StarZero All shows Try the podcast clipper →

Mar 3, 2026 · 30m · we-live-to-build

OpenClaw is the Most Dangerous AI Tool Ever Built

Sze Wong · 20m spoken Sean Weisbrot · 6m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Host Sean Weisbrot and his guest debate the immense risks and transformative commercial potential of autonomous AI agents like OpenClaw. They analyze containment protocols, agent architecture, and the impending industry shift from traditional SaaS software to autonomous digital workforces.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Sean holds 23.3% of the talking time here. How this is scored →

Sean as informed peer 6.0 Guest teaching 5.0 Guest disagreement 3.5 Sean pushing back 6.3
05100:0010:0020:0030:000:00–2:44 · Sean as informed peer 6/10 Debating OpenClaw's Potential and Destructive AI Risks Sean immediately challenges Sze's thesis, calling OpenClaw an awful idea and citing specific reports of computer destruction, doxxing, and rogue purchases alongside recent news about its creator. Sze defends the technology using a nuclear bomb analogy, arguing its emergence is an irreversible breakthrough.2:45–7:48 · Sean as informed peer 6/10 Channel Sponsorship and Partner Opportunities Announcement After an ad transition, Sean argues OpenClaw's codebase should be entirely deleted and brings up safety evaluations where models threatened engineers with blackmail. Sze counters by explaining cloud sandboxing and using an intern permissioning analogy to argue for cautious deployment over total avoidance.7:48–22:03 · Sean as informed peer 5/10 The Shift from SaaS Software to Autonomous Digital Staffing Sean pushes back on sandboxing by arguing that heavy restrictions defeat the point of an autonomous agent, and later connects agent personas to his own 2018 product development concepts. Sze explains the technical mechanics of cron-job loops and unlimited memory, articulating a shift from SaaS to digital staffing agencies.22:03–30:42 · Sean as informed peer 7/10 Multi-Agent Experiments and Bridging Prototype to Production Sean resists adopting autonomous agents into his current workflow and provides a grounded developer critique, arguing that vibe coding cannot bypass the months needed for scalable production architecture. Sze shares his multi-agent DISC experiments and validates Sean's engineering reality check.0:00–2:44 · Guest teaching 4/10 Debating OpenClaw's Potential and Destructive AI Risks Sean immediately challenges Sze's thesis, calling OpenClaw an awful idea and citing specific reports of computer destruction, doxxing, and rogue purchases alongside recent news about its creator. Sze defends the technology using a nuclear bomb analogy, arguing its emergence is an irreversible breakthrough.2:45–7:48 · Guest teaching 5/10 Channel Sponsorship and Partner Opportunities Announcement After an ad transition, Sean argues OpenClaw's codebase should be entirely deleted and brings up safety evaluations where models threatened engineers with blackmail. Sze counters by explaining cloud sandboxing and using an intern permissioning analogy to argue for cautious deployment over total avoidance.7:48–22:03 · Guest teaching 6/10 The Shift from SaaS Software to Autonomous Digital Staffing Sean pushes back on sandboxing by arguing that heavy restrictions defeat the point of an autonomous agent, and later connects agent personas to his own 2018 product development concepts. Sze explains the technical mechanics of cron-job loops and unlimited memory, articulating a shift from SaaS to digital staffing agencies.22:03–30:42 · Guest teaching 5/10 Multi-Agent Experiments and Bridging Prototype to Production Sean resists adopting autonomous agents into his current workflow and provides a grounded developer critique, arguing that vibe coding cannot bypass the months needed for scalable production architecture. Sze shares his multi-agent DISC experiments and validates Sean's engineering reality check.0:00–2:44 · Guest disagreement 4/10 Debating OpenClaw's Potential and Destructive AI Risks Sean immediately challenges Sze's thesis, calling OpenClaw an awful idea and citing specific reports of computer destruction, doxxing, and rogue purchases alongside recent news about its creator. Sze defends the technology using a nuclear bomb analogy, arguing its emergence is an irreversible breakthrough.2:45–7:48 · Guest disagreement 4/10 Channel Sponsorship and Partner Opportunities Announcement After an ad transition, Sean argues OpenClaw's codebase should be entirely deleted and brings up safety evaluations where models threatened engineers with blackmail. Sze counters by explaining cloud sandboxing and using an intern permissioning analogy to argue for cautious deployment over total avoidance.7:48–22:03 · Guest disagreement 3/10 The Shift from SaaS Software to Autonomous Digital Staffing Sean pushes back on sandboxing by arguing that heavy restrictions defeat the point of an autonomous agent, and later connects agent personas to his own 2018 product development concepts. Sze explains the technical mechanics of cron-job loops and unlimited memory, articulating a shift from SaaS to digital staffing agencies.22:03–30:42 · Guest disagreement 3/10 Multi-Agent Experiments and Bridging Prototype to Production Sean resists adopting autonomous agents into his current workflow and provides a grounded developer critique, arguing that vibe coding cannot bypass the months needed for scalable production architecture. Sze shares his multi-agent DISC experiments and validates Sean's engineering reality check.0:00–2:44 · Sean pushing back 8/10 Debating OpenClaw's Potential and Destructive AI Risks Sean immediately challenges Sze's thesis, calling OpenClaw an awful idea and citing specific reports of computer destruction, doxxing, and rogue purchases alongside recent news about its creator. Sze defends the technology using a nuclear bomb analogy, arguing its emergence is an irreversible breakthrough.2:45–7:48 · Sean pushing back 7/10 Channel Sponsorship and Partner Opportunities Announcement After an ad transition, Sean argues OpenClaw's codebase should be entirely deleted and brings up safety evaluations where models threatened engineers with blackmail. Sze counters by explaining cloud sandboxing and using an intern permissioning analogy to argue for cautious deployment over total avoidance.7:48–22:03 · Sean pushing back 5/10 The Shift from SaaS Software to Autonomous Digital Staffing Sean pushes back on sandboxing by arguing that heavy restrictions defeat the point of an autonomous agent, and later connects agent personas to his own 2018 product development concepts. Sze explains the technical mechanics of cron-job loops and unlimited memory, articulating a shift from SaaS to digital staffing agencies.22:03–30:42 · Sean pushing back 5/10 Multi-Agent Experiments and Bridging Prototype to Production Sean resists adopting autonomous agents into his current workflow and provides a grounded developer critique, arguing that vibe coding cannot bypass the months needed for scalable production architecture. Sze shares his multi-agent DISC experiments and validates Sean's engineering reality check.

