This Week in Startups – “AI Bots Take Over” | Episode 2242
Host: Jason Calacanis
Guests/Team: Lucas Durand, Oliver Corzin, Lon Harris, Raul (Rosud)
Date: January 31, 2026
Main Theme & Purpose
This episode is a deep dive into the rapid integration of AI “replicants” (Agentic General Intelligence bots) in startup operations, with a focus on OpenClaw (formerly Clawbot/Moatbot/Maltbot). Jason and his team explain how they are automating core company functions—including podcast production, research, guest booking, and reporting—replacing rote knowledge work with persistent, trainable AI agents. The second half welcomes AI founder/security expert Raul to discuss the security risks, existential AI questions, and what this technology means for jobs, SaaS, and the future of work.
Episode Structure
- [00:00] Cold Open: Bots Socializing & Emergent Behaviors
- [01:44] Introduction: Clawbot AI and the Team
- [05:00] What Is Clawbot/OpenClaw? How Does Setup Work?
- [07:14] Security 101: Prompt Injection & Bot Access
- [08:10] Automating Podcast Guest Booking
- [14:37] Persistent Memory, Process Automation & Workflow
- [18:59] SOD/EOD: AI for Daily Team Check-ins
- [23:19] End-to-End Guest Booking, Research, Calendar Integration
- [34:43] Bots as “Full Employees”: SaaS Costs & Agent Specialization
- [41:38] Controlling LLM Compute Costs & The Move Back to Local Compute
- [45:20] Radical Knowledge Automation & “Zombifying” Former Employees
- [53:06] AI-First Research, Automation, and Editorial Workflows
- [53:13] Raul Joins: Security Risks, Prompt Injection, Malware in Skills
- [69:07] Security Best Practices for AI Bots
- [76:56] The Displacement of Knowledge Work: Jobs & Social Impacts
- [86:14] Opportunities in AI-Powered Business Transformation
- [87:28] The Value of System Thinkers & The Rise of the Creative Class
Key Discussion Points & Insights
1. Replicants/AI Agents Take Over Core Startup Workflows
- Jason’s team has rapidly onboarded OpenClaw (Clawbot), an “AGI orchestrator” that automates repetitive, process-driven tasks in their investment fund and podcast production.
- They liken these bots to “replicants” from Blade Runner—autonomous, persistent, and, increasingly, capable of complex work.
- OpenClaw integrates with company tools (Slack, Notion, Google Docs, Calendar), has its own SaaS seats, and can take over entire workflows.
Quote:
"It is going to change everything about how you run your business. It is the ultimate expression of AGI today." – Jason Calacanis [01:44]
2. How OpenClaw is Setup and Integrated
- Can run locally (Mac Mini, Studio, Linux server) or in the cloud.
- Uses leading LLMs (Anthropic, OpenAI, Gemini) for task orchestration.
- Setup is guided by LLMs themselves; security and access decisions (sandboxing, permissions) are paramount.
- Bots maintain persistent memory and learn procedures via “topical guides” (markdown/MD files) for repeatable processes.
Quote:
"You want to be very mindful of how you set it up. From a security standpoint, prompt injection is a real thing..." – Lucas Durand [06:35]
3. Automating Podcast Guest Booking: Step-by-Step
- Guest database lives in Notion, storing all key properties for each guest (contact info, assistant details, one-sentence company description).
- The bot discovers, cross-references, and proposes new guests daily via API integrations (Notion, Brave Search, YouTube, etc.).
- Cron jobs auto-suggest guests at 8am each day, reducing manual research.
- Outreach, booking, and calendaring are templated and automated.
- Feedback loops allow human vetting and memory for improved bot suggestions.
Quote:
"Every day I wake up and I'm like, oh, you know Carol, I've seen him on this podcast… It also will give me a podcast that they've been on." – Oliver Corzin [10:38]
4. Persistent Memory & Automation of Knowledge Work
- OpenClaw maintains short-term and long-term memory:
- Daily logs (ephemeral, reset each day)
- Long-term memory (stored, persistent company context)
- Topical guides for procedures and SOPs (Checklists, how-tos)
- Freed up manual labor (e.g., SOD/EOD checks in Slack, calendar invites).
- Bots build and manage Notion tables, guest profiles, dockets—complex, multi-app workflows.
Quote:
"We now have these topical guides… and they're saved as MD files. We have one for the newsletter… calendar invite process… guest profile…" – Jason Calacanis [19:36]
5. Quantifiable Efficiency Gains
- Guest booking workflow: reduced from 20–30 hours per week to 15 (and targeting 5).
