AI Agent Use Reveals 50% Software Dominance, Exposing Untapped Vertical Markets

Edgen Stock·Feb 23 2026, 08:18

Share to

Share to

Copy link

Key Takeaways

A new study from AI research firm Anthropic reveals a significant imbalance in the application of AI agents. Software engineering workflows constitute nearly half of all usage, while major industries like healthcare, finance, and law remain virtually untouched. This gap between AI's advanced capabilities and its current deployment signals a major opportunity for startups to create specialized, high-value solutions in these underserved vertical markets.

Extreme Market Imbalance: An Anthropic study shows software engineering commands 49.7% of all AI agent tool usage, while major sectors like healthcare and law each represent less than 2% of the market.
Capability vs. Deployment Gap: AI models possess capabilities for complex tasks requiring nearly five hours of human effort, yet the longest user sessions last only around 42 minutes, highlighting a lag driven by user trust and product maturity.
The Next Unicorn Frontier: Industry leaders from Y Combinator and Box predict the market could support 300 new vertical AI unicorns by building specialized agents for these empty sectors, an opportunity potentially ten times larger than the SaaS boom.

Software Engineering Commands 49.7% of AI Agent Use

A comprehensive study by AI firm Anthropic has revealed a stark concentration in the use of AI agents, with software engineering applications dominating at 49.7% of all tool calls on its API. This figure overshadows the activity in other key industries, where adoption is minimal. The data shows healthcare applications at just 1%, law at 0.9%, and education at 1.8%. In total, 16 distinct vertical industries outside of software engineering show single-digit market share, indicating they are not saturated markets but rather almost entirely new frontiers. This lopsided distribution is attributed to developers being natural early adopters and the lower technical barriers for coding applications compared to highly regulated and data-sensitive fields.

AI Capabilities Outpace User Trust, Creating an Opportunity

The study uncovers a critical "deployment lag," where AI models are far more capable than users currently allow them to be. For example, models like Claude can solve problems that would take a human expert nearly five hours, but the 99.9th percentile of actual user sessions lasts only about 42 minutes. This gap represents a clear product development opportunity. Evidence shows user trust is building steadily. From October of last year to January, the 99.9th percentile session duration nearly doubled from under 25 minutes to over 45 minutes. Furthermore, internal data showed Claude Code's success rate on challenging tasks doubled between August and December, as the average number of required human interventions per session fell from 5.4 to 3.3. This demonstrates that as users become more familiar with the tools, they grant them greater autonomy.

The Path to 300 Vertical AI Unicorns Is Now Clear

This market imbalance has led prominent tech figures, including Y Combinator's Garry Tan and Box's Aaron Levie, to predict the rise of 300 vertical AI unicorns. The opportunity lies in building defensible businesses by developing AI agents that can navigate the proprietary data, regulatory constraints, and complex workflows of specific industries. While anyone can create a generic AI tool, few can master the intricacies of medical billing, legal discovery, or construction permitting. This strategy mirrors the rise of SaaS, which produced over 170 unicorns in the last two decades. The vertical AI opportunity could be ten times larger, as these new systems have the potential to replace not just software, but also the human operators, fundamentally reshaping enterprise cost structures and productivity.