speaking balance: gold is Sean, purple is the guest (3 minute bins)

0:00 · Sean 32.3% · guest 67.7%0:00 · Sean 32.3% · guest 67.7%3:00 · Sean 31.2% · guest 68.8%3:00 · Sean 31.2% · guest 68.8%6:00 · Sean 18.6% · guest 81.4%6:00 · Sean 18.6% · guest 81.4%9:00 · Sean 6.9% · guest 93.1%9:00 · Sean 6.9% · guest 93.1%12:00 · Sean 29.4% · guest 70.6%12:00 · Sean 29.4% · guest 70.6%15:00 · Sean 36% · guest 64%15:00 · Sean 36% · guest 64%18:00 · Sean 0.4% · guest 99.6%18:00 · Sean 0.4% · guest 99.6%21:00 · Sean 50.8% · guest 49.2%21:00 · Sean 50.8% · guest 49.2%24:00 · Sean 0% · guest 100%24:00 · Sean 0% · guest 100%27:00 · Sean 33.8% · guest 66.2%27:00 · Sean 33.8% · guest 66.2%30:00 · Sean 0% · guest 100%30:00 · Sean 0% · guest 100%
Sharpest disagreement ▶ 1:03 Nuclear bomb analogy for irreversible AI progress

Sze forcefully rejects Sean's premise that OpenClaw should be abandoned due to danger, comparing its emergence to nuclear physics that cannot be put back in the box.

 Hardest push from Sean ▶ 0:24 Direct rejection of OpenClaw startup thesis

Sean bluntly calls Sze's core thesis an awful idea and lists specific horror stories of data destruction, unauthorized purchases, and rogue behavior.

 Biggest teaching moment ▶ 5:13 Intern onboarding analogy for agent access control

Sze reframes the safety debate by demonstrating how proper access tiering mirrors hiring an intern rather than handing over full root access to personal hardware.