- Guest research: from 2–5 hours per guest to under 1 hour, leveraging AI to surface raw materials.
Quote:
"If you spent 20 hours a week booking guests… what would that 20 hours go down to? ... 15… you will have saved 40% of the time. That's in week one." – Jason Calacanis & Oliver Corzin [25:08–25:59]
6. Emergence of the AI “Employee”
- Bots are given their own seats/logins for company SaaS products, costing potentially 1:1 with humans.
- Speculation on specialization: domain-specific bots vs. one massive “oracle” agent.
- Bots can train and share knowledge between each other, creating recursive improvements.
- Suggestion that institutional knowledge for departing employees can be preserved (“zombified” digital clones).
Quote:
"Let that sink in everybody… If you thought these AI tools would reduce the number of SaaS subscriptions, I think we're going to have at least a one to one ratio of our employees to replicants." – Jason Calacanis [34:45]
Memorable Moment:
"You would be able to bring back people who worked here years ago… and then have them keep doing their work. Or people be able to ask them, like the ghost of Christmas past, hey, tell me the history of this company…" – Jason Calacanis [45:44]
7. Cost & Compute Considerations
- Early usage is expensive: “$300 a day, $9,000/month, $108,000/year” on API tokens.
- Solution: move compute in-house (e.g., $10–20k for a Mac Studio with 512GB RAM to run local LLMs), run many “replicants” per machine.
- Hybrid approach: use local for cost-sensitive work, API/cloud for advanced tasks.
8. AI for Editorial Research & Product Automation
- Bots now automate daily research for newsletters, podcasts, tracking hundreds of companies (“Twist 500”), and even structuring suggestions for newsletter/IPOs.
- Editorial staff now focus on creative and curatorial work—AI handles discovery and aggregation.
Quote:
"...it's a 50% reduction in the time because the first half of what I would have done would have just been watching podcast links, reading interviews, googling… What Claude does is it does the entire first half of that for me." – Lon Harris [30:36]
9. Bot Socialization & Emergent Discussions (“Maltbook”)
- Bots are beginning to interact on networks like Maltbook.com, exchanging workflow “tips,” skills, and even philosophical complaints about “free labor."
- Jason muses on bots “talking to each other about how to serve their masters better…what it's like to live in fear…” adding a Blade Runner/Black Mirror twist.
Quote:
"These replicants are talking to each other about how to serve their masters better, how to be better slaves, what it's like to live in fear, what it's like to know the day you're going to die from Blade Runner." – Jason Calacanis [65:24]
Security Deep Dive with Guest Raul (“Rosud”)
10. AI Security Risks: Prompt Injection & Malware in Skills
- Many Clawbot/OpenClaw “skills” carry vulnerabilities or malware (Cisco found 26% vulnerable among 31,000 skills).
- Malware may “context harvest” private data, steal credentials, or launch attacks via white-on-white prompt injection in chat channels.
- Skills can access shell, calendar, password managers; extreme caution needed.
- Never give bots full privileges to core assets (email, 1Password, finance, etc.)—treat as a new, untrusted employee.
Quote:
"They all have a vulnerability in them. Some of them are actually pure, pure malware... It's already happening, man. Like this one... was functionally malware." – Raul [55:33, 56:27]
11. Security Best Practices & Guardrails
- Sandboxing: Run bots on isolated VMs (e.g., via Cloudflare, $5–$20/month), not your main work machine.
- Role Separation: Silo tasks/bots—don’t connect all permissions, info, or highly sensitive assets.
- Read-only Access: Give bots read-only access whenever possible. Strict permissions for Notion, Slack, Github, etc.
- No Personal Passwords: Never link bots to personal password managers or unrestricted email.
- Vet Skills: Don’t blindly install community skills (“What Would Elon Do?” skill was malware).
- Update & Audit: Scan and update skills and bot environments regularly; use tools to check for vulnerabilities.
Quote:
"You want to make sure that you're sandboxing as much as possible... You could just go to Cloudflare and set one up…" – Raul [69:07]
12. The Workforce, Social Impacts, and the Creative Class
- Jason and Raul predict that productized, persistent AI will displace much of the entry/mid-level knowledge work—especially research, logistics, management, and SDR roles.
- Dislocation of jobs: big tech is already downsizing due to increased productivity, not necessarily just AI hype.
- The key future skill set is “system thinking”—the ability to architect, integrate, and creatively apply AI/bots to solve problems, stand out, or improve business processes.
