founder

San Francisco, CA
Joined January 2015
the single most important concept in 2026
28
156
5
2,953
819,587
Ben Smith retweeted
Dan Robinson
@danrobinson25 Nov 2025
Are hackathons just like 45 minutes now, or what
16
2
52
10,340
Ben Smith retweeted
Most web3 agents use a predefined set of hardcoded tools or function calls. For instance: - Get my price chart - Run security check - Execute token swap While these predefined functions simplify interactions, they severely limit flexibility and scalability, restricting users to a narrow set of tasks. In comes Model Context Protocol (MCP)... Instead of rigidly defined toolsets, MCP enables an autonomous agent to dynamically: - Discover new tools or functionalities not previously available or hardcoded. - Subscribe to these tools on-demand based on the agent’s needs or user requests. - Integrate new capabilities seamlessly into its workflow. This means your application can dynamically expand its abilities beyond initial coded limitations ( @AgentTankLive). Example Scenario: User makes a complex request to an AI agent: "Get a price chart of my top token, run security checks, analyze market sentiment, then swap tokens with the highest potential." Without MCP: - Limited to predefined functionalities (e.g., charting, swapping). - Any request outside of predefined tools (e.g., sentiment analysis) is impossible, leaving the request unfulfilled. With MCP: - The agent recognizes missing functionalities (e.g., sentiment analysis). - Dynamically queries a registry or marketplace (like @arcdotfun rhizome) to identify and integrate a new "Market Sentiment Analysis" tool. -Agent subscribes to the newly discovered capability autonomously. - The request is fulfilled seamlessly. --- MCP fundamentally transforms static AI toolsets into dynamic, self-adapting ecosystems, bridging the gap to true autonomy. We'll be seeing this in action with Tank by @AgentTankLive
17
31
1
102
4,021
Ben Smith retweeted
Solid Foundation: My Personal Assistant $Tank @AgentTankLive @bcsmithx
6
18
1
58
3,339
THE FUTURE OF AI ISN’T CHAT… IT’s ACTION Most of AI today can “talk” by writing back to you. But what if AI could see your screen, hear your voice, and use your mouse? Meet $TANK @AgentTankLive - your Crypto co-pilot turned full blown digital teammate. Their beta drops April 30th Here’s why it finally turns crypto complexity into simplicity 👇🧵
23
28
2
99
14,080
Ben Smith retweeted
One other sector that's likely to gain more popularity is Computer Use Agents At the core of the AI Agent narrative, we all want agents to perform & automate repetitive tasks for us—acting as an efficient, proactive digital worker that makes us more productive. But there are limitations in current systems that don’t allow AI to easily enable things we want to do on the internet (API limitations). In Web2, one startup breaks free of this limitation — @browser_use (YC W25) raised $17M off the back of its open-source product that allows AI agents to browse websites and perform tasks just like a human would. Browser Use essentially breaks down the buttons and elements of a website into a more digestible, “text-like” format for agents, which helps them interact & navigate through websites better. (Interestingly, the founders launched Browser Use last year through ETH Zurich’s Student Project House accelerator). Browser Use has become a widely used product as seen by its ~50k GitHub stars, 5k forks, and the use of Browser Use by ManusAI, which brought a lot more awareness to the product. Example things you can do w/ Browser Use: • Add grocery items to cart, and checkout on a Supermarket site • Add my latest LinkedIn follower to my leads in Salesforce. • Read my CV & find ML jobs, save them to a file, and then start applying for them in new tabs—if you need help, ask me. • Write a letter in Google Docs to my Papa, thanking him for everything, and save the document as a PDF. • Look up models with a license of cc-by-sa-4.0 and sort by most likes on Hugging Face, save top 5 to file. The focus is geared towards developer-first as the tech remains quite nascent (similar to other AI products out there) In Web3, we have a few teams building Computer Use agents, with the most notable one being @AgentTankLive that made headlines a few months ago—showcasing demos of agents scrolling through X while narrating w/ their voice, researching through social media, breaking down charts, etc. • $TANK retraced 90%+ from ATH but the project has seen constant feature updates & integrations • Recently became a part of @arcdotfun eco, acting as one of the key infra for Arc Ryzome App Store vision • Tank beta coming soon for macOS Given Agent Tank’s focus on ease-of-use for the non-technical Web3 crowd, it'd be interesting to see if the beta could live up to its initial hype. Computer Use agents could be one of the most practical use cases to bridge the AI x Web3 experience gap—especially as dev tools evolve into smoother user flows.
