A recent study on the practical applications of AI intelligent agents reveals an extremely imbalanced market landscape: software engineering dominates nearly half of the market, while more than a dozen vertical fields such as healthcare, law, and finance together account for only the other half, with each individual field accounting for less than 5%. This situation points entrepreneurs in the right direction—real opportunities lie not in already cultivated fields, but in those nearly untouched blue ocean markets.
A comprehensive study released by Anthropic shows that software engineering accounts for as much as 49.7% of AI intelligent agent tool calls on its API. In contrast, healthcare accounts for only 1%, law 0.9%, and education 1.8%. These are not saturated markets, but rather markets that barely exist.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/65a91b35-77c5-45c8-bae8-a601f2ac494e.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/65a91b35-77c5-45c8-bae8-a601f2ac494e.png"/>
The research also reveals a key finding: the actual capabilities of AI models have far exceeded users' trust levels in them. METR's capability assessment shows that Claude can solve tasks that would take a human nearly five hours to complete, but in practical applications, the 99.9th percentile of conversation duration is only about 42 minutes. This significant gap between capability and deployment represents a product opportunity that entrepreneurs can seize.
Garry Tan, president of Y Combinator, and Aaron Levie, CEO of Box, both believe that this pattern indicates the future will see the emergence of 300 vertical AI unicorn companies, corresponding to the more than 170 unicorns that emerged during the previous SaaS era. The scale of AI versions may expand tenfold, as they not only replace software but can also substitute operators.
<h2>Software Engineering Dominates, Vertical Fields Almost Blank</h2>
Anthropic's data shows that software engineering occupies half of all AI intelligent agent activities, while the other half is dispersed across 16 vertical fields, none of which exceed 9%. The market shares of healthcare, law, education, customer service, logistics, and other fields are all in single digits.
This distribution is not because these vertical fields do not need AI intelligent agents, but because relevant applications have not yet been truly developed. Software engineering dominates because developers are naturally early adopters of AI tools, and the technical barriers are relatively low.
In contrast, vertical fields such as healthcare and law involve proprietary data, regulatory constraints, and complex organizational processes. These seemingly obstructive factors actually constitute defensible competitive barriers. Anyone can build a generic wrapper, but few can deeply understand the specific workflows of medical billing, legal discovery, or building permits.
<h2>Capability Outpacing Trust in Deployment Gap</h2>
The &#34;deployment lag&#34; phenomenon revealed by the research is worth entrepreneurs' attention. The capabilities that models already possess far exceed the levels users are willing to let them operate at
From October 2025 to January 2026, the 99.9% percentile of conversation duration nearly doubled, increasing from less than 25 minutes to over 45 minutes. This growth remains steady across multiple model versions. This is not only an improvement in model capability but also a cumulative trust from users—users learn to collaborate with the agent through successive conversations.
Anthropic researchers, including Miles McCain, pointed out that from August to December, the success rate of Claude Code on the most challenging tasks for internal users doubled, while the average number of human interventions per session decreased from 5.4 to 3.3. This indicates that as users deepen their understanding of the agent's capabilities, they are willing to grant more autonomy.
The capability is already there, but deployment has not kept pace. This is not a problem but rather an opportunity for the product.
<h2>The Paradoxical Pattern of Trust Evolution</h2>
The research discovered a phenomenon in the evolution of user trust: experienced users automatically approve more sessions while also intervening more.
New users automatically approve about 20% of Claude Code sessions. After 750 sessions, this proportion increases to over 40%. However, new users intervene in only 5% of rounds, while experienced users have an intervention rate of 9%.
This is not contradictory. The research team explains that this reflects a shift in supervision strategy. Novices seek approval before each step, while experienced users adopt a model of delegation with post-hoc intervention, shifting from prior approval to active monitoring.
The research also uncovered an important safety feature: in complex tasks, Claude Code requests clarification more than twice as often as human intervention. The agent pauses to confirm when uncertain, rather than pushing forward blindly. Researchers believe that &#34;the autonomy exercised by the agent in practice is co-constructed by the model, the user, and the product. Claude limits its independence by pausing to ask questions when uncertain.&#34;
73% of tool calls involve human participation, with only 0.8% of operations being irreversible. The highest-risk deployment scenarios, such as API key extraction or autonomous cryptocurrency trading, primarily involve safety assessments rather than actual production environments.
<h2>Defensive Strategies for Vertical AI</h2>
Aaron Levie's vertical AI strategy reveals a path to building defensible enterprises: develop agent software that can access proprietary data; ensure the software genuinely solves real problems; leverage context to maximize intelligent output; and address a critical aspect often overlooked by most founders—driving change management for customers.
The last point is precisely why vertical AI is defensible. In vertical fields, navigating traditional workflows, regulatory constraints, and organizational friction is key to distinguishing defensible companies from generic wrappers.
The SaaS industry has grown tenfold every decade over the past few decades. Over the past 20 years, more than 40% of venture capital has flowed into SaaS companies, which have produced over 170 unicorns. The logic of vertical AI is similar: every SaaS unicorn has a corresponding vertical AI version waiting, and the scale of the AI version could be ten times larger, as it not only replaces software but also the operators
Researchers point out that policies requiring &#34;approval for every operation&#34; will stifle productivity gains without increasing safety. A better goal is to ensure that humans can monitor and intervene, rather than mandating specific approval workflows.
<h2>The Hiding Places of 300 Unicorns</h2>
The market landscape has become clear. The software engineering field has found its place, while 16 verticals including healthcare, law, finance, education, customer service, and logistics each hold single-digit market shares, waiting for someone to integrate domain expertise into intelligent agents.
Models have been able to work for five hours, but users only allow them to operate for 42 minutes. This gap indicates that the market is still in its very early stages, with a lot of work to be done, and many fields have yet to see even a minute of intelligent application.
Previously, 300 SaaS unicorns were born, and the next 300 vertical AI unicorns will emerge. Founders who choose a vertical field, build domain expertise into intelligent agents, and address change management issues will dominate the next decade of enterprise software.
Risk Warning and Disclaimer
The market carries risks, and investments should be made cautiously. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at one's own risk

CLOU

A study on the application of AI agents shows that software engineering occupies 49.7% of the market, while sectors such as healthcare, law, and finance each have a market share of less than 5%. This indicates that entrepreneurs should focus on the undeveloped blue ocean market. The research also points out that the capabilities of AI models far exceed user trust, presenting significant application potential. Executives from Y Combinator and Box believe that there could be 300 vertical AI unicorn companies in the future, with the market size potentially expanding tenfold

- A recent study on AI applications reveals a significantly imbalanced market, with software engineering dominating at nearly 50%, while vertical fields like healthcare and law barely contribute 5% collectively.  
- This indicates untapped opportunities for entrepreneurs in under-explored markets, as AI's capabilities exceed current user trust and engagement.  
- Experts predict the emergence of 300 vertical AI unicorns, reflecting the potential for AI systems to expand significantly beyond software replacement, addressing real-world challenges within these fields.

Wallstreetcn

Global X Cloud Computing ETF

VanEck Semiconductor ETF

ISHRS S&P Glb It

iShares Semiconductor ETF

iShares Expanded Tech Software Sector ETF

SPDR S&P Software

Franklin Exponential Data ETF

Invesco Semiconductors ETF

SPDR S&P Semicon

Proshares Big Data Refiners ETF

The enormous "blue ocean market for intelligent agents": software programming accounts for half, while medical, financial, legal, and other sectors are "few and far between."