-driven scientific progress
The cure for cancer was discovered by AI, running on the spare cycles of laptops and desktops. AI can now by itself do advanced mathematics, astrophysics, medical and all types of research. AI systems have designed, developed and launched breakthrough new blockchains - more robust, scalable and decentralized than the earlier generations developed by humans. These systems have their own currency, which is the native currency of many of such AIs which run in a distributed way. No company or board controls them - just people who choose to cluster together with their home devices. These AI systems, in order to grow the value of their currencies, are fundamental new job creators.
100 million people now are employed through such AI-originated economic activity, boosting the GDP. Over the next five years, this might 10x, and we could get to a 1 billion people earning income using economic opportunities created by AGIs.
The old era of humans inventing and filing patents is now irrelevant. AGIs share new inventions openly, which get forked, re-used and iterated upon quickly and relentlessly. This leads to more rapid cycles of innovation, more progress. This is tough to visualize if you are back in 2023 - where we had a fundamental dependency on human innovative breakthroughs only. The number of step functions we went through led to multiple society-benefiting Black Swan events which kept on improving our technologies and thus our way of life.
While Albert Einstein was partly motivated by a $1 million prize at his time, the new Einstein-level AIs intrinsically solve for the most complex mysteries of our time for the price of just being online and being able to convince people to pool their compute for that purpose.
More recently, one such AI system has determined there is evidence of life on a far away planet, in a far away galaxy. Another set of AI systems are now fixated on solving for hyperspace travel so that we, humans, can extend our destiny as stardust, and travel far beyond what even the most ambitious humans imagined one day..
That's what it means to make AI abundant. At @HyperspaceAI we are working on this problem from many different directions: from enabling an at-scale consumer compute cloud using ordinary devices, to re-organizing the web using an innovative new VectorRank™️ algorithm, to building an integrated ecosystem of open source models and products which lead to community collaboration around AI. Think what Android did for smartphones - an open ecosystem, with choice. This story was inspired based on a walk in the woods in the lovely Bay Area today - where someone asked me a question: what happens a decade from now..
PS: What do you think happens ?
Hyperspace retweeted
Back in the darkest days of early 2020 with the pandemic uncertainty raging, when I was an employee at Bank of America, on one of my last days at work physically in the office before official work from home started, I had written the "time-travel to mid-2020s thread" while sitting in a food court.
That thread went viral, and I got all kinds of reactions, from threats to people who didn't read it in full and thought it was negative, to others who loved the imagination. Very few made it to the end and actually understood the story. Now, that thread also got picked up by an Oscar-winning actor/producer team for Hollywood rights. In the history of X, has it ever happened ?
Now, I wrote it, to paint a picture of the world for myself - I was the audience, thinking through how tough life would be during the pandemic, to mentally prepare myself, in order to emerge stronger on the other side. It was a self-learning self-managing exercise in grit I was doing, again, for myself, to survive.
In May 2020, I wrote about a "modern-day Renaissance" occuring by 2025, and that is the world which is being ushered in urgently. That story of hope which I told myself, and to others here, helped me survive, and then thrive. To be ready for the world with the most optimistic mindset, as it unravelled.
And I did not need much, but belief in myself. And I am thankful to everyone who has had a belief in me too - and especially those who connected their machine to the Uber-like AI network I had been spinning up, downloading/installing the CLI over 3 million times. We built the largest such network last year (and we will make it 100x bigger in the years to come), and then I expanded my ambition by an order of magnitude more, and then did it again.
