Big Technology Podcast: OpenAI vs. Anthropic's Direct Faceoff + Future of Agents — With Aaron Levie

Host: Alex Kantrowitz
Guest: Aaron Levie (CEO, Box)
Release Date: April 8, 2026

Episode Overview

This episode dives deeply into the state of the competition between OpenAI and Anthropic, two leading AI labs, focusing on their converging product strategies, the maturing landscape of AI agents, and what the future holds for applied AI at work and beyond. Alex Kantrowitz sits down with Box CEO Aaron Levie to analyze which company is best positioned for the next era of AI, examine the real challenges for AI agent adoption, and debate who will capture the most value: the foundational labs or those building on top.

Key Discussion Points & Insights

1. OpenAI vs. Anthropic: Now Building the Same Product?

• Early Divergence, Eventual Convergence:

  • Anthropic initially led in enterprise and coding use cases, while OpenAI focused on consumer AI with ChatGPT.
  • Now, both offer similar “super app” visions—AI assistants that do much more than chat, competing head-to-head for the full scope of use cases.
  • “If you have this AI model that is super intelligence packed into a model, it eventually has to converge on all of this. All the same use cases will be represented by that.” — Aaron Levie [01:35]

• Enterprise vs. Consumer Origins:

  • Both have made inroads into the other’s strongholds—enterprise buying ChatGPT for internal use, while Anthropic gained ground in enterprise tools and coding.
  • “Both have done actually extremely well in The Enterprise and ChatGPT obviously even more focused on the consumer historically.” — Levie [03:25]

2. The Rise (and Reality) of AI Agents

• The Cowork Agent Paradigm:

  • Major recent breakthrough: agents that combine coding, tool calling, and workflow automation to become “expert superworkers” for any knowledge task—not just software but marketing, legal, life sciences, and research.
  • “The mental model is what if everybody was truly an expert at using their computer...and they could write code for any task...That's basically the power of agents today, more and more.” — Levie [04:36]
  • The dream: Agents move from chatbots to autonomous coworkers that can use multiple tools, act across different systems, and run workflows end-to-end, expanding the addressable market from engineers to every knowledge worker—“a 30-50x larger market.” [06:45]

• Enterprise Will Lead Adoption:

  • Economic value (ROI on compute) highest in the enterprise, so big businesses will adopt agentic workflows first.
  • “The ROI on those tokens will just be much higher in the enterprise because it'll be generating something that impacts the GDP in some way.” — Levie [07:22]

3. Will People Actually Use AI Agents? (And Will They Work?)

• Task Depth and Subjectivity:

  • Simple, automatable, and verifiable tasks (like coding) have been easier for agents to master, leading to rapid tech insider adoption.
  • Many knowledge work tasks aren’t easily verifiable or codifiable (e.g., subjective editing or legal judgement), slowing broader diffusion.
  • “How well do these models work on much more subjective tasks, like editing a video ... actually in many cases a harder task than coding…that the diffusion of these types of technologies will take many, many years as they go through the rest of the world.” — Levie [10:10]

• Barriers: Context, Learning Curve, and Governance:

  • Most knowledge workers aren’t highly technical; context for automating tasks is fragmented across dozens of systems.
  • Enterprises need to solve hard data organization and access problems before agents can deliver their full value.
  • “If you now want to go deploy an AI agent ... you can almost think about it like a new employee joining that company. ... they're not going to know which is the one that really is the authoritative copy of that research plan or contract.” — Levie [25:13]

4. Subjectivity, Benchmarking & The “Taste” Problem

• Evaluating AI Work:

  • In creative and subjective fields (e.g., video editing), even smart agents need taste: “Do you then build ‘taste agents’ that watch the video and vote on what’s better?” — Kantrowitz [15:39]
  • Levie compares this to Hollywood productions with many editing/producers—agents might generate options but humans will still guide final choices. [15:54–16:50]

• Algorithms and “Systematized Work”:

  • Debate on whether AI-driven optimization will render the world “too systematized” and algorithmic, especially when agents do experimentation and user-facing creative work.
  • Levie believes proliferation of agents in areas like marketing, finance, and drug discovery will be broadly beneficial, increasing experimentation and enabling more efficient outcomes—“a net positive for society.” [19:22]

5. Trust, Control, and the Limits of Autonomous Agents

• People’s Reluctance:

  • Real utility comes when agents get full access to your files, inbox, data. Many, including Kantrowitz, feel uneasy about this, even though benefits are clear.
  • “To make these agents work really well...you have to be like, here’s my computer…And honestly, they work better when you take the guardrails off...” — Kantrowitz [24:40]
  • “The common practice and state of the art is: don't give [the agent] access to your inbox. Create a separate inbox for the agent and really treat that agent as another colleague that you're working with.” — Levie [30:59]

• Security and Compliance Roadblocks:

  • New vectors for cybersecurity breaches (e.g., “prompt injection” attacks)
  • Regulatory and legal liability for agent actions is still undefined—especially critical in health, finance, and legal sectors.
  • “We have massive, hundred-plus years of legal frameworks that just always assume that a user, a human, is on the other end of every transaction...When agents are doing that, this opens up a whole new field of questions...” — Levie [32:05]

