Podcast Summary: Implementing and Scaling AI Agents in Business

Podcast: Intelligence Squared
Host: Kamal Ahmed
Guest: Ben Kuss (CTO, Box)
Release Date: January 29, 2026
Length: ~40 minutes

Episode Overview

In this episode, Intelligence Squared explores the concrete steps organizations must take to successfully implement and scale AI agents in business. Host Kamal Ahmed is joined by Ben Kuss, CTO of Box, who outlines the five essential steps for becoming AI-ready: auditing data, establishing a single source of truth, starting small, pushing beyond chatbots to true AI agents, and, crucially, measuring value. The conversation is rich with practical insights and examples, emphasizing the need for strong foundations, thoughtful governance, and a culture that supports responsible experimentation.

Key Discussion Points and Insights

1. Revisiting Legacy Systems and Preparing for AI

• [00:01] Ben Kuss: Highlights that companies need to revisit legacy systems to stay responsive to AI innovation.

"It's time for them to revisit the old way of doing things, to basically prepare not just for the current benefit of AI, but for the expectation that as AI agents do more, companies need to lay the right foundation."

2. The Five Essential Steps to AI Readiness

[02:05] Kamal Ahmed:

• Introduces Ben Kuss and outlines the episode’s structure:
  1. Auditing data
  2. Creating a single source of truth
  3. Starting small
  4. Pushing beyond chatbots
  5. Measuring value

Step 1: Audit Your Data Architecture

[03:58]

• Many AI project failures are rooted in data issues, not AI issues itself.
• [04:28] Ben Kuss:

  "We often hear from customers that they don't actually have an AI problem. Instead, they have a data problem."

• Data silos and legacy systems prevent effective AI deployment.
• Both structured (e.g., databases) and unstructured (e.g., documents, emails) data must be accessible and well-organized.
• Treat AI agents as a new employee—if they can't access data, they can't contribute value.

How to audit:

• [06:33] Ben:
  • Do a top-down audit: Where is all company data stored? Which tools are modern, and can they integrate with AI?
  • Focus especially on unstructured data, as generative AI excels at leveraging this information.

[09:04] Ben:

• The concept is simple: Centralize data, select suitable platforms. The challenge is change management—getting people to actually adapt and migrate systems.

[10:27] Ben:

• Red flag: Trying to tackle overly complex use cases first.

"Have big ambitions, but start small is typically a focus area for people that we see who are more successful in their projects."

Step 2: Establish a Single Source of Truth

[11:39] Kamal:

• Organizations must consolidate their critical data—especially important datasets (CRM, HR, financial)—into accessible, modern systems, at least for key use cases:

"Let's focus on the things that are the most important... use that as your test case." ([11:49] Ben)

Governance and Access Control

[12:56] - [13:34]

• Issue: “AI doesn’t keep secrets”—AI will provide whatever information it can access.

"If you ask AI something, it will tell you everything it knows about that thing, whether or not you should have access." ([12:49] Kamal) "It's absolutely critical that you do not give AI access to what that person doesn't have access to... role-based access controls are key." ([14:08] Ben)

• Rely on platforms that apply strong access controls and don’t try to roll your own from scratch.

Step 3: Start Small and Learn

[15:47] Kamal:

• Focus on clear, simple use cases that provide early wins and confidence.
• [16:15] Ben: Example—A client advisory firm wanted AI to make holistic recommendations, but had fragmented data. They began by using AI for basic data extraction and consolidation, building capability in manageable increments.

[18:04] Kamal:

• Company culture matters for adoption. How do you measure pilot success and know when to stop?
• [18:57] Ben:

  "The key... is to make sure you have directly responsible people who have clear exit criteria and a clear goal that matters to them."

• Focus on automating mundane tasks with clear ROI, not high-level, vague ambitions.

Step 4: Push Beyond Chat—Embrace Agentic AI

[21:26] Kamal:

• Public AI hype focused on chatbots—companies must now move beyond simple Q&A and use AI agents for substantial, background workflows.
• [22:16] Ben:

  "Chatting with an intelligent system... has become a new interface... but you want AI to do real work for you, not just answer questions."

• Agentic experience (AX): AI agents can perform tasks, not just respond via chat.
• Modern workflows let agents anticipate needs and prepare work in the background.

"Why not have it know that I have a meeting, and have it prepare the work before I need it?" ([24:30] Ben)

• Managing and learning to work with AI agents is quickly becoming a core business skill.

Step 5: Measurement and Continuous Improvement

[29:16] Kamal:

• Success is about embedding measurement—track ROI, set and meet KPIs/OKRs that reflect true business value improvement.
• [30:04] Ben:

  "Start with the most important business value, but identify a more immediate, direct metric to measure... Start small but make sure you're making processes 10x faster, and those add up."

• Do not become attached to vague, large-scale KPIs without clear links to day-to-day improvements.

Notable Quotes & Memorable Moments

• "AI doesn't keep secrets." – Kamal Ahmed [12:49]
• "We often hear from customers they don't actually have an AI problem. Instead, they have a data problem." – Ben Kuss [04:28]
• "Have big ambitions, but start small." – Ben Kuss [10:27]
• "The agentic experience... is almost like talking to somebody in your organization who has a job, a role, and access to certain systems." – Ben Kuss [22:45]
• "Make sure you have directly responsible people who have clear exit criteria and a clear goal." – Ben Kuss [19:35]
• "Start with direct metrics... processes that improve 10x add up over time to shift higher-level company KPIs." – Ben Kuss [30:15]

Timestamps for Key Segments

• 00:01 – Introduction: Revisiting legacy systems for AI
• 02:05 – The five essential steps to AI readiness
• 04:28 – "We don't have an AI problem; we have a data problem"
• 06:33 – How to audit your data
• 09:04 – The practicalities and challenges of data centralization
• 11:39 – Achieving a single source of truth
• 12:49 – "AI doesn't keep secrets" (Governance/access control)
• 14:08 – Role-based access controls for responsible AI use
• 16:15 – Real-world example: Starting small for impact
• 18:57 – Measuring pilot success with clear criteria
• 22:16 – Beyond chat: What “agentic AI” means in practice
• 24:30 – AI agents proactively preparing work
• 29:16 – Embedding measurement and driving ROI
• 30:04 – Practical measurement advice and pitfalls to avoid
• 33:05 – Strategic advice for leadership: Next 3-6 months
• 34:32 – Final reflections: Foundations are what matter most

Final Strategic Takeaways

• Foundations matter more than tools: Break down data silos, modernize systems, and ensure access/governance first.
• Start small but act strategically: Early, focused projects build organizational confidence and capability.
• Harness real agents, not just chatbots: Use AI for substantive, background tasks and leverage the growing agentic ecosystem.
• Measure what matters: Define direct, practical metrics for success; aggregate incremental gains.
• Understanding evolves: Stay current with the fast-moving AI landscape, and continuously experiment with the latest capabilities.

For leaders:
Begin by making your organization’s data accessible and well-governed, create testable use cases that matter, empower responsible champions, and foster a culture of continuous measurement and curiosity. The winning organizations are not those with the most hype, but with the best foundations and the ability to adapt at scale.