2026年AI智能体发展状况报告

The 2026 State ofAIAgents Report How enterprises are building and Contents Foreword Research insights Part I: The current landscape Part II: Going deeper Part III: The path forward AI agents in production23 Outlooks and perspectives38 Additional resources Foreword Over the past several months, AI agents have moved from experimental tech-nology to infrastructure that enterprises use in production. Unlike traditionalsoftware that waits for human input, agents reason through problems, make This shift toward automated workflows and multi-step agentic systems funda-mentally changes what organizations require from AI: models that are secure In partnership with research firmMaterial,we surveyed over 500 technicalleaders in the United States across company sizes and industries to understandhow organizations are using agents today and where they see opportunity in The data shows this shift in concrete terms. According to our research, morethan half of organizations (57%) now deploy agents for multi-stage workflows,including 16% that have progressed to cross-functional processes spanning Given the growth of agentic coding over the past 12 months, it's unsurprisingthat nearly 90% of organizations surveyed use AI to assist with coding today. Organizations report AI agents free up more time across the entire developmentlifecycle—from planning and ideation (58%) to code generation, documenta- The impact also extends well beyond software development. Beyond engineer-ing, the highest-impact use cases include data analysis and report generation(60%) and internal process automation (48%), with 56% planning to implement Eight in 10 organizations believe AI agents have already delivered measurableROI, with another 1 in 10 saying they expect them to deliver more economicimpact in the future. The question facing leaders in 2026 isn't whether to adopt Read on to learn more about how today’s leaders are building AI agents in the Survey methodology In partnership with Material, Anthropic surveyed over 500 technical leadersacross company sizes and industries in late 2025 to understand current AI agentadoption patterns and future plans. Respondents included engineering leaders, Research Research insightsPart I: The current landscape How organizations are deploying AI agents today More than half of businesses are deployingAI agents with multi-step workflows Trend summary Organizations are deploying AI agentsfor work that goes well beyond chatinterfaces and single-step automation.More than half (57%) now use agents Why this matters The shift from task automation toprocess orchestration represents afundamentally different use case—and a different value proposition.Organizations that master multi- Nearly all organizations are adoptingcoding agents Trend summary More than 9 in 10 organizations nowuse AI to assist with coding. The vastmajority (86%) have moved beyondexperimentation and are deployingAI coding agents for production code, Why this matters AI coding agents have moved fromexperimental to mainstream, withthe majority of organizations alreadydeploying them in productionenvironments. Organizations thatembrace these tools strategicallyare accelerating delivery timelines, AI coding agents boost developer productivity Trend summary AI agents are increasing productivityacross the entire developmentlifecycle, not just code generation.Organizations report time gains infour key areas at nearly identical rates: Why this matters The impact spans every phaseof software development, whichmeans teams can improve bothengineering velocity and code qualitysimultaneously. Organizations thatintegrate AI agents across the fulldevelopment process can compoundthese gains, turning what might be Leaders favor hybrid approaches to agenticdevelopment over building from scratch Trend summary Most organizations (47%) takea hybrid approach to AI agents,combining off-the-shelf solutions withcustom-built components. About onein five (21%) rely entirely on pre-built Why this matters The hybrid model dominance suggestsno single approach delivers everythingorganizations need. Off-the-shelfagents get teams running quickly butoften lack the customization requiredfor specific workflows or proprietarysystems. Fully custom builds offer Research insightsPart II: Going deeper Expanding use cases and measuring ROI Agentic use cases are expanding beyond coding Trend summary Organizations expect AI agents toexpand well beyond engineering andIT functions over the next 12 months.Research and reporting leads adoptionplans at 56%—particularly among mid- Why this matters Research and reporting work spansevery function and level of anorganization, making it a high-leveragestarting point that builds institutionalcomfort with AI agents beforedeploying them in more sensitive orcomplex workflows. Organizationsthat successfully implement agents In addition to coding, data analysis andprocess automation are the enterprise’s most T