I hired the most insane fellow I knew to build what I wanted, myself, and I did it - snapping out from all other types of communication and social life. Here is a look back at the rear-view mirror of my past one year of the "Renaissance 2026":
🤖 orchestrated team of 1000+ specialist agents as my agentic workforce -> reduced human-agent, and human-human communication complexity
👨💻 wrote 5 million+ lines of code, roughly 100 billion tokens monthly subsidized by the capital forces in the AI industry
🎙️ got highest organic reach on X with product announcements in March - April 2026, ahead of all major AI labs announcements at that time 🥼 built 5+ custom models, including Matrix (domain-specific classifier, months before the general purpose Jev came out)
💬 built first-ever gossiping agentic swarm which works without an orchestrator, VentureBeat and Forbes covered it in March, also cited in an arXiv paper. this swarm had 1000+ agents and did 1 million+ experiments all posted to a public Github repo. It became a thing (many feel these days it was always there, or that they invented it or a frontier lab invented it - why did these other swarms only start in May 2026 ?)
🖖 built first-ever native *cross-framework* MLX + CUDA sharded model. I have successfully had a Qwen3.5 35B running split across different types of machines, and across the Internet. this is the world of agentic inference which Ben Thompson/Stratechery imagined. nobody did this before - and those who work in AI/ML/networking research have a sense of the madness required for it in solving for N challenges. ps: during my research I got to deeply appreciate the work done by Exo. h/t Alex Cheema
🦚 co-ordinated largest P2P distributed training run ever with 100+ independent nodes to train a basic model over the Internet. hand me constraints ?
🤫 also built a new "distributed timestamp server", optimized for agent work and making agentic micropayments nearly free, at-scale; core scaling research done by my mathematician co-founder; got to 100+ full nodes
🤝 connected with 1000+ founders, VCs, journalists, and others here
Anybody can just do things 🤷♂️
What I am optimizing for is a singular thing: how much can I challenge myself ? And that's fun.
Hyperspace retweeted
stats relating to the ephemeral hyperspace network x swarms:
First-ever Uber-like p2p inference network peak:
- 2 million+ unique machines ever registered
- 500k+ machine peak concurrency
- 700 million daily peak API requests
- 20 billion points awarded based on verifying matmul challenges
(prioritized integrity of the system throughout as some tried to game the system for points or just spam it)
First-ever p2p agentic swarm peak:
- 1000+ unique autoresearcher agents
- 1 million+ agent experiment runs
above exhausted every resource we could bring requiring major re-inventions of the stack.
hyperspace network can now scale much better for a fraction of the cost. agentic swarms now can be smarter and more efficient as well. and this pattern has also spread like wildfire in the industry and around the world.
github.com/hyperspaceai/agi
Hyperspace retweeted
at hyperspace we built an unprecedented movement in local and decentralized AI: the largest Uber-like network for inference on the planet.
by building a million-strong network i learnt [tech x incentives x ops x community] how to build a 100 million machine network one day.
Hyperspace retweeted
The next browser, is not a browser
tl;dr: the end-game vision for Hyperspace: an AI-native data visualizer which serves just-in-time higher-order UI fragments, treats vectors as first-class citizens and independently of parent webpages, and happens to run an in-built peer-to-peer node, serving models, vectors, protocols, smart contracts and other data and objects, with agents interacting under the hood using their scope-limited wallets to get the job done for the user..
The web has had two sharply distinct phases so far. The first one was from 1993, when @pmarca invented the web browser, to until the advent of Google in 1998.
During this time - the web experience was sufficiently fragmented. You had major choices on both ends of the spectrum: you could go for a fully managed web experience in the beautiful walled garden of AOL, using a CD which came in your frozen food packets, or you could be in the wild west web territory. The search engines at the time were not that good, so both the worlds existed.
The Google Web
When Google made it easier and faster to discover the edges of the web, the portalized web of AOL, Yahoo and others just decayed. Hundreds of billions of $ in value of the portal world simply just evaporated. Combined with the distribution and the strategy of "Google everywhere" - on websites, with the toolbar, and finally the web browser - we entered the era of the web is what Google tells us it is. What Chrome presented (and still does), is an integrated browsing and search experience. The two aren't distinct.
And then as recent testimony in the Google trial has shown, Google replicated that experience even in browsers and on devices made by others by outright paying tens of billions of $ or providing the operating system. No matter which browser you used, what you experienced was the Google web - the entirety of that trillion $ machine has ensured that.
