Intelligence Squared

What Are The Essentials for Reimagining Work with AI Agents?

October 29, 2025

Key Takeaways Copied to clipboard!

  • Most enterprises are currently underutilizing AI, operating at the basic 'chatbot' level, while the real value lies in deploying AI agents capable of looping, reasoning, and completing complex, multi-step tasks autonomously. 
  • A company's AI readiness is fundamentally a data strategy problem, not an AI problem; success with agents requires governing and cleaning up unstructured data silos to establish clear sources of truth and security boundaries. 
  • Business leaders should adopt an experimental, lean approach to AI adoption now, pushing models beyond perceived limits (e.g., expecting 5x or 10x gains instead of 1x) and focusing on augmenting human value creation rather than fearing job replacement. 

Segments

Current State of AI Adoption
Copied to clipboard!
(00:01:48)
  • Key Takeaway: Most current AI adoption is limited to basic question-answering and summarization use cases, similar to ChatGPT.
  • Summary: AI adoption across the UK and US is largely confined to using models for information assistance, such as asking questions or summarizing content. The next phase involves AI agents automating real tasks within workflows, with task size capabilities growing significantly over the past year. This transformation is still in its earliest stages, estimated to be only one or two percent complete.
Defining AI Agents vs. Current AI
Copied to clipboard!
(00:04:44)
  • Key Takeaway: AI agents differ from standard AI by possessing the capability to loop and execute tasks continuously beyond a single prompt pass-through.
  • Summary: Standard AI models operate on a single pass-through, generating a response based on internal knowledge or a quick search. An AI agent, conversely, can loop through steps, utilizing temporary memory to complete entire, complex tasks, such as analyzing 100 due diligence documents to generate a full report. This agentic workflow represents the major shift toward real task automation.
Structured vs. Unstructured Data Value
Copied to clipboard!
(00:07:47)
  • Key Takeaway: Unstructured data, comprising about 90% of enterprise information, is becoming an untapped goldmine unlocked by AI’s ability to understand text and multimodal content.
  • Summary: Structured data resides in databases (CRM, ERP) and has historically been queryable, but unstructured data (contracts, memos, research files) requires human review to extract value. Large Language Models enable computers to finally understand this messy data at scale, allowing for complex queries like instantly finding all contracts expiring in 30 days with specific clauses.
Governance and Data Hygiene
Copied to clipboard!
(00:11:34)
  • Key Takeaway: Most companies have a data problem, not an AI problem; deploying agents across siloed systems creates security risks and authoritative source-of-truth issues.
  • Summary: When AI agents roam across multiple systems (e-signature, document management, CRM), they risk finding the wrong document or revealing unauthorized information if access controls are not perfectly synchronized. Leaders must establish clear systems of record per data type (e.g., one source for all HR data) to govern information effectively before deploying agents at scale.
Experimentation and Scaling Strategy
Copied to clipboard!
(00:16:53)
  • Key Takeaway: The recommended approach for AI adoption is a lean methodology: experiment widely to find successes, scale those that work, and aggressively push models beyond current perceived capabilities.
  • Summary: Companies should try many different things to build a flywheel of success, while also being prepared to ‘go big’ on proven methodologies learned from others. A common pitfall is settling for limited gains; enterprises must actively push AI models to deliver capabilities far exceeding initial expectations (e.g., 5X output instead of 1X).
Measuring ROI Beyond Cost Savings
Copied to clipboard!
(00:20:15)
  • Key Takeaway: The most powerful ROI from AI agents comes from using them as a competitive weapon to drive revenue, increase output, and accelerate sales cycles, not just to achieve minor cost efficiencies.
  • Summary: Companies should bias their measurement toward revenue generation, such as improving win rates or speeding up the revenue cycle, rather than solely focusing on saving a small percentage of costs. AI augmentation frees up time, which should be reinvested into value-creating activities like innovation, better customer service, and shipping more product.
Litmus Test for Agent Deployment
Copied to clipboard!
(00:22:49)
  • Key Takeaway: The primary litmus test for immediate AI agent upside is identifying bottlenecks where human time is consumed by reviewing or generating digital information.
  • Summary: If a workflow is bottlenecked by the time it takes a human to review a contract, create marketing assets, or analyze sales data, that area offers the greatest near-term upside for AI agents. Tasks that are primarily about human interaction with the outside world, rather than information processing, will see limited immediate benefit.
Hurdles: Overwhelm and Hallucinations
Copied to clipboard!
(00:25:39)
  • Key Takeaway: Business leaders are often stalled by the overwhelming rate of AI model breakthroughs and the need to manage potential errors by treating agents as new employees requiring review.
  • Summary: The rapid succession of new AI models (Anthropic, Gemini, OpenAI) creates decision paralysis, suggesting an architecture that can leverage any breakthrough is necessary. While hallucinations are improving, humans must review agent outputs, especially for factual data like financial filings, accepting that AI currently delivers the 90-95% of the work, not the perfect 100%.
Future of Work and Job Evolution
Copied to clipboard!
(00:34:56)
  • Key Takeaway: AI agents will automate discrete tasks, shifting human roles toward coordination, review, and value creation activities like client understanding and innovation, rather than causing mass unemployment.
  • Summary: Employees often resist AI due to perceived competition, but AI excels at the drudgery workers dislike (e.g., data collation, report writing). This reallocation of time means organizations will spend less time on manual data movement and more on high-value activities like client engagement and innovation, leading to new job creation across the economy.
Actionable Next Steps for Leaders
Copied to clipboard!
(00:42:42)
  • Key Takeaway: Leaders must start experimenting with AI agents immediately to build crucial feedback loops and organizational ‘wherewithal’ rather than waiting until 2026 or 2027.
  • Summary: Starting small allows companies to learn quickly what works and what doesn’t, which builds momentum that snowballs over time. Waiting to dive in means missing out on the competitive advantage gained from early lessons in this rapidly evolving environment. The time to begin experimenting is now.