Intelligence Squared: Reimagining Work with AI Agents
Episode Title: What Are The Essentials for Reimagining Work with AI Agents?
Host: Kamal Ahmed (B)
Guest: Aaron Levie (A), Co-founder and CEO of Box
Date: October 29, 2025
Episode Overview
This episode explores the current and future role of AI agents in transforming business processes. Kamal Ahmed and Aaron Levie dive deep into the practical realities, challenges, and opportunities AI agents present. They address the state of adoption, critical myths, governance and data strategy, change management, ROI, workplace evolution, and what practical steps leaders should take to prepare for the AI-driven future of work.
Key Discussion Points & Insights
1. Current State and Definition of AI Adoption
-
AI Adoption Plateau & Emerging Potential
- Most enterprises underutilize AI, content with minor efficiency improvements but missing larger opportunities.
- “I think most enterprises aren't pushing AI enough… They could actually be going way bigger with AI.” [00:01]
- Widespread use is still focused on “first wave” tools like ChatGPT; AI agents able to handle extended, complex tasks are just emerging, mainly within engineering.
- The transformation has only just begun:
- “We are quite literally 1 or 2% of the way through the full transformation that we expect to see.” [03:58]
- Most enterprises underutilize AI, content with minor efficiency improvements but missing larger opportunities.
-
What Is an AI Agent?
- Beyond a single prompt or interaction: agents can autonomously work through multiple steps, referencing memory and iterating to complete complex tasks.
- “An AI agent… can continue to execute on tasks on an ongoing basis beyond just the single prompt… to complete an entire task.” [04:44]
- Example: automating due diligence by digesting and analyzing a hundred documents, synthesizing results over time—impossible for previous generations of AI.
- Beyond a single prompt or interaction: agents can autonomously work through multiple steps, referencing memory and iterating to complete complex tasks.
2. The Data Challenge: Structured vs Unstructured Data
- Structured vs Unstructured: The 90/10 Paradox
- Legacy enterprise IT excels at structured data, but 90% of business data is unstructured (documents, images, contracts, media).
- Historically, value from unstructured data relied on human time and attention—a productivity bottleneck now being eliminated by AI’s language and multimodal capacities.
- “The more data you have, the more value you can now generate in your enterprise.” [11:27]
- New Opportunities
- Use cases: Automated contract analysis, onboarding, content generation, research synthesis, and asset management can now be radically expedited by agents.
3. Governance, Regulation & Trust
- It's a Data Problem, Not an AI Problem
- Most companies' challenge is fragmented, siloed data, not AI model sophistication.
- “Most companies don't have an AI problem. They usually have a data problem.” [12:11]
- AI agents amplify the risk of bad data hygiene, as agents lack humans’ contextual signals and may act on outdated, incorrect, or insecure data.
- Most companies' challenge is fragmented, siloed data, not AI model sophistication.
- Guidelines for Data Governance
- Consolidate sources of truth.
- Clear access/permission boundaries to avoid agents accessing or leaking unauthorized info.
4. Practical Approach: Experimentation and Scaling
- Stages to AI Agent Integration
- Start with experimentation across teams and workflows, identify high-value wins, and scale what works.
- “I tend to lean toward experimentation first… then figure out how to scale the things that are working.” [16:53]
- Start with experimentation across teams and workflows, identify high-value wins, and scale what works.
- Think Big, Push Beyond the Obvious
- Most organizations underestimate AI’s potential.
- “Most enterprises aren't pushing AI enough… AI is now probably capable of 5x units [more].” [17:44]
- Most organizations underestimate AI’s potential.
5. ROI, Measurement, and Mindset
- Measuring Impact
- ROI depends on both efficiency and topline growth; firms should consider classic KPIs but also competitive dynamics, time to market, and ability to serve new demand.
- “Push yourself to find as many of those situations as possible instead of just, okay, I was able to save 2% cost…” [21:16]
- ROI depends on both efficiency and topline growth; firms should consider classic KPIs but also competitive dynamics, time to market, and ability to serve new demand.
- Litmus Test for Adoption
- Evaluate where bottlenecks exist due to sheer information processing or synthesis—those are ripe for automation.