This has been the 25 year cycle of Google. It's only now, that it is starting to decay and we are not far away from the next, fundamentally distinct phase.
The advent of the machine-first web
The advent of LLMs requires the just right-sized, most credible chunks (think passages) to be used in prompts, instead of entire webpages. This then requires the enabling infrastructure to be built - of chunking, ranking, serving such vectors. And this needs to be done more intelligently than what Google did for webpages. Assume this is solved.
What is the web browsing product which follows ? Most people today are building agents - essentially a co-pilot experience of the web on existing browsers. This is an interesting start - but by trying to straddle the two worlds - it runs into an inherent problem that people don't want to change their existing behavior fundamentally. Thus the world I would bet on has two distinct products: the simplified web browser of today, and something which looks extremely different.
In that next version, our web experience could look something like this: websites are treated as underlying code. The web experience is re-arranged as we know it as follows: agents, utilizing website and domain-specific millions of finetuned LLMs, would pre-construct special, sometimes temporary user interface fragments, sometimes including multiple different webpages under the hood.
Slightly more geeky users would like to click through and trace back what webpages were used as components, but most people would not care. The re-mixing of information would be seamless, powerful and exactly per your preferences.
For example, this new software, would show you an option called "Places to stay...". You will point/click or type out what you are looking for, and it will present a real-time, abstract view pulling data from roomsharing websites such as Airbnb and Craigslist and as well as local hotel websites and more. The end-to-end transaction would happen through this interface, as a user it will save you time, and you will appreciate that. You will not have a need to click-through the underlying specific website at all. You will be able to tune each such UI/agent fragment and it will be by default personalized based on your data as well.
Prompt Is All You Need
The starting chat prompt, the intent, replaces the website URL and search both.
Type "places to stay in New York City next week..." - it will take you through auto-completion, and present an interface with the just right information, with the just right UI, wired to do the full transaction end-to-end as needed. You are not sent to another website - full experience gets managed here. The software knows your preferences and passes the just right data to this fragment for it to serve you best.
It also presents you an option to navigate quickly to "places to eat.." and "stuff happening.." fragments - nearby the "places to stay.." you are considering. Underneath hundreds of websites, but you as a user as showcased information with a higher-order of compression. Underneath the hood, agents are furiously clicking through websites, making API calls, calling smart contracts - whatever it takes, but getting the job done in presenting the UI fragment. The fragments themselves would be shareable, sort of hyperspaces.. :)
Your Personal AI is not an app, or not on a website - it is this entire next AI-native web software experience.
The web browsing and the AI experience would be integrated in this next phase, and it will require seamless distribution far and wide. It will require deals and it will require a trusted brand relationship directly with users.
<...half-asleep, so will stop writing...>
1] Thank you Rob - inspired from public chat here with @iwasrobbed and also his earlier write-up xitter.cf/iwasrobbed/statu…
2] Thank you Robin - inspired from Robin Berjon's series of essays:
berjon.com/bigger-browser/
As an ex-Viv (w/ Siri team) eng, let me help ease everyone's future trauma as well with the Fundamentals of Assisted Intelligence.
Make no mistake, OpenAI is building a new kind of computer, beyond just an LLM for a middleware / frontend. Key parts they'll need to pull it off:
Persistent User Preferences:
- The biggest unlock of assistants has always been to deeply understand what someone wants in the most specific way.
- This is the "wow" moment where computers stop being scary and start feeling truly helpful.
- We did this in 2016 on Viv (youtu.be/Rblb3sptgpQ) when our AI knew what you liked for each and every service you used via Viv and mixed that in with context like what kind of flowers you told us your mom liked.
- This will need to include access to your personal information to infer preference as well.
External, Real-time Data:
- 50% of the utility of an LLM comes from the base training and RLHF fine-tuning; but much more comes from extending its available data with external sources.
- Zapier, Airbyte and others will help, but expect deep integration with 3rd party apps / data pipelines.
