Sema4.ai

AI Agents: The Next Big Thing in Generative AI OpenAI's third level toward AGI is all about Agents—systems that take action and get things done for us. Companies like Salesforce and HubSpot have pivoted toward AI agents. In 2025, AI agents are expected to become reliable and more widely used. So, what are AI Agents, and why should you care? While there is no universally agreed definition, I’ve come across two valuable perspectives recently. Bret Taylor, Chairman of the Board at OpenAI and CEO of Sierra, a company building AI Agents for customer service (ex CTO at Facebook, Co-CEO of Salesforce), classifies AI agents into three main types in No Priors Podcast Ep 82: Personal Agents: General-purpose AI assistants that help with tasks like managing email, scheduling, and vacation planning. Imagine an agent that finds items, compares options, negotiates prices, makes purchases, and arranges delivery—all without you spending hours browsing. Widespread adoption of personal agents is still a bit far off due to challenges in human-computer interaction and system integration. Persona-Based Agents: Specialized for specific domains, like Harvey for legal work or Cognition for coding are already making a significant impact. We'll see AI agents in sales and marketing, such as Salesforce's "Agentforce Sales SDR" for handling inbound leads. Social Media, Content, and Newsletter Agents are around the corner. https://agent.ai, is an early marketplace of AI agents from HubSpot. Company Agents: represent a company’s digital presence, handling customer service, inquiries, and transactions, like Sierra. Bret envisions a future where customers interact with a "blank box" on a website, asking anything and getting things done—an intuitive, free-form interface where the multi-modal AI agent takes real actions. This is the next evolution of conversational AI, offering more sophisticated and capable systems at scale. Menlo Ventures recently examined how AI agents are advancing enterprise automation. They categorized the AI agent market based on domain specificity and LLM autonomy. Enterprise AI Agents are versatile, handling various functions and workflows using natural language SOPs or rulebooks, similar to training new employees. Platforms like Sema4.ai are in this space. To learn more about Enterprise AI Agents, check out this webinar by Digital Workforce Services Plc and Sema4.ai: https://lnkd.in/d38rMH7S Although AI agents may seem futuristic, they are already a reality. As definitions and capabilities evolve alongside GenAI, one thing is clear: AI agents will fundamentally change how work gets done. They’ll take on tasks and processes and eventually assume roles that redefine business operations. As agents become more capable, they will revolutionize workflows, boost productivity, and increasingly act as integral colleagues in the workplace. #AIAgents #GenAI

Generative AI applications have strong product-market fit in three core areas today: search, synthesis, and content or code generation. However, the technology's potential extends far beyond these initial use cases. While AI that can read and write is valuable, the truly transformative applications lie in AI systems capable of planning and taking action. This is the biggest opportunity for AI apps.?Recent advancements in multi-step reasoning, external memory integration, and tool use have paved the way for an emerging category of AI agents to capitalize on this potential. Led by companies including Anterior (formerly Co:Helm), Sema4.ai, and Cognition, these systems are pushing the boundaries of AI capabilities, enabling end-to-end process automation. In two new articles, my partners and I have lay out Menlo Ventures' thesis on this emerging market. The first piece defines AI agents and traces the architectural evolution of the modern AI stack, from few-shot prompting through Retrieval-Augmented Generation (RAG) to fully-fledged agentic systems: https://lnkd.in/gaBBVS4f The second article surveys the current agent landscape and examines how this paradigm shift is reshaping the enterprise, from horizontal automation to verticals like healthcare: https://lnkd.in/gFyqwiuZ cc: Tim Tully Joff Redfern Deedy Das Steve Sloane Jean-Paul (J.P.) Sanday Naomi Pilosof Ionita

OpenAI's recent o1 model is a step forward in reasoning. OpenAI recently announced o1, introducing built-in chain-of-thought (CoT) reasoning. This development follows the GPT-4o release, which brought multi-modal features to the forefront. CoT is nothing new to AI developers, and most agents use that method extensively in various configurations. What's new here is that CoT-focused training has been baked directly into the LLM. The Good: - Enhanced reasoning performance in complex scenarios, which will improve reasoning for tasks like contract analysis - ChatGPT now behaves more like an agent in terms of reasoning and can do multiple iterations on a problem The Not-So-Good: - Lacks tool calling support for now - Limited customization options and fixed reasoning process as a black box While o1 may not be an immediate game changer for developers, it paves the way for future advancements. Personally, I'm eager to see other model providers incorporate similar reasoning-focused training with more granular controls. Right now, I personally think 4o was a bigger leap forward for Agents than o1, but it's too soon to tell.