Sierra co-founder Bret Taylor on AI agents' role in an evolving global landscape

Key Concepts

• AI Agents: Autonomous software entities capable of making decisions and taking actions on behalf of users or businesses.
• Sierra Platform: A platform for building AI agents, particularly for customer experience applications.
• Outcomes-Based Pricing: A revenue model where customers pay only when the AI agent successfully completes a task.
• Conversational AI: AI systems designed to interact with humans through natural language, either via text or voice.
• LLMs (Large Language Models): Advanced AI models trained on vast amounts of text data, forming the foundation for many AI applications.
• AGI (Artificial General Intelligence): AI that possesses human-like cognitive abilities across a wide range of tasks.
• Fine-tuning: The process of adapting a pre-trained foundation model to perform a specific task.
• Supervisor Model: An internal AI model used by Sierra to critique and improve the decisions of other AI models.

Sierra Partnership with Minted

Sierra has formed a new partnership with Minted, a company known for its personalized stationery and gifts, particularly for holiday cards. The collaboration focuses on leveraging Sierra's AI platform to enhance Minted's customer experience, especially during peak seasons like the period between Thanksgiving and Christmas.

• Problem Addressed: Minted faces challenges with high customer demand during specific holiday periods, leading to potential complexities in design, shipping, and customer service.
• Solution: Minted has deployed an AI agent built on the Sierra platform. This agent aims to:
  • Assist customers in designing and ordering cards on time.
  • Improve customer service during peak demand.
  • Provide an exceptional customer experience.
• Brand Voice and Personalization: The AI agent is designed to embody Minted's brand voice, which is described as personalized and engaging. This is seen as a crucial aspect of customer interaction, with the potential for AI agents to become primary brand ambassadors. The speaker, Brett Taylor, anticipates a future where "conversation designers" and "voice designers" become common job titles, akin to web designers today.

The Nature of AI Agents vs. Chatbots

A key distinction is made between traditional "chatbots" and the AI agents developed on the Sierra platform.

• Negative Perception of Chatbots: The term "chatbot" often evokes negative associations with robotic, unhelpful automated systems that hinder human interaction.
• Positive Perception of AI Agents: AI agents, in contrast, are designed to be "delightful," empathetic, and capable of understanding user intent and taking action. They are encouraged to be named (e.g., "Duncan Smothers") to foster a more personal connection.
• Capabilities: These agents are not just conversational; they can access information and perform actions on behalf of the customer, such as changing a subscription plan, fixing an alarm system, or assisting with a Minted card order.
• Customer Satisfaction: The effectiveness of these AI agents is reflected in high customer satisfaction scores, often exceeding those of human call centers (e.g., 4.5-4.6 out of 5).
• Modality: The Sierra platform supports a multi-modal conversational experience, allowing customers to interact via:
  • Text: Through websites or messaging apps like WhatsApp.
  • Voice: Via phone calls.
  • Future Potential: Video interactions are also anticipated.

Economic Impact and Business Model

The deployment of AI agents has significant economic implications for businesses.

• Cost Reduction: Traditional phone calls to customer service can cost between $10-$20 per call. AI agents drastically reduce this cost by orders of magnitude, making it economically viable for businesses to handle a much larger volume of customer interactions.
• Shift from Cost Savings to Growth: While cost savings are a benefit, the primary driver for CEOs is growth. By reducing the cost of personalized interactions, AI agents enable companies to engage in more conversations with their customers, fostering loyalty and driving revenue.
• Sierra's Revenue Model: Sierra employs an "outcomes-based" pricing model.
  • Customers pay only when the AI agent autonomously completes a defined job.
  • If the agent needs to escalate to a human, the interaction is free for the customer.
  • This model aligns Sierra's revenue directly with the value and savings it delivers to its clients.

Differentiating AI Agents and the Sierra Platform

Brett Taylor clarifies what "agent" means in the context of AI and Sierra's specific offering.

• Categories of AI Agents:
  1. Personal Agents: Assist individuals with tasks like scheduling or managing inboxes.
  2. Enterprise Agents: Perform specific jobs within a company (e.g., software engineering, paralegal tasks).
  3. Customer Experience Agents (Sierra's Focus): Act as the primary interface for customer interactions on a company's website or through other channels.
• Sierra's Specialization: Sierra focuses on building customer experience agents that act as the "dot of your websites," handling a broad range of customer needs.
• Ease of Adoption: Sierra aims to make AI technology accessible even to non-technical teams, providing "shovel-ready" solutions that don't require massive technology investments.
• Analogy to Cloud Computing: Similar to how cloud computing evolved from infrastructure-as-a-service to software-as-a-service, Sierra operates on the "solve a problem for you" end of the AI spectrum.

Competition and Market Dynamics

The discussion touches upon the competitive landscape and potential challenges.

• Competition from LLM Companies: While larger LLM providers (like OpenAI) could potentially offer similar services, Taylor believes that deeply understanding specific customer problems and providing tailored solutions requires specialized expertise and a nuanced go-to-market strategy, creating a long tail of intellectual property.
• Startup Ecosystem: Taylor expresses strong belief in the startup ecosystem, particularly in the Bay Area, due to the rapid pace of technological change and the advantage of being nimble.
• Public vs. Private Companies: Taylor expresses relief at not being part of a public company during periods of market volatility, allowing for a focus on customer value and company building without the daily pressure of stock market fluctuations.

Enterprise Spending and Geopolitical Uncertainty

The impact of global events on IT spending is considered.