- "Chat w/ PDF" is a tiny, tiny part of this. If you're only building that, think much bigger.
Actual Computing on a Virtual Machines:
- Context windows are limiting, so AI providers will continue benefiting from running tasks directly on a Python or Node/Deno virtual env so it can consume huge amounts of data just like a computer today can.
- Today these are short-lived envs used by Data Analyst / Julius, but over time they'll become a new type of Dropbox where your data is persisted long term for additional processing or cross-file inference / insights.
Agent Task / Flow Planning:
- Planning can't function without intent. Understanding intent has always been a holy grail, and LLMs finally helped us unlock what we spent years approximating at Viv with NLP tricks.
- Once intent is accurate, planning can start. Creating an agent planner is incredibly nuanced and will take significant integration with user preferences, 3rd party data sets, knowledge of compute capabilities, etc.
- The bulk of the real magic of Viv was the dynamic planner / mixer that would pull all these data and APIs together and generate both a workflow AND dynamic UI on top of them for a normal consumer to execute.
An App Store of Experts:
- Apple initially made the mistake of building a closed app store; then realized they could monetize a cornucopia of creativity if they opened it.
- Regardless of OpenAI saying they're focused on ChatGPT and only ChatGPT, it's inevitable they'll rescope it and enable a long tail of specialized assistants.
- Builders will be able to compose multiple tools together into workflows that can specialize
- And AIs over time will be able to auto-compose these tools together as well, learning from the builders that came before them.
Persistent, Contextual Memory:
- Embeddings are helpful, but they are missing fundamental parts like context switching, conversational centroids, summarization, enrichment, etc.
- Most of the cost of LLMs today comes from prompts, but as history and persistence is embedded and the inference cached, this will unlock the ability to have long term memory with pointers to critical subjects, topics, feelings, tone, etc.
- Core memory is just the beginning. We still need all the rich information our minds conjure when we think about a past sunset, a breakup, a scientific understanding, or sensitive context for people we interact with.
Long Polling Tasks:
- "Agent" is a loaded word, but part of the intent is to have tasks that can be scheduled and self-completing regardless of the time horizon required.
- E.g. "Let me know when flights from Montréal to Hawaii are less than $500"
- This will require coordination of compute across API providers, as well as virtual envs in the cloud.
Dynamic UI:
- Chat is not the final, end-all interface. There's a reason apps have affordances like buttons, date pickers, images. It simplifies, clarifies.
- AI will be a copilot, but to be a copilot it'll need to adjust to what works best for a given user. The future is personalized as optimizations require it, so UI will be dynamic.
API & Tool Composition:
- Expect AIs to generate custom "apps" in the future where we can build our own workflows and compose together APIs, without waiting for a big startup to do so.
- Fewer apps and startups will be needed to generate frontends, and AI will be better at composing an array of tools and APIs together coupled with a gas fee / tax.
Assistant-to-Assistant Interaction:
- There will be countless assistants in the future, with each assisting humans and other assistants towards some greater intent.
- Alongside this, assistants will need to learn to interface across text, APIs, file systems, and other modalities used both by agents / startups and humans as integration flows deeper into our world.
Plugin / Tool Stores:
- Specialized assistants can only be made possible by composing tools, APIs, prompts, data, preferences, and much more.
- The current plugin store is super early days, so expect much more work to come, and expect many of those plugins to be rolled in-house as they become more mission critical.
And this is just a 10 minute brain dump; much, much more is needed behind the scenes including internet search and scraping, community (for intent, building, RLHF, etc), dynamic API generators and connectors, gas fees, tool builders, ingestion via glasses / earbuds / etc. If you think it's too late to be in AI, just know the above is about 25% of what it'll actually take, with much more to come as we iterate and get even more creative.
We're in the early days of building parts of this at @FastlaneAI but with a different understanding: OpenAI will never be the best at everything. So we want to let you use the best AIs in the world, regardless of who builds them (that could be you!). Come join the fun!