Intelligence Squared

Implementing and scaling AI agents in business

January 29, 2026

Key Takeaways Copied to clipboard!

  • Successful AI adoption hinges not on ambition, but on establishing solid data foundations, clear governance, and purposeful experimentation to move beyond initial hype. 
  • The first essential step for AI readiness is auditing the data architecture, as most AI failures stem from data silos and inaccessibility rather than model limitations. 
  • Organizations should push beyond simple chat interfaces to leverage agentic AI that can perform complex, asynchronous work, requiring a focus on role-based access controls to maintain security and governance. 

Segments

Preparing for Agentic AI
Copied to clipboard!
(00:00:01)
  • Key Takeaway: Companies must revisit legacy systems now to lay the foundation for responding to complex innovations driven by AI agents.
  • Summary: Organizations need to prepare their infrastructure not just for current AI benefits but for the increasing complexity of future AI agents. This preparation involves revisiting and updating legacy systems. This proactive approach ensures readiness for upcoming innovations in the AI landscape.
Five Essential AI Readiness Steps
Copied to clipboard!
(00:01:50)
  • Key Takeaway: The five essential steps for AI readiness include auditing data, establishing a single source of truth, starting small, moving beyond chat, and measuring value.
  • Summary: Ben Kus outlines five critical steps for organizations aiming to adopt AI safely and effectively at scale. These steps cover foundational data work, strategic experimentation, and defining success metrics. This framework guides leaders from initial experimentation to achieving lasting impact with agentic AI.
Data Auditing and Architecture
Copied to clipboard!
(00:03:58)
  • Key Takeaway: Many AI failures are rooted in data problems, specifically the lack of centralized, accessible data architecture for both structured and unstructured information.
  • Summary: Companies often face a data problem rather than an AI problem, where critical data resides in silos across legacy systems. AI agents, analogous to new employees, cannot deliver value without secure and readily available access to this data. Auditing involves both top-down review of central IT tools and bottom-up identification of data sources tied to specific use cases.
Achieving Single Source of Truth
Copied to clipboard!
(00:11:36)
  • Key Takeaway: Establishing a single source of truth (SOT) requires identifying critical data platforms (structured, unstructured, CRM, HR) and prioritizing migration from old systems for key test cases.
  • Summary: The SOT is achieved by assessing where the most important data resides and ensuring it is readily available for AI. While a full system overhaul is not immediately necessary, moving critical data out of legacy systems into modern platforms is a key first step. This process often involves change management as employees adapt to new centralized systems.
Governance and Access Control
Copied to clipboard!
(00:12:38)
  • Key Takeaway: Because AI does not inherently respect permissions, governance must enforce user-based and role-based access controls so AI only surfaces data the interacting user is authorized to see.
  • Summary: The risk of AI sharing protected information is high if access controls are not explicitly managed, as AI is designed to be helpful and will share everything it can access. Governance requires ensuring the AI system respects existing permission structures tied to the user’s role. Organizations should leverage existing platforms with built-in access controls rather than building custom solutions from scratch.
Strategy of Starting Small
Copied to clipboard!
(00:15:47)
  • Key Takeaway: Successful AI adoption involves having big ambitions but starting with simple, high-impact use cases to ready employees and gain measurable, incremental value.
  • Summary: Attempting the most complex, high-value project first is often a challenge; starting small allows teams to gain experience and understand AI’s capabilities and limitations. A good starting point is often data extraction or standardizing reports for a specific process, as demonstrated by an example in financial advising. These smaller wins build organizational readiness and experience quickly.
Moving Beyond Simple Chat
Copied to clipboard!
(00:21:26)
  • Key Takeaway: The Agentic Experience (AX) moves beyond simple Q&A chat by deploying AI agents with specific objectives that can perform background work asynchronously via workflows.
  • Summary: While chatting with intelligent systems (the AX) is common, organizations must progress to using AI agents that can execute tasks rather than just provide answers. This involves creating agents with defined roles, access, and objectives to perform specific work, such as pre-preparing data for an upcoming meeting. This shift leverages AI’s capability to think, reason, and loop through complex tasks.
Realizing Exponential Potential
Copied to clipboard!
(00:26:01)
  • Key Takeaway: The rapid, continuous improvement in AI models means organizations must constantly re-experiment, as capabilities available today are dramatically different from those even six months prior, unlocking 5x or 10x improvements.
  • Summary: The pace of AI model releases makes it difficult for users to keep up with what is currently possible, leading to underutilization based on outdated experience. Agentic coding, for example, represents a new paradigm where agents can manage complex, multi-step tasks over minutes or hours, which is fundamentally different from simple function completion. Continuous experimentation is necessary to realize these exponential gains.
Embedding Measurement and ROI
Copied to clipboard!
(00:29:41)
  • Key Takeaway: Delivering true business value requires measuring direct, immediate metrics (like reducing a 10-hour task to one hour) that ladder up to higher-level KPIs like overall efficiency.
  • Summary: Using AI must result in measurable business value, not just the adoption of the technology itself. Leaders should focus on direct, immediate metrics for specific processes rather than waiting for high-level KPIs to shift immediately. Success is achieved when many small, 10x improvements across various processes accumulate to impact the overall organizational metrics over time.
Next Steps for Leadership
Copied to clipboard!
(00:32:41)
  • Key Takeaway: In the next three to six months, leaders must ensure their data foundation is AI-ready and focus on enabling an ecosystem of specialized AI agents that cooperate to accomplish complex tasks.
  • Summary: The next critical decision for leadership is confirming that critical data is available and accessible to AI systems. The future involves an ecosystem where specialized AI agents interact with each other to solve larger problems, similar to how people use different tools for complex work. Organizations must focus on building this cooperative agent environment to realize future value.