AI代理圣经

The AI agent landscape is evolving rapidly, with500+ startups founded since 2023. Here is our topAI agent research — featuring the trends, keyplayers, and 6 predictions to watch all based onCB Insights predictive intelligence. AI agents are defining the next wave of tech innovation. Every big tech company and a rapidly growing private market landscape are building agentofferings targeting horizontal use cases and industries fromfinancial services tomanufacturing. For enterprises across sectors, one question is becoming unavoidable:Which AI agentstrategies will separate market leaders from those left behind? Enterprises are under pressure to build and implement agents as these LLM-based systemschange how companies operate, hire, and scale. Across 9 reports, discover where startup innovation is pointing, promising partnership andacquisition targets, and key trends to watch based on CB Insights predictive intelligence. AI Agent Bible Foreword Manlio Carrelli CEO of CB Insights Seven years ago, I watched companies struggle with primitive chatbots that broke themoment customers went off-script. Today, those early systems have given way toAI agents, which in 2 short years have movedfrom experiment to enterprise priority. I’m seeing agent mentions on earnings calls10xsince 2023. The velocity is unlike anything we’ve tracked before. What strikes me most: agents are climbing the value chain faster than any technology I’veseen. They gained traction in customer service – which made sense, given the sector’s earlyAI adoption.82% of organizations we surveyed in June 2025said they will be using AI agentsin customer support in the next 12 months. But now they’resupporting clinical decisionsathospitals,assessingfinancial riskat banks, anddrafting legal memosat lawfirms. This is why25% of AI agent startups founded since 2023are already deploying, achievinglevels of commercial distribution that previously took startups 5+ years, according to ourCommercial Maturity scores. The momentum is accelerating. Among the1,500+ tech marketsCB Insights tracks,fiveofthe top 10 by deal activity in 2025 are agent-related.One infive new unicornsis buildingagent technology. Half of thetop 20 revenue-generating agent startupsdidn’t even exist threeyears ago. But here’s what I keep telling our enterprise clients:agents are only as good as the data theyrun on. The real competitive advantage isn’t in the AI – it’s in operationalizing accurate,comprehensive data and the orchestration layers that make agents actually useful. That’s where the roadblocks are. Enterprises understand agents matter, but they’re strugglingwith integration complexity, security concerns, and vendor sprawl. The opportunity ismassive for those who can navigate this transition. The research and data that follows breaks down this rapidly evolving landscape by vertical,maturity, and momentum.Remember: in the age of AI agents, the edge doesn’t live in themodel. It lives in connecting purpose-built agents to purpose-specific data. Those who get this right will shape the economy agents are creating. Predictive Intelligence on Private Companies With CB Insights, you see your next move –first. Rare data straight from companiesbuilding the future. Powerful AI that automates the analysis. Table of Contents Outlook on AI agents 6 AI agent predictions looking into 2026 The AI agent ecosystem 13 The AI agent market mapThe AI agent tech stackThe AI agent revenue race 131929 AI agents make inroads across enterprise workflows Y Combinator’s 2025 Spring batch reveals the future of agentic AI31Building the agent economy: How cloud leaders are shaping AI’s next frontier36The summer of vibe coding is over43 Industry-specific applications gain momentum 51 3 markets fueling the shift to agentic commerce51The industrial AI agents & copilots market map58100 real-world applications of genAI acrossfinancial services and insurance64 Additional resources 68 AI Agent Bible Outlook on AI agents 2025: Agents with guardrails An AI agent is an LLM-based system designed to independently perform tasks on a user'sbehalf by reasoning, planning, leveraging memory, and interacting with external tools andother agents. Most agents today operate within constrained environments, utilizing structured workflowsand “guardrails” to complete specific objectives while retaining some decision-makingcontrol. As foundation model capabilities improve, agents are expected to become increasinglyautonomous. Beyond 2025: Fully autonomous agents Agents will operate without human involvement, with more sophisticated decision-making,adaptability, and task execution. Looking into 2026, watch for new form factors outside of the copilot/chatbot interface thatwill push the boundaries of what an “agent” is. AI Agent Bible Early indications of this include “AI-native” workspaces — tools and platforms built from theground up around AI capabilities, rather than layering AI features on top of a tr