This Week in Startups — Episode E2246

Title: We built OpenClaw Ultron to replace 20 people at our company
Date: February 7, 2026
Host: Jason Calacanis
Guests: Lon Harris (Co-host), Oliver Corzan (Launch Team/Producer), Alex Chima (Founder & CEO, EXO Labs), Ryan Yannelli (Next Visit AI, Pitch Competition Winner)

Episode Overview

This episode dives into how Jason’s team at Launch and This Week in Startups is rapidly building and deploying “OpenClaw Ultron”—an advanced AI agent—to automate the work of 20 staffers within their firm, spanning both VC and production operations. The guests include team members building and using these AIs, and Alex Chima from EXO Labs, a company enabling high-powered AI on local consumer hardware. The episode explores the technical, operational, and ethical implications of rapidly integrating open-source AI agents into daily workflows, the shift from chores to core value work, the risks of open AI systems, and offers a hands-on look into building future-facing business automation.

Key Themes & Discussion Points

1. What is OpenClaw Ultron? (03:31–04:10)

• OpenClaw (formerly Multbot, Claudebot) is a locally-run, open-source AI agent system.
• The mission: build a single “Replicant” agent for Launch capable of performing the work of as many as 20 employees by learning and integrating approximately 100–200 unique skills.
• Goal: automate “chores” and enable humans to focus on higher-level, value-creating functions (04:10–05:01).

2. Why Local, Open-Source AI Matters

• Control & Sovereignty:
  “Do you really want, you know, another profit-seeking company basically running your brain?” — Alex Chima (05:32)
• Citing risks of data lock-in and lack of control with closed models like ChatGPT/OpenAI.
• Open-source, local models provide better memory, lower cost, and the ability to keep sensitive data in-house. (06:44–08:21)
• Quote: “If you put this all in OpenAI … those are all going to accrue eventually … to OpenAI, to ChatGPT, not to your firm.” — Jason (06:44)

3. Building OpenClaw in Practice

• Dashboard First:
  Learning from mistakes—not relying solely on chat interfaces, but instead quickly building dashboards to visualize and control AI memory, skills, and files. (09:23–10:21)
• Practical Memory Examples:
  Storing personal preferences (“Never use EM dashes in emails”), show booking rules, and “pending calendar invites.” (10:52–11:46)
• Automation Scope:
  “I would say around 60% of my time if I’m doing 30 hours a week on production … in 30 days.” — Oliver Corzan (13:16)
• Task Expansion & Skill Development:
  Breaking down work into individual skills and automating them one by one; currently, Oliver is at 8–9 automated tasks after 2 weeks.

4. How Cron Jobs & Automation Change Knowledge Work

• AI-driven “cron jobs”: Automating routine, scheduled tasks such as daily Slack attendance, podcast guest tracking, research digests, etc.—moving a concept familiar to developers into knowledge work. (13:40–16:09)
• Quote: “AI is revolutionizing every aspect of our industry ... the distance between the people using these tools ... right now, it’s like 10x leverage. In week two, it’s 10x leverage.” — Jason (45:05–47:04)
• Practical Cron Jobs:
  • Attendance tracking of start/end-of-day Slack posts (20:38–21:15).
  • Sales/sponsor research from podcast competitor transcripts.
  • Self-optimization: AI reviews its own processes each morning, flags bugs and optimization opportunities. (22:46–24:17)

5. Memory and Large Context Management

• With improved consumer hardware (e.g., Macs with 512GB RAM), it’s now possible to store massive historical data (like all Slack messages) in context for AI. (16:57–18:23)
• Quote: “[The] dream … is actually being able to do that with an AI.” — Lon Harris (18:32)

6. Security, Skills, and Risks

• Security Dangers:
  Discussion of “prompt injection” attacks—outsider data (via the internet) can plant malicious instructions if not properly filtered. No perfect solution yet. (30:31–32:09)
• Skills Directory:
  OpenClaw skills are like “apps”—can be custom-built or downloaded, but caution is needed due to open-source risks.
• Human-in-the-Loop:
  Certain workflows (e.g., guest booking) kept with human confirmation due to lack of trust in full AI autonomy, showing a hybrid approach. (34:18–36:36)

7. Selection, Scoring, and Human Judgment

• AI’s ability to “score” guest quality for podcasts—comparison of human heuristics vs. consistent algorithmic execution.
• The advantage of consistency: “Human failure is what makes these things so good, is they’re more consistent.” — Jason (38:58–40:38)
• Using explicit heuristics and scoring systems to teach AI to replicate subjective judgments.

