This Week in Startups – Episode 2260

How Agents Will Change Banking Forever

Date: March 10, 2026
Host: Jason Calacanis
Co-host: Alex
Guests: Suresh Ramamurthy (NetXD), Rohan Arun (getsupers/PhoneClaw), Eugene Stuckless (EIR Inc.)

Episode Overview

This episode explores the rapid advancements in agentic AI—software agents able to execute complex tasks, self-improve, and transform industries. The discussion focuses on how these agents are poised to fundamentally alter the landscape of banking, entrepreneurship, and productivity. Demos from founders pioneering AI-powered banking and mobile automation highlight real-world progress. The hosts weigh AI optimism against American skepticism, debate job displacement, and offer practical productivity hacks for the agent era.

Key Discussion Points & Insights

1. Auto-Improving AI and the Democratization of Research

[01:05 – 06:52]

• Andrej Karpathy's "Auto Research":

  • Karpathy, prominent former AI head at Tesla and OpenAI, released "Auto Research", a tool for running recursive LLM improvements in five-minute loops.
  • Alex: "You bring your own AI model to be an agent… it tries to improve its own code… over 8 hours managed to find nearly a 1/5 improvement in results." [03:45]
  • Non-experts (e.g., Shopify CEO, Toby Lutke) are experimenting with AI models, showing barriers to entry are dropping.

• Jason: Compares the shift to past democratizations (iPhone apps, database tech)—predicts a leap from 3,000 PhDs to 300,000 AI practitioners as knowledge spreads.

  • “This is the damn cracking from the developers owning the world to everybody building the future. And I'm here for it. It's exciting.” [05:44]

• Implications:

  • Pace of improvement in AI is accelerating (“bullish for startups, for VCs, for everyone who's put money behind a data center.” – Alex [06:17])
  • Open source models and increased tinkering are globalizing AI innovation.

2. Open-Source Agents and Global AI Adoption

[06:54 – 15:42]

• OpenClaw's Viral Spread:

  • OpenClaw (an open-source, agentic AI platform) is embraced especially rapidly in China—grassroots meetups, government support.
  • In the US, however, public opinion on AI is mostly negative: "NBC poll… 26% of people in the US are pro AI and 46% are opposed to it for a minus 20% differential. I was surprised by this." – Alex [07:57]

• Causes of Diverging Attitudes:

  • Jason: Chinese citizens have vivid, positive memories of technological progress; Americans have seen stagnant wages, student debt, and a severed social contract between corporate profits and worker prosperity.
    • "For the first time...if profits are surging, bonuses and headcount will increase...Now profits are surging because we're lowering comp and moving jobs offshore or we're cutting headcount. This is apparent to Americans now." [16:49]
  • Gig economy's rise and potential eclipse by automation increase anxiety about the future.

• Trust and AI's Reputation:

  • Jason draws a stark comparison: “Why is AI as unpopular as a brutal dictatorship … and ICE agents?” [15:51]
  • Criticizes tech industry for not effectively communicating or safeguarding worker interests.

3. AI, Job Loss Fears, and Industry Responsibility

[21:08 – 26:25]

• Alex: Balances optimism for tech progress with concern over displacement: "I've always struggled to match my AI enthusiasm...with the potential for quicker job loss than we can replace with new jobs."
• Jason:
  • “Americans should not trust AI or the AI industry, because until that social contract is fixed, they should assume the worst.” [16:49]
  • "Assume the worst and then be delighted if your job ... becomes a reality. Let's hope that this becomes the Star Trek version." [22:01]
  • Outlines the need for political awareness: "This is not just people saying no data centers in my backyard. This is people potentially voting people into Congress who want to put relatively strict guards ... which could disembowel the industry." – Alex [21:08]

4. Surviving in an Agentic Future: Career and Productivity Advice

[26:27 – 33:58]

• Work Advice:

  • Those in corporate jobs should "be the person who knows how to manage, being a maestro who manages AI." [26:27]
  • For gig workers: “You might need to buckle down and say, you know what, I'm going to learn a trade.”
    • “The robot will build a fence in five years…Then you're going to need to learn how to make the beautiful desk…In 10 years, forget it. It's over anyway. You'll have a robot...” [28:21]

• Productivity Hacks:

  • Jason’s "super distribution": Use assistants and agents to multiply and tailor content distribution, cross-posting, and engagement with minimal human input.
    • “Having a human in the loop with your Athena assistant who can make phone calls. You'll have your agent who is openclaw ... But having a human in aloof loop is going to make you even more productive.” [31:19]
  • Real-world examples: Using assistants to pre-order from exclusive bakeries, coordinate couriers, keep children supplied with the best food.

