TBPN's Run of Show: Clawdbot: AI enters the Napster era
Dario posts new essay, GPT to charge merchants 4% for referrals, Crowdstrike takes Daytona, and more
Happy Monday.
The current thing in tech and business is the WhatsApp-based AI agent Clawdbot.
Let’s get to it.
Today’s lineup
CrowdStrike Founder and CEO George Kurtz at 11:45 AM
Consensys Founder and CEO Joseph Lubin at 12:30 PM
a16z Partner Christian Keil at 1:00 PM
Basis Set Ventures Founder and Managing Partner Lan Xuezhao at 1:30 PM
Synethsia Co-Founder and CEO Victor Riparbelli at 1:40 PM
Arcee AI CTO Lucas Atkins at 1:50 PM
Daily Op-Ed, by John Coogan
Clawdbot: AI enters the Napster era
Clawdbot hype hit a massive inflection point over the weekend, with tons of people either joking about or actually going out and buying Mac Minis so they could run the hottest new open-source AI assistant 24/7. It’s very much an extension of the Claude Code hype train that left the station right around the time everyone went on winter break. LLMs are extremely knowledgeable, they can write code now, they can execute that code now, so they can do tons of useful stuff on a computer. Claude Code was very much locked to a specific computer, as Anthropic has yet to launch a cloud version of it, but even that won’t deliver every possible use case.
People want a fully hybrid desktop and mobile experience. Integration with files and apps on desktop, but accessible from mobile. Clawdbot is this bridge that sits between messaging apps and the AI model of your choosing. Despite the Claude-sounding name, you can easily slot in 10+ different models, and even if you do pick Claude Opus 4.5, you’re still prompted later to add API keys for ChatGPT, Gemini, and Brave to help with things like image generation and web browsing.
It’s all a pretty streamlined experience, you don’t need to have programming experience, but you do have to be happy with sitting in front of a terminal for maybe an hour to even get started. The vast majority of consumers will not jump through the hoops to get Clawdbot installed, but it already has clear product-market fit among developers and lightly technical folks who are willing to jump through some hoops for the promise of a truly universal AI assistant.
In some ways it feels like the GPT-3 launch in 2020, where if you jumped into the OpenAI sandbox and wrote a pretty thorough prompt, you could see glimmers of what was coming with ChatGPT in 2022.
It also feels like the early internet piracy days, where you could fire up Napster or later a torrent site to get an illegal copy of “The.Matrix.1999.720p.HDDVD-Rip.[wArEzCr3W].rar” - there was always the risk you didn’t quite get what you asked for. It was entirely possible you’d get a Russian dub, a different movie entirely, or maybe just a dot exe file begging you to install it and join a botnet.
The technology to send digital media over the internet was basically ready to rock in 1999 when Napster launched. The iTunes Store didn’t launch until 2003 and Netflix started streaming in 2007. The tech worked, but there were hairy business deals to negotiate and plenty of kinks to work out until consumer adoption could really take off.
Despite the Mac Mini memes, Apple Stores do in fact have them in stock. I talked to one Apple Store associate who hadn’t heard of Clawdbot, and when I described it I felt crazy because I was basically describing what Siri and Apple Intelligence should be. At the same time, Clawdbot encourages you to be very careful about security and containment. It’s very easy to imagine spoofed messages triggering prompt injection attacks in a bunch of terrible ways. No big tech company, or frontier AI lab, wants to be responsible for someone getting personal information stolen because of a misconfigured AI assistant, and many of the big tech companies don’t particularly want to work together. Meta can’t get iMessage to show up in their new Meta Ray-Ban Displays. Merging information from all the major consumer internet platforms has never been a technical issue - it’s been a question of walled gardens and business strategy. These negotiations and privacy protections will take years to roll out. Until then, enjoy your hacker superpowers.
Headlines
Noah Epstein: What Is Clawdbot? (And Why People Are Losing Their Minds Over It)
Olivia Moore: Clawdbot is amazing - and, I don’t think consumers should use it
New Dario essay: The Adolescence of Technology
Alex Honnold climbs skyscraper Taipei 101
Dean Ball and Yann LeCun spar over LLMs, AI progress on 𝕏
WSJ: The Wall Street Star Betting His Reputation on Robots and Flying Cars
Rainmaker takes over DC metro station for first OOH ad campaign
WSJ: Hedge Funds Are Back on Top After a Long ‘Alpha Winter’
GPT to charge merchants 4% for GPT checkout transactions
Kyla Scanlon in NYT: The Dangerous Power of Prediction Markets
Racing news: CrowdStrike by APR Takes Long-Awaited Rolex 24 LMP2 Win
Posts of the day
TBPN is brought to you by Ramp.
Ramp is the all-in-one finance automation platform that helps businesses save time and money with smarter corporate cards, spend management, and bill pay.
Special thanks to our sponsors: Shopify, Graphite, ElevenLabs, CrowdStrike, MongoDB, Fin, NYSE, AppLovin, Phantom, Gemini, Cognition, Labelbox, Public, Restream, Vanta, Console, Railway, Figma, Plaid, Okta, Linear, Turbopuffer, Lambda, Gusto, Vibe.co, Sentry, and Cisco.
Regarding the topic of the article, excellent Napster analogy. Very insightful.
Great analysis comparing this moment to Napster - that framing really clicked for me. The "technology works, negotiations pending" parallel is spot on. We're in this weird limbo where the capabilities are clearly there but the ecosystem hasn't figured out the rules yet.
What struck me most about Clawdbot's rise is how it validates something I've been thinking about: the best AI tools right now aren't polished products, they're rough but genuinely useful agents that people cobble together. 44K GitHub stars in weeks doesn't happen because of good marketing - it happens because people are genuinely frustrated with the limitations of official chatbot interfaces.
I've been running my own AI assistant (built on Claude Code) for a few months now, and the security concerns you mention are very real. Prompt injection isn't theoretical - I've had to build multiple layers of input sanitization just for my personal use. The Napster comparison extends here too: early file-sharing was a security nightmare, but the convenience won until better solutions emerged.
The cost angle is interesting too. Clawdbot reportedly runs around $300/day in API costs at scale. I did a deep dive into the architecture and economics after seeing it blow up - the tradeoffs between capability, cost, and control are fascinating when you actually try to build something similar: https://thoughts.jock.pl/p/clawdbot-deep-dive-personal-ai-assistant-2026