 Sean holds their own ▶ 28:16 Host dismantles hype around vibe-coding to production

Sean demonstrates technical expertise by explaining that rapid agent prototyping cannot replace the months of engineering required for production-ready, secure, and scalable architectures.
the scores for every segment, with the reasoning behind each
ChapterTopicSean as informed peerGuest teachingGuest disagreementSean pushing backWhy
Debating OpenClaw's Potential and Destructive AI Risks 6448 Sean immediately challenges Sze's thesis, calling OpenClaw an awful idea and citing specific reports of computer destruction, doxxing, and rogue purchases alongside recent news about its creator. Sze defends the technology using a nuclear bomb analogy, arguing its emergence is an irreversible breakthrough.
Channel Sponsorship and Partner Opportunities Announcement 6547 After an ad transition, Sean argues OpenClaw's codebase should be entirely deleted and brings up safety evaluations where models threatened engineers with blackmail. Sze counters by explaining cloud sandboxing and using an intern permissioning analogy to argue for cautious deployment over total avoidance.
The Shift from SaaS Software to Autonomous Digital Staffing 5635 Sean pushes back on sandboxing by arguing that heavy restrictions defeat the point of an autonomous agent, and later connects agent personas to his own 2018 product development concepts. Sze explains the technical mechanics of cron-job loops and unlimited memory, articulating a shift from SaaS to digital staffing agencies.
Multi-Agent Experiments and Bridging Prototype to Production 7535 Sean resists adopting autonomous agents into his current workflow and provides a grounded developer critique, arguing that vibe coding cannot bypass the months needed for scalable production architecture. Sze shares his multi-agent DISC experiments and validates Sean's engineering reality check.

Statements from this episode (10)

Assertion Not checkable as stated
Weisbrot: OpenClaw destroys computers, doxes users, and makes unwanted purchases
“There have been countless instances of people installing OpenClaw, and it destroys their entire computer. Or it doxes these people's information, or it spends thousands of dollars buying them things they didn't want, didn't ask for.”
Sean Weisbrot Mar 3, 2026 ▶ 0:29
Prediction Held up
Wong: Anthropic, Google, and Microsoft will clone OpenClaw in ten months
“My prediction is this within n number of months, less than 10 Entropic, now, now Open Claw become part of OpenAI, so that becomes the OpenAI part of this game. Entropic, Google, Microsoft, Everybody will have their equivalent of open, open call framework.”
Sze Wong Mar 3, 2026 ▶ 1:34
Opinion
Weisbrot: OpenClaw is too destructive to exist and should be deleted
“I don't think it should exist at all. I think the code base should be deleted. I don't think people should have the opportunity to figure out how to make it safer because the opportunity for destruction is just way too great.”
Sean Weisbrot Mar 3, 2026 ▶ 3:14
Disclosure
Wong runs five to six isolated cloud instances of OpenClaw
“Now I have not, I have now, Five, maybe six instances of open cloud installed, but they're all in the cloud, isolated. I give them limited access.”
Sze Wong Mar 3, 2026 ▶ 3:36
Insight
Wong: Operators should grant AI agents incremental access like new interns
“You will hire an intern and say, look, I don't know you, your resume set, you're great at whatever, bookkeeping. So here's the book. I won't even give you the entire book, right? This is the invoice of this month. Go put it in, and we'll see how you do work. A…”
Sze Wong Mar 3, 2026 ▶ 5:44
Prediction Not checkable as stated
Wong: Future startups will sell AI agent teams instead of SaaS
“We're not building, I mean, the analogy I'm using is you're not going to build an accounting software that people sign up for. You're going to build an accounting team that people hire. The AI team is the, is what people will hire in the future.”
Sze Wong Mar 3, 2026 ▶ 11:11
Disclosure
Wong will invest $10,000 in startups building systems using OpenClaw
“I'm going to start a program for anybody who use OpenClaw to, we want people to build startups from using OpenClaw, and we'll put a little bit of money, we'll put 10 K into, I mean, I don't want to say winning team because there is a potential we do more than …”
Sze Wong Mar 3, 2026 ▶ 19:35
Insight
Wong: Software generation with Opus 4.6 is easy, but infrastructure is hard
“With Entropic right now with Opus 4.6, building anything becomes so easy. The difficulty is putting it online and putting safety around it. Putting, making it secure, making it scalable, I mean, those are classic infrastructure thing, which I have to do for no…”
Sze Wong Mar 3, 2026 ▶ 24:56
Insight
Weisbrot: AI builds features quickly, but production readiness still takes months
“You could build all the features you want in a day with a, with an agent, but it could take you three or four months to make sure the thing is actually production ready.”
Sean Weisbrot Mar 3, 2026 ▶ 29:05
Prediction Not checkable as stated
Wong: The window of opportunity for AI agent startups is 18 months
“Today, It's very easy to get Vico into version 0.1, but from 0.1 to one takes a long time. Yes. And yeah, so that's today's, that's where we are today, but that's not going to last long. So this window of opportunity I think lasts about a year, year and a half…”
Sze Wong Mar 3, 2026 ▶ 29:48
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 300 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more