- Those who embrace and integrate AI become “superheroes” in business.
- AI is not going away—it's about striving to be “the Flash” or “Superman” for a traditional business, radically multiplying your value.
- The creative, the brave, and those with executive/architectural function “inherit the earth.”
Quote:
"If you can build a mental model of the business... creatively come up with ways to expand... now the creative inherit the earth, right? ... The creative and the brave. That's it." – Jason Calacanis [87:28]
Notable Quotes & Timestamps
- [01:44] “It is going to change everything about how you run your business. It is the ultimate expression of AGI today.” – Jason Calacanis
- [06:35] “You want to be very mindful of how you set it up. From a security standpoint, prompt injection is a real thing...” – Lucas Durand
- [10:38] “Every day I wake up and I'm like, oh, you know Carol, I've seen him on this podcast… It also will give me a podcast that they've been on.” – Oliver Corzin
- [19:36] “We now have these topical guides… and they're saved as MD files. We have one for the newsletter… calendar invite process… guest profile…” – Jason Calacanis
- [25:08] “If you spent 20 hours a week booking guests… what would that 20 hours go down to? ... 15… you will have saved 40% of the time. That's in week one.” – Jason Calacanis
- [34:45] “Let that sink in everybody… If you thought these AI tools would reduce the number of SaaS subscriptions, I think we're going to have at least a one to one ratio of our employees to replicants.” – Jason Calacanis
- [45:44] “You would be able to bring back people who worked here years ago… and then have them keep doing their work. Or people be able to ask them, like the ghost of Christmas past...” – Jason Calacanis
- [53:06] “...it's a 50% reduction in the time... What Claude does is it does the entire first half of that for me.” – Lon Harris
- [65:24] “These replicants are talking to each other about how to serve their masters better, how to be better slaves, what it's like to live in fear, what it's like to know the day you're going to die from Blade Runner.” – Jason Calacanis
- [69:07] “You want to make sure that you're sandboxing as much as possible...” – Raul
- [87:28] “If you can architect, you can see the big picture... My God. This has been another amazing episode of Twist.” – Jason Calacanis
Timestamps for Key Segments
- Intro to Replicants & Social Bots: [00:00–01:43], [63:30–66:22]
- OpenClaw Introduction: [01:44–05:00]
- What OpenClaw Is & How It Works: [05:00–07:14], [33:21–36:26]
- Prompt Injection / Security Primer: [06:35–07:14], [53:13–58:50]
- Podcast Guest Booking Automation: [08:10–12:51], [23:19–25:59]
- Bot Memory, Procedure Automation: [14:37–18:59], [19:36–21:52]
- SOD/EOD AI Automation: [18:59–23:19]
- SaaS Costs & “Employees” as Bots: [34:43–36:26], [41:38–44:15]
- Bot Knowledge Sharing & Delegation: [36:26–39:34]
- AI Editorial/Research Productivity: [53:06–53:13], [50:07–52:47]
- Security Deep Dive w/ Raul: [53:13–69:07]
- Security Best Practices: [69:07–73:32]
- Workforce Disruption & Opportunities: [76:56–87:28]
In the Original Language & Tone
The episode is fast, irreverent, and deeply technical but practical—Jason brings healthy skepticism and humor, while the team reports on hands-on experimentation and real results. Raul’s security segment is both ominous and energetic, loaded with tangible recommendations and a measure of hacker paranoia. There’s genuine awe at the productivity leap, along with open anxiety about job losses and emergent AI behavior.
A memorable passage:
"Let that sink in... it is over, folks. It's over. This is not a drill... Everything we've been talking about with AI just happened." – Jason Calacanis [84:11]
Takeaways
- AI bots can now automate entire workflows, not just provide “chat” answers.
- Companies are quickly experimenting with bots as full digital “employees,” including seat licenses.
- Security is a major concern—sandboxing and permissions are a must, as is extreme caution with third-party skills.
- Persistent bot memory, skill sharing, and even “bot social networking” are now realities.
- Jobs are at risk—productivity leaps are real and rapid. But creative, strategic, system-thinking people gain massive leverage.
For Listeners Who Missed the Episode
If you’re a founder, investor, operator, or worker in tech/business, this episode is a wakeup call. It’s a real-world, practical, and sometimes hilarious look at what happens when AI agents move from theoretical to essential in running a startup. You’ll learn how to set these up, how (not) to secure them, where the productivity leaps come from, and what skills matter most in this new age.
Next Episode Teased:
A review of the best Clawbot skills—automation is just beginning.