Demos: Agents Changing Banking and Productivity

1. NetXD: Secure Banking Automation with OpenClaw Agents

[34:14 – 42:25] Guest: Suresh Ramamurthy

• What It Does:

  • NetXD enables AI agents (like OpenClaw) to connect with your bank for reading balances and queuing up transactions, secured by cryptographic authentication that keeps humans in approval loop.
  • Personal optimization rules (buffer management, auto-transfers) executed by agent, requiring human biometric approval for actual money movement.

• Jason: “This is the productivity we need. This is the power of an open source... And the fact that there's an open source agent makes it so easy.”

• Future Directions:

  • Plans to let anyone use NetXD’s open banking API.

2. PhoneClaw: Agentic Automation of Mobile Devices

[42:25 – 47:13] Guest: Rohan Arun

• What It Does:

  • PhoneClaw allows agents to control multiple Android devices, automating multi-step tasks (e.g., posting to various social networks).
  • Demo included AR glasses interface showing available command options on each fingertip for quick navigation.

• Use Cases:

  • Social media management, research, and even large-scale automation (with Jason joking, "Putin and Xi ... need more phones and more social accounts to cause chaos in our elections in 2026!" [44:39]).
  • Potential for full mobile app testing and research—"unlimited access to app stores and apps" [47:15]

3. EIR Inc.: Self-Improving AI Agents for Automated Website Testing

[49:22 – 57:49] Guest: Eugene Stuckless

• What It Does:

  • Automates deep AI-driven testing of websites (SEO, UX, security, performance) using recursive agent workflows.
  • System tracks improvements over time, distributes cost between expensive and cheap tasks, and isolates risk (“blast radius”) from prompt injection or malicious input.

• Innovation:

  • Attribution mechanism so contributors’ knowledge/work powering the AI is credited and even compensated.
  • Eugene: “Allows for you to pay somebody for knowledge work as they transfer it into the AI.” [57:28]

• Vision for Next Steps:

  • Inference efficiency and domain modeling will deliver better, faster agentic work, with less “token burn” for rote tasks.

Memorable Quotes

• Jason: “This is the damn cracking from the developers owning the world to everybody building the future. And I'm here for it.” [05:44]
• Alex: “The pace of improvement in AI this year should be insane. And that's very bullish for startups ... one of the most bullish stories that I've seen in weeks, if not months.” [06:17]
• Jason: “Americans should not trust AI or the AI industry, because until that social contract is fixed, they should assume the worst.” [16:49]
• Jason (NetXD demo): “You have given it optimization rules in memory.... This is something everybody has... [NetXD] queues up the payment, I approve it. That's the productivity we need.” [39:51, 41:04]
• Jason (on AR + agents): “Being able to use AR ... but then actually giving your openclaw, your assistant, its own phone to do things ... kind of interesting, kind of compelling.” [46:46]
• Eugene (EIR Inc.): “It creates attribution. It allows for you to pay somebody for knowledge work as they transfer it into the AI.” [57:28]

Timestamps for Major Segments

| Topic | Timestamp | |-----------------------------------------------------------|--------------------| | Auto-improving AI and Karpathy's "Auto Research" | 01:05 – 06:52 | | OpenClaw adoption — US vs China, public skepticism | 06:54 – 15:42 | | Poll data & American tech pessimism | 14:00 – 16:48 | | AI job loss and broken social contract | 16:48 – 21:08 | | Industry's responsibility to public & regulation risk | 21:08 – 26:25 | | Concrete career advice in agentic era | 26:27 – 33:58 | | NetXD/Banks agentic demo (Suresh) | 34:14 – 42:25 | | PhoneClaw/mobile automation demo (Rohan) | 42:25 – 47:13 | | EIR Inc. agentic QA/testing demo (Eugene) | 49:22 – 57:49 |

Tone & Closing Thoughts

• The episode maintains a fast-paced, witty, and inside-baseball tone, with Jason’s blend of skeptical realism and pragmatic optimism.
• It balances technical detail for insider listeners with broader strategic implications for founders and the public.
• Jason repeatedly emphasizes both the empowerment potential (“the damn cracking”) and the existential disruption AI agents are unleashing.

Takeaways for Non-Listeners

• Agentic AI is crossing into practical, everyday workflows: self-improving models, agent-driven banking, automating device fleets, and intelligent testing.
• The democratization of AI tools is accelerating experimentation and productivity, but public trust in the US remains low compared to China's optimism.
• For individuals: learn to wrangle agents, stay ahead of automation, and invest in skills robots can’t (yet) replace.
• For founders: open-source platforms enable rapid prototyping, competitive advantage, and global reach.
• For society: a new social contract is needed to match the speed and disruptive power of agentic technologies.