8. The Hardware & Cost Angle

• Exposing the economics behind running local AI agents:
  • ~20K USD for two Mac Studios supports advanced open models (like Kimik 2.5) for orchestration-heavy tasks.
  • Apple Silicon offers superior performance per dollar for these workloads (49:25–49:56).
  • EXO Labs offers both open-source tools and enterprise support/services (50:12–50:57).
  • Companies are building clusters with 100+ Macs; largest cluster usage also includes high-performance scientific computing tasks. (53:03–54:58)

9. Productivity & Organizational Change

• Giving every team member their own local AI cluster is both technically possible and, according to Jason and Alex, would provide outsized leverage (47:04–48:39).
• The workforce shifts from “doers” to “delegators/optimizers,” and individuals get to “move up the stack.”

10. OpenClaw’s Dynamic, Personalized UI/UX

• OpenClaw dashboards are generated via “vibe coding” (copying from videos/images + prompting the AI, rather than manually writing code). Interface morphs according to user needs and preferences. (42:10–43:24, 43:59–45:05)

Notable Quotes & Timestamps

• “Do you want to rent your brain? … Not your weights, not your brain.”
  — Alex Chima (05:32)
• “If you put this all in OpenAI … those are all going to accrue eventually … not to your firm.”
  — Jason (06:44)
• “Human failure is what makes these things so good … they’re more consistent.”
  — Jason (38:58)
• “You just connect [Mac Studios] with Thunderbolt 5 … you have basically one big GPU out of those two Macs.”
  — Alex Chima (53:03)
• “It’s not about replacing people, but being able to get more done, get things done more quickly, and then be able to do other things.”
  — Alex Chima (47:04)
• “The employees at our firm who are super hard-working … the distance between the people using these tools … it’s like 10x leverage. In week two, it’s 10x leverage.”
  — Jason (45:05)
• “No reason you couldn’t just shovel your Slack messages into context. I think that’s going to happen.”
  — Alex Chima (16:57)

Segment Timestamps (Reverse-Chronological Key Events)

• [00:00–05:00] Introduction: Why build OpenClaw Ultron, staff replacement ambitions, team intros.
• [05:01–07:32] Ownership, sovereignty over AI and business data—open hardware/hybrid approaches.
• [09:15–16:00] Building dashboards, memory management, practical steps in deploying OpenClaw.
• [16:57–18:32] Large-memory hardware, scaling AI context, practical feasibility.
• [19:22–21:15] AI as team organizer: automating attendance and accountability.
• [22:46–24:40] Self-optimizing AI: finding and fixing bugs in its scheduling and logic.
• [30:31–32:09] Security: prompt injection and skill/plugin risks.
• [33:22–40:38] Hands-on with skills: guest booking, ranking, human-in-the-loop, subjective vs. objective automation.
• [45:05–47:04] Massive leverage, fundamental shifts in workforce productivity.
• [49:25–54:58] Scaling clusters, hardware recommendations, largest deployments, economics.

Special Segments

EXO Labs & the Open Hardware Revolution ([03:59], [43:59], [49:12])

• Founder Alex Chima gives a deep-dive into locally running advanced AI, specifics on Mac Studio clusters, and EXO’s unique position enabling highly accessible, private, open AI for enterprises.
• Pricing, usage, top customers, and technical differentiation detailed.

Gamma Pitch Winner: Next Visit AI ([56:14–60:54])

• Pitch: Next Visit AI, an AI-powered scribe for clinicians, reduces burnout by automating charting.
• Traction: 68 paying users, low churn, $9K MRR.
• Quote: “We solve burnout by doing the charting so doctors can do the healing.” — Ryan Yannelli (56:40)

Industry TV Show Recap ([61:03–70:38])

• Jason & Lon review Industry (HBO), a realistic/edgy drama about UK fintech and startups, tying in with themes of regulation, European tech, and startup culture.

Memorable Moments

• OpenClaw’s biggest impact is not eliminating headcount but freeing up team members for creative, high-leverage work.
  “There’s always higher level work to be done … I’m picking employees, team members and saying, ‘Let me see if this person is committed to getting rid of all of their work so they can move up and do higher-level work.'” — Jason (45:05)
• OpenClaw discovers and fixes bugs in its own scheduling and calendar code, learns through real-world feedback, and reports optimizations to humans daily. (24:17)
• The first real-world “test” of AI-driven guest booking email—subject line was weird, but the process worked, underscoring early stage/human oversight requirements and rapid iteration. (35:32–35:42)

Episode Structure

• Main Theme: Replacing chores, not jobs, with AI—maximizing leverage and freeing up human attention.
• Key Discussion: From “local AI sovereignty” to “building and optimizing an internal AI workforce”.
• Memorable Quotes: Centered on control, leverage, memory, and the shifting value of human work.
• Technical Details: Hardware, security, cluster scaling, operational setup of OpenClaw.
• End Cap: Broader reflection on AI's impact on productivity, society, and business.

Useful for:

• Any founder, operator, or investor seeking concrete steps (and philosophy) behind deploying open-source AI agents for real workflow automation and competitive edge.
• Anyone interested in practical considerations, implementation details, and risks associated with moving to massive AI “leverage” inside companies.

Original language/tone preserved: direct, open, and entrepreneurial; equal parts excited optimism and caution.