超越炒作：衡量真正的企业人工智能代理采用率

Arjun Bhatia+1 312 364 5696abhatia@williamblair.com Willow Miller, CFA+1 312 801 7885wmiller@williamblair.com Software Alinda Li+1 312 364 8241ali@williamblair.com The State of Agentic AI: The Four Key Questions Defining the Enterprise AIRace Contrary to the market narrative, we do not think software is dying, but the sector is evolving toincorporate the new AI opportunity. The transition and buildout of agentic AI offerings, alongwith unknowns regarding the new competitive dynamics, are leading to sentiment in the sectorbeing quite negative as investors have broadly placed software into the “too hard” bucket.There are many questions being asked about the future of sector and how it needs to evolve:some of these questions are noise, while others are critical. In this report, we narrow down thelist to the four most important questions we think are defining the enterprise AI race: 1. How big is the agentic AI market opportunity? 2. How far along is enterprise AI adoption? 3. Who has the right to win the new agentic AI enterprise TAM? (the most importantquestion, in our view) 4. What does an AI business model actually look like? We attempt to provide answers to these questions where we can, though clean answers mostlydo not exist at this early phase in the AI cycle. So, we provide frameworks, estimates, context,and our perspective on where consensus might be wrong and what to monitor going forwardas a guide to help investors navigate the AI turbulence. We also provide some thoughts on ourcoverage to conclude the report. Key highlights and takeaways include: •AI and agentic ROI is likely to come from displacing a portion of the $26 trillionwhite-collar labor market.We believe enterprise AI is more disruptive to the roughly $26trillion in white-collar compensation across advanced economies than to the $1.3 trillionglobal software budget (we estimate white-collar labor is about 20x larger than softwarespend).•We estimate the agentic AI TAM at about $4 trillion, roughly 3x the size of globalsoftware spending today.Applying a 40%-60% AI productivity benefit to the $26 trillionlabor pool and a 25%-35% AI/tech-vendor value capture assumption yields a $2.6 trillion-$5.4 trillion range, or about $4 trillion at the midpoint.•Enterprise AI adoption is a mile wide and an inch deep; most of the spending ramp isstill ahead.Roughly 50% of U.S. businesses have a paid AI subscription, but only 3% report"extensive" or "fully integrated" agentic deployments. We expect adoption to increase incoming years as enterprise readiness increases. Please refer to important disclosures on pages 21 – 23. Analyst certification is on page 21.William Blair or an affiliate does and seeks to do business with companies covered in its research reports. As aresult, investors should be aware that the firm may have a conflict of interest that could affect the objectivity ofthis report. This report is not intended to provide personal investment advice. The opinions and recommendationsherein do not take into account individual client circumstances, objectives, or needs and are not intended asrecommendations of particular securities, financial instruments, or strategies to particular clients. The recipient ofthis report must make its own independent decisions regarding any securities or financial instruments mentionedherein. William Blair We see three major players making a claim to the enterprise agentic AI TAM. This includes the frontier modelproviders, incumbent software vendors, and AI native startups. Each has advantages and challenges to overcome, whichwe walk through in detail.Incumbents have a credible right to win the workflows they already own, if they deliver quality AI offerings.Incumbent systems of record have the data, integrations, domain expertise, and trust to win the right to address theagentic TAM in workflows they already own. Their data provides context required to power enterprise AI.Net-new agentic workflows are likely a battle between frontier models and AI-natives.Coding, legal, and investmentbanking have no established system of record to defend, and the frontier model providers have already shown willingnessto move up the stack with Claude Code, Codex, Cowork, and Frontier. AI natives are also making progress here, but have toaddress long-term defensibility.The pricing model transition is a second-order question.If the incumbent has the right to win the agentic TAM with agood AI offering, the move from seat-based to consumption or outcome-based pricing is not existential; the risk is inproduct quality, not pricing mechanics.AI-era economics look different from SaaS, but not necessarily worse.We estimate 60%-70% gross margins as thelikely steady state versus about 80% for traditional SaaS margins. However, operating expense leverage and a materiallylarger revenue base get us to an EBIT margin around 30%, above today's 25% benchmark. Contents Question 1: How Big Is the Agentic AI Market Opportunity?........