6. Who Will Capture the Most Value: Labs or Builders?

• Horizontal vs. Vertical Agents:

  • Will the “super app” horizontal assistants (from labs) win, or will domain-specific agents (built by startups/verticals on top of labs) deliver the most value?
  • Levie leans toward both succeeding: “There's always domain-specific context...Even in traditional SaaS software we saw 30, 40, $50 billion vertical software companies emerge in categories where there was already plenty of horizontal products...” [44:40]

• The “Bitter Lesson” Debate:

  • As models get better, do wrapper companies (verticals) lose their edge? Not necessarily, as new use cases and greater depth always surface, creating new wrappers and integration opportunities. [49:05–51:04]

7. State of the Models: What’s Coming Next?

• New Model Releases:

  • OpenAI and Anthropic each have next-gen, “gigantic capability” models about to drop, built on massive compute and newer architectures—major jumps in agents’ abilities expected.
  • “We are nowhere close to hitting a wall...I would expect that you'll just see major improvements on all of those [agentic work areas].” — Levie [51:41]

• Evaluation and Progress Examples:

  • Levie’s team uses complex document-based knowledge tasks as benchmarks for new LLMs—already seeing double-digit improvements each model cycle.
  • “Already we’ve seen double-digit point improvement gains just in the last sort of model family update ... If we see that again ... that's just another category of enterprise work that will be unlocked.” [53:41]

8. Winners: OpenAI vs. Anthropic?

• Too Early to Call:
  • Levie, who partners with both companies, won’t pick a winner and compares today’s market to the early days of cloud computing: “It’s like trying to predict anything about the Cloud wars in 2008.” [54:17]
  • Whichever company wins lead positions, the markets are so vast (cloud infrastructure grew from $500M to hundreds of billions in 15 years) that “all of these companies will be much bigger in the future.” [55:21]
  • “Unless there's some so kind of closed proprietary research event and breakthrough that happens that just simply nobody else knows about ... I think any one lab probably has a six month to one year lead on the breakthrough AI model.” — Levie [57:44]

Notable Quotes & Memorable Moments

• On Converging Product Visions
  “If you have this AI model that is super intelligence packed...it eventually has to converge on all of this...the labs eventually need to compete head to head for all those use cases.”
  – Aaron Levie [01:35]

• On The Power of AI Agents
  “The mental model is what if everybody was truly an expert at using their computer ... and writing code ... was a lawyer...a marketer...did research. That’s basically the power of agents today.”
  – Aaron Levie [04:36]

• On Subjective Tasks & AI’s Limits
  “Editing a video is going to be actually in many cases a harder task than coding, because again ... the reward function is a lot trickier...that diffusion ... will take many, many years as they go through the rest of the world.”
  – Aaron Levie [12:29]

• On Security & Trust Issues
  “I just can’t get there. I can’t get to the point trustwise, even though I know how good it would be. I don’t want an AI system that can act autonomously in my inbox or text messages.”
  – Alex Kantrowitz [30:04]

• On Legal & Regulatory Challenges
  “We have massive hundred-plus years of legal frameworks that just always assume that a user, a human, is on the other end…When agents are doing that, this opens up a whole new field of questions...”
  – Aaron Levie [32:05]

• On Who Wins: Labs or Verticals?
  “In all the scenarios, the labs win a very big prize...the only question is how much value is created on top of the labs for the applied layer. It’s just very early to see how that plays out.”
  – Aaron Levie [44:40]

• On Big Model Drops
  “We are nowhere close to hitting a wall...I would expect that you’ll just see major improvements on all of those [agentic work areas].”
  – Aaron Levie [51:41]

• On Market Dynamics
  “It’s like trying to predict anything about the Cloud wars in 2008 ... all of these companies will be much bigger in the future.”
  – Aaron Levie [55:21]

Timestamps for Major Segments

• 01:18 — Framing the OpenAI vs. Anthropic battle & agent super-app vision
• 04:36 — The evolution and future of AI agents in knowledge work
• 07:11 — Why enterprise is the primary driver for AI agent adoption
• 10:10 — Barriers to agent use: subjective work & real-world complexity
• 15:39 — The “taste agent” challenge in subjective creative tasks
• 19:22 — How agentic experimentation will transform industries
• 24:40 — The trust and security barriers to widespread agent adoption
• 30:59 — Current best practices for agent security/compliance
• 44:40 — Who wins: foundational labs vs. applied/vertical builders?
• 51:41 — What’s coming in the next generation of LLMs
• 54:17 — Aaron Levie’s take on who wins: “It’s too early”

Final Thoughts

Aaron Levie gives a nuanced, clear-eyed take on AI’s trajectory, emphasizing business realities over hype. While the OpenAI/Anthropic rivalry attracts attention, the real revolution is the slow, complex shift as AI agents become integral to enterprise workflows, with huge but uneven progress coming as new models roll out. The winner won't be one company, but the entire AI ecosystem—provided industry solves the thorny problems of trust, integration, subjectivity, and regulation.

If you’re following the AI race, this episode is a must-listen for its inside perspective and skeptical, no-nonsense examination of where the hype is justified—and where it’s not.