Why companies are not ready for agentic AI | Ep. 209

The Future of AI Agents

Introduction to AI Agents

• The discussion opens with the concept of AI agents, which are pieces of AI code designed to assist individuals or companies in completing tasks with minimal human intervention.

• Keith Shaw introduces Anuka Gupta, Chief Product Officer at Rubric and lecturer at Stanford Graduate School of Business.

Current Trends in Generative AI

• Anuka shares insights on how companies are deploying generative AI, highlighting both successes and hesitations within the ecosystem.

• Many business leaders recognize the potential for significant productivity gains through AI, but acknowledge that it is still early in its adoption phase.

Successful Use Cases

• Organizations with large call center operations have successfully implemented generative AI alongside machine learning to expedite call resolution processes.

• The clear return on investment (ROI) from these implementations has led to immediate benefits in cost reduction for managing call centers.

Challenges and Hesitations

• Broader adoption of generative AI is hindered by challenges such as integrating technology effectively and concerns regarding data privacy and security.

• There is a risk that generative AI tools could inadvertently expose sensitive data due to improper permission settings within enterprises.

Data Privacy Concerns

• Anuka explains that while individual employees may not know where sensitive data resides, introducing an AI tool can lead to unintended data exposure if not managed properly.

• Concerns arise around tools like Microsoft Co-Pilot potentially leaking confidential information when employees misuse sharing permissions.

The Evolution of Agentic AI

• As discussions about agentic AI grow, there are apprehensions since many companies have yet to fully embrace previous generations of generative AI technologies.

The Future of Agentic AI: Opportunities and Challenges

The Rise of Agentic AI

• Discussion on the transformative potential of agentic AI across various organizational use cases, highlighting its scalability and impact.

• Notable companies like Microsoft and OpenAI are leading the charge in developing agentic AI, similar to their previous efforts with general-purpose AI tools.

Innovation in Startups vs. Large Corporations

• A belief that the landscape will be chaotic as both large corporations and startups innovate in agentic AI, with startups potentially having an edge due to fewer constraints.

• Startups can rapidly iterate on agentic AI solutions without being hindered by security, privacy, or legal issues that larger firms face.

Concerns Around Privacy and Data Security

• The discussion raises questions about privacy concerns associated with agentic tools compared to general generative tools, emphasizing user context as a critical factor.

• Consumer-focused tools for tasks like video editing may not raise many complaints; however, enterprises—especially regulated ones—will have significant concerns regarding data architecture and compliance.

Integration of Agentic AI into Existing Platforms

• Companies are likely to integrate agentic AI into existing platforms to enhance automation and personalization while reducing user fatigue from dashboards.

• There is potential for new startups focusing on specific agents that bridge gaps between different technologies used within organizations.

User Experience and Emerging Technologies

• The power of agentic AI lies in its ability to span multiple functions rather than serving only one persona or use case effectively.

• New technologies emerging from innovative startups will focus on creating intuitive user experiences that ensure correct actions by agents.

External Tools Impacting Hiring Practices

• Mention of external tools aiding interview preparation raises concerns about their implications for hiring processes.

The Impact of AI on Recruitment and Security

The Role of AI in Recruitment

• The advancement of technology allows AI agents to operate silently during calls, often without the other participant's knowledge.

• In recruitment, AI agents are increasingly used to assist candidates by providing real-time recommendations based on extensive training from previous interviews.

• This raises concerns about authenticity in interviews, as candidates may not be providing genuine responses but rather relaying suggestions from an AI.

• A notable example involved a candidate using ChatGPT during a Zoom interview, where the AI provided answers that were then repeated by the candidate.

• The potential future could see candidates interviewing with AI avatars instead of real people, blurring the lines between human and machine interactions.

Threat Actors and Generative AI

• Discussion shifts to how threat actors might exploit generative AI for malicious purposes beyond just improving phishing emails.

• There is a concern that these actors will conduct more sophisticated attacks using deep fakes across text, audio, and video formats to compromise identities.

• Generative AI can adaptively learn while executing tasks, enabling it to launch various types of attacks against systems and quickly identify vulnerabilities.

• The speed at which zero-day vulnerabilities can be exploited may increase significantly due to advancements in generative AI technologies.

• This creates a race between security measures implemented by organizations and the evolving tactics employed by cybercriminals.

Human Oversight in Security Decisions

• Current security tools alert humans about potential attacks; however, human judgment is still required to assess these alerts' validity.

• There's a need for human decision-making regarding significant actions during security incidents (e.g., whether to close retail stores under attack).

• Future scenarios may involve AIs investigating alerts and making recommendations based on data analysis before presenting findings to humans for final decisions.

Generative AI: Balancing Innovation and Data Security

The Need for a Dual Approach in Generative AI Implementation

• Organizations are currently working on solutions to reduce the need for numerous security analysts to sift through alerts, aiming for more efficient data management.

• Companies considering generative AI should assess their current data protection measures before fully diving into new technologies, ensuring they can prevent potential data leaks.

• A dual-pronged approach is essential: implementing protective technologies while also exploring advancements in generative AI to stay competitive in a rapidly evolving landscape.

Staying Ahead of Competitors

• Traditional organizations should not ignore emerging technologies; having someone dedicated to monitoring technological trends is crucial to avoid falling behind competitors.

• Current productivity gains from generative AI are estimated at 20-30%, but there is potential for much larger improvements that remain untapped.

Addressing Productivity Gains and Workforce Implications

• As productivity increases due to AI, concerns arise about how saved time will be utilized—risking backfilling with mundane tasks rather than fostering creativity.

• Historical patterns suggest that significant shifts in productivity often lead to societal upheaval as labor markets and regulations struggle to adapt quickly enough.

The Future of Work with Generative AI

• Discussions around whether AI replaces jobs or merely tasks will become increasingly relevant as technology evolves throughout the year.

Consumer Use Cases of Agentic AI

• There is skepticism regarding consumer applications of agentic AI, with concerns about data privacy overshadowing potential benefits.

• Despite initial hesitations, significant advancements in generative technologies are occurring within the consumer space, indicating a shift towards broader adoption.

Practical Applications of Generative Technologies

• Emerging applications include automated note-taking during meetings that merge personal notes and generate action items, showcasing practical business use cases.

The Future of Agentic AI: Personal Assistants and Consumer Use Cases

The Evolution of AI in Content Creation

• Advances in AI technology allow users to create realistic videos or images, placing themselves or family members into various scenarios effortlessly.

• Currently, the primary application of this technology is among consumers, including artists and hobbyists experimenting with AI-generated content.

Understanding Agentic AI

• Agentic AI represents a combination of advanced language models (LLMs) and tailored workflows that function like personal assistants for tasks such as household management and work-related duties.

• Users may desire a single agentic AI capable of handling multiple tasks—like managing taxes, sending emails, and planning trips—rather than separate AIs for each function.

Trust and Data Privacy Concerns

• There are existing voice technologies that effectively replace human coordinators in logistics companies, showcasing the potential for comprehensive personal assistants.

• The willingness to share sensitive data with agentic AIs often hinges on perceived benefits; however, safeguards must be established to protect user information from potential threats.

User Experience Challenges

• Users may face challenges if their agentic AI makes assumptions based on past behavior (e.g., suggesting fast food due to frequent visits), highlighting the need for context-aware interactions.

• While much data is generated daily by users (like grocery lists), it often remains unstructured. However, LLMs can learn from this data without requiring it to be organized beforehand.

Predictions for 2025 Adoption Rates