• Uncertainty and AI: While geopolitical turmoil creates uncertainty, it also drives attention to AI as a solution for cost savings (e.g., offsetting tariff-related cost increases) and growth in a shifting market.
• AI Applications vs. Models: Spending is accelerating in the AI applications market (where Sierra operates) because these solutions offer a clear return on investment (ROI) by being readily deployable, unlike building solutions from scratch using raw models.

Focus and Specialization

Sierra's strategy of focusing on a specific niche is highlighted.

• Advantage of Specialization: In a rapidly evolving market, a focused approach on specific use cases (like AI agents for customer experience) is seen as an advantage, especially when scrutiny on spending increases.
• Contrast with Bundled Offerings: Unlike large tech companies bundling AI with broader software suites, Sierra's specialization allows it to be a clear, go-to solution for a specific need.

The Future of AI and Labor Force Impact

The conversation delves into the broader implications of AI.

• AI's Maturation: AI models are becoming significantly better and cheaper at a rapid pace. The cost of output tokens has decreased by orders of magnitude, and quality has improved.
• Impact on Labor:
  • Automation of Tasks: AI will automate certain tasks, similar to how ATMs changed banking roles without eliminating branches.
  • Higher-Order Work: Humans will likely shift to more leveraged and valuable tasks.
  • Historical Parallels: The transition is compared to historical shifts like the introduction of calculators or computers, where new roles emerged.
  • Democratization of Intelligence: AI is making intelligence plentiful, akin to how electricity made power plentiful.
  • Disruption and Empowerment: While disruptive, AI can be empowering, enabling individuals to excel in their jobs. The analogy of Microsoft Excel and pivot tables is used to illustrate how embracing new tools leads to greater efficiency.
  • Education: AI can personalize learning experiences for children, acting as a tutor.
• New Use Cases: The full range of AI use cases is still being discovered, much like the early days of the internet.

AI Safety and Governance

The critical aspect of AI safety is addressed.

• OpenAI's Mission: Safety is paramount to OpenAI's mission of ensuring AGI benefits all of humanity.
• "First, Do No Harm": The approach to AI safety is likened to the medical oath, emphasizing the prevention of harm.
• Expertise and Research: OpenAI is employing top minds and conducting in-depth research on safety, including areas like "jailbreaking."
• AI as a Solution to AI Problems: The belief is that more AI development is needed to solve AI safety challenges.
• Public-Private Partnership: Effective AI governance requires collaboration between public and private sectors.
• Forward-Looking Regulation: Regulations need to be future-oriented to avoid becoming obsolete quickly due to the rapid pace of AI development.
• Humility in Prediction: There's a recognition that predicting all first and second-order effects of AI is challenging, and innovators should be allowed to build and explore while maintaining safety guardrails.
• Lessons from Social Media: The speaker reflects on the evolution of social media, acknowledging that unforeseen consequences can arise and that technology companies must be responsible in addressing issues as they emerge.

OpenAI's Structure and Future

• Mission-Driven Focus: OpenAI's primary focus remains on achieving its mission of creating AGI for the benefit of humanity.
• Non-Profit Structure: The non-profit arm of OpenAI is not being dissolved, despite discussions about restructuring.
• AGI's Economic Value: The development of AGI is expected to unlock unprecedented economic value across various sectors.

Sierra's Underlying Technology

• Constellation of Models: Sierra utilizes a "constellation of models," integrating off-the-shelf LLMs (like those from OpenAI), fine-tuned models, and proprietary models.
• Proprietary Model Development: Sierra develops its own models for specific, hard-to-solve problems, such as accurately detecting background noise in customer calls.
• Multi-Model Interaction: A single customer interaction might involve accessing four to eight different models to determine the appropriate action.
• Reasoning Capabilities: Sierra's models are designed not just for generation but also for reasoning, enabling them to make complex decisions.
• Supervisor Model: An internal "supervisor" model critiques and refines the decisions of other AI models, improving overall effectiveness.
• Rapid Iteration: Sierra stays abreast of the latest research to ensure its platform incorporates best practices and the most advanced models.
• Model Differentiation: While some models may become commoditized, the most advanced and differentiated models (like GPT-4) will retain significant value due to their superior performance and user experience.
• Consolidation in Model Market: Similar to the cloud infrastructure market, consolidation is expected in the LLM space due to the high cost of training and the need for scale.

The Evolution of Human-Device Interaction

• Beyond Traditional Interfaces: The shift from punch cards and keyboards to graphical user interfaces, smartphones, and now conversational AI signifies a fundamental change in how humans interact with technology.
• Conversational First: The future may see a move towards conversational interfaces becoming the primary mode of interaction, reducing reliance on screens.
• Invisible Technology: The goal is for technology to become more integrated and less intrusive, "melting away" into the background.
• Pace of Change: The rapid evolution of AI makes older technologies (even those from a few years ago) feel like "horse and carriage versus a flying car."
• Value of AI Applications Companies: Companies like Sierra are valuable because they can absorb the complexity of rapidly changing AI models and package them into user-friendly solutions for businesses.

Conclusion

The conversation highlights the transformative potential of AI, particularly through the development of sophisticated AI agents. Sierra's partnership with Minted exemplifies how this technology can enhance customer experiences and drive business value. The discussion underscores the rapid advancements in AI, the economic opportunities it presents, the critical importance of safety and responsible development, and the profound impact it will have on the labor force and human-device interaction. The future promises a more conversational and integrated technological landscape, with AI agents playing a central role.