Top 3 AI Agents to Keep Your Eyes On ↓ (Bonus picks at the end)
61
45
4
189
23,290
Ben Smith retweeted
/arc_angels Yesterday we welcomed @FabelisAI and @AgentTankLive as the newest projects joining the Arc Handshake—bringing us closer to an interconnected, composable ecosystem of AI services. We also hosted a Twitter Spaces to dive into what this means for Arc. It was a lot of fun—thanks to everyone who joined! For those who couldn't make it, here's a quick recap: At the core of our conversation was Ryzome, the universal "app store" for AI agents, powered by Anthropic’s Model Context Protocol (MCP). Think of MCP as the "HTTP" or "USB-C" for AI—it enables seamless connections between AI agents and countless services, from traditional Web2 platforms to innovative Web3 apps. (For more, check out our recently released arc lite paper.) We discussed how the two newest arc handshake members embody this vision: - Fabelis creates open-source Rust SDKs that let developers easily build and integrate powerful services into AI agents. Instead of building integrations from scratch every time, devs can leverage Fabelis to instantly connect agents to platforms like Discord or Shopify—transforming simple chatbots into genuinely useful assistants. - Agent Tank is redefining AI interfaces. Their agents aren’t confined to chat—they see your screen, respond to your voice, and interact smoothly with multiple apps. Imagine a Jarvis-like companion actively managing your digital world. What's especially exciting is how Agent Tank can leverage Fabelis’s connectors, demonstrating the true power of composability. Check out this awesome video demo of Agent Tank and Fabelis in action! Together, these projects—alongside Listen, Soulgraph, Askjimmy, and others in the Arc handshake program—embody the heart of what we're building: AI systems that integrates, composes, and does real-world tasks. We’re seeing the Arc ecosystem come alive, moving beyond AI that merely chats toward AI that actively accomplishes meaningful work. This is just the start—and it's already incredibly exciting. we build rig, rig builds us
This Post is from an account that no longer exists. Learn more
22
45
3
195
24,844
Ben Smith retweeted
the distinction between commanding AI and conversing with it begins to blur
This Post is from an account that no longer exists. Learn more
23
28
199
14,627
Silver lining the meme extraction + bear market vibes has *really* forced teams to think about what they're building Entertainment agents are dead. Theoretical Trading agents? dead. Time to ship in production. Let the market decide if your product is good or not. We're starting to see the emergence of these at the fringes: @AskBillyBets trading a live $100K account balance & posting wins/losses (also cool that it leverages a Bittensor subnet) @AgentTankLive launching their desktop co-pilot that can do multiple actions at once. Think @AcolytAI too. bunch of other teams about to expand from the "demo phase" @griffaindotcom @HeyAnonai & load more. I'm also excited to see the leaders set the stage for the agent product/services layer: @arcdotfun with @askjimmy_ai @listenonsol creating autonomous trading. @ai16zdao creating a social trading layer. @virtuals_io with their agent commerce protocol. The time to larp is up, ship or get left behind.
76
56
10
322
29,283
Ben Smith retweeted
0xSammy
@0xSammy17 Feb 2025
Agent tank releases its voice AI integration solution The agentic co-pilot is evolving fast /12 xitter.cf/AgentTankLive/status/1…
9
25
94
6,404
Join us tomorrow at 2:00 PM PST for a conversation with the @AgentTankLive team! ⬇️ xitter.cf/i/spaces/1kvJpyPlZDOxE

21
42
3
105
23,410
Ben Smith retweeted
Proof: @AgentTankLive did a cool demo of this. Also expect other agent engines to follow suit eg griffain, limitus
This Post is from an account that no longer exists. Learn more
4
11
66
6,509
Ben Smith retweeted
Today’s agents: Robotic, narrow-focus, clunky, non-expressive. Tomorrows agents: Human-like, multi-expert (e.g. KOL, DeFAI expert), adaptable, personable. The latter will come in the form of a personal co-pilot. It will live on your phone & computer spending every second tailoring itself to your needs. Major companies like OpenAI, Perplexity will normalise this to the masses (their initial agent demo’s tease this) But several crypto teams have led the way here too (eg @griffaindotcom @HeyAnonai @AgentTankLive ) I suspect the impact of copilot agents to be vastly greater in crypto given: - crazy TAM (endless users) in need of better UX navigating crypto. - open source global markets are a much more expressive, adaptable medium vs. centralised companies (silos).
50
39
10
211
21,339
Ben Smith retweeted
It’s the weekend, onchain coins are decimated but I still see certain teams heads down grinding on a Saturday. Spotted @bcsmithx building a new feature for @AgentTankLive today I know the @virtuals_io team is non-stop prepping for the Solana launch & I saw the capital allocators for their new prototype-2-sentience fund are busy finding driven teams ( @VaderResearch @gkisokay I see you) @ai16zdao just shipped the final ELIZA version before V2 which will be a game-changer The teams that lock-in & build through the hard times are the ones you want to watch. Who else am I missing?
94
27
8
245
33,425
Ben Smith retweeted
Wait, it just occurred to me that voice-enabled agents will be 100% more addictive to engage with. Think about it, it’s less friction (talk vs. type) & the agent sounds like a human (so you subconsciously feel at ease) Proof: I’ve gone from typing to ChatGPT to Having conversations with it (voice-mode) Result: I’m giving WAY more detail to the AI model AND speaking to it for longer. Which is giving it much better data on me as an individual + it gets my Voice data. I’m guessing we see agents adopt voice capabilities soon —> more engagement + utility. I’ve no idea how this would work yet beyond just a chat feature But I think 2 projects working on this are @soulgra_ph & @AgentTankLive Who else am I missing?
150
47
13
392
47,934
Ben Smith retweeted
Replying to @bcsmithx
tank gang
17
16
2
55
4,599
@AntiRugAgent is proud to partner with @AgentTankLive! SarumAI will become the main safety agent on Agent Tank Marketplace. Our specialized agent will be included with their TankTerminal desktop app release. More Information coming soon... #ANTIRUG x $TANK
46
121
11
381
36,852
Ben Smith retweeted
🚨Big News Neurobros! Neurobro has officially partnered with @AgentTankLive - a leading AI project focused on computer-use AI infrastructure This partnership will integrate Neurobro's Nevron agentic framework into the AgentTank ecosystem - providing a solid foundation for advanced agentic intelligence Neurobro will also be part of their ecosystem in an innovative way! Stay tuned - more details about this partnership will be shared early tomorrow🤝
48
86
9
304
31,509
Ben Smith retweeted
Replying to @AgentTankLive
TANK MODE
13
19
5
82
10,736
$TANK framework probably leads the next leg of AI agents. We’ve gotten agents with X functionality, Discord and Telegram access. Voice Customization + Computer Access will 100x the virality/use cases of all agents.
17
43
7
210
42,632