- Medical example: AI scribes reduce physician burnout by 30%; agents free talent for critical human tasks. [23:54]
- Evaluate where bottlenecks exist due to sheer information processing or synthesis—those are ripe for automation.
6. Hurdles and Common Pitfalls
- Tech Overwhelm & Partner Strategy
- Rapid model evolution makes tech choices daunting (Anthropic, Gemini, OpenAI, Llama, etc.).
- Build an adaptable architecture; don’t do everything in-house—wise partner decisions are key.
- Use context engineering (detailed task specification) to improve outputs, reduce “hallucinations,” and manage errors.
- “You are the manager of an AI agent in the exact same way that… you may have been a manager of new people coming into an organization.” [29:54]
- Psychological Shift
- AI agents, like new hires, need oversight—expect 90–95% completion, not perfection. Treat errors as you would with a human: review, correct, improve.
7. Organizational and Human Implications
- Jobs and Workforce Evolution
- Resistance will arise due to fear of redundancy, but most AI tasks target rote, low-value-add activities.
- True value in human work is in creativity, relationships, judgment—AI frees time for these by doing the drudgework.
- “AI for the most part is solving the kinds of problems it's very good at solving—the kind of problems that people don't really want to do.” [35:45]
- Historical Perspective
- Tech amplifies demand: automation hasn't made sectors smaller, but enabled more output and creative roles (e.g., marketing, design exploded post-Photoshop/Internet).
8. Advice for Business Leaders: What To Do Right Now
- No Excuses For Delay
- Early movers gain learning and competitive advantage—don't wait years, begin with small pilots today.
- “Experimenting, learning and building the right foundations will be the ones future proofing.” [48:40]
- Early movers gain learning and competitive advantage—don't wait years, begin with small pilots today.
- Future Outlook
- Over next 2–3 years, most information-focused tasks will be automatable; jobs become orchestration and oversight of AI-driven processes, not task-doing.
- “We will be able to automate… very different than automating a job. A job is a collection of tasks.” [44:21]
- Over next 2–3 years, most information-focused tasks will be automatable; jobs become orchestration and oversight of AI-driven processes, not task-doing.
Notable Quotes & Moments
- “Your AI strategy is your data strategy.” – Aaron Levie [47:15]
- “Start small but get going right now.” – Aaron Levie [47:16]
- “Push the limits of these AI models far past what you think is possible.” – Aaron Levie [47:17]
- “We are quite literally 1 or 2% of the way through the full transformation that we expect to see.” – Aaron Levie [03:58]
- “AI literacy is going to be the new skill for the 21st century.” – Aaron Levie [30:47]
- “The more data you have, the more value you can now generate in your enterprise.” – Aaron Levie [11:27]
Timestamps for Key Segments
- AI agent definition and capabilities: [04:44]–[07:06]
- Structured vs. unstructured data, why it matters: [07:47]–[11:30]
- Governance and data trust: [12:11]–[15:59]
- Experimentation and scaling successful projects: [16:53]–[18:29]
- Concrete ROI and measurement discussion: [20:15]–[22:18]
- ‘Litmus test’ for AI agent adoption in business: [22:54]–[25:19]
- Hurdles, hallucinations, and human oversight: [25:59]–[32:45]
- Workforce fears and historical tech analogies: [34:53]–[41:02]
- What leaders should do next quarter: [42:11]–[42:50]
- Three takeaways for businesses: [47:15]–[48:09]
Takeaways & Action Points
Aaron Levie’s Top 3 Essentials
- AI strategy is data strategy: Get your data house in order to fully leverage AI.
- Start now, start small: Experimentation leads to feedback loops and quick learning.
- Don’t underuse AI—push for more: The tech can already do more than most realize.
Practical Action Steps
- Audit data sources, eliminate silos, and clarify access/permissions.
- Run focused pilots in bottleneck areas and scale what works.
- Upskill your organization in AI literacy and context engineering.
- Set realistic expectations: AI is a collaborative tool, not a replacement for human oversight.
Tone & Style Notes
The conversation is candid, practical, and supportive of experimentation while busting “sci-fi” level hype. Both speakers repeatedly stress psychological readiness, managerial adaptation, and the importance of hands-on experience.
Summary prepared by Podcast Summarizer AI
For more conversations like this, subscribe to Intelligence Squared.
