Trending agent use cases this year, with numbers
Where businesses actually use AI agents in 2026, from coding and support to sales research, based on public surveys and usage data.
Updated 2026-11-19. "(source N)" refers to the numbered list at the end.
Where the numbers come from
Agent hype is easy to find. Hard numbers are not. This piece sticks to three public sources: LangChain's survey of people who build agents, Anthropic's Economic Index, which studies how Claude is used, and Stanford's 2026 AI Index (source 7).
Each has limits. LangChain surveyed 1,340 people between November and December 2025, published in June 2026, and most of them work in tech (source 1). Anthropic only sees its own traffic (source 2). Stanford covers the whole field but mostly through benchmarks (source 3). Read the numbers as signals, not a census.
1. Coding is still number one
Software work is the clearest agent success so far. In Anthropic's March 2026 report, coding was the most common use on both its chat app and its API. Computer and maths tasks made up 35% of conversations on Claude.ai (source 2). The same report says coding is moving from chat, where a person works alongside the model, to the API, where tools like Claude Code run more of the work directly (source 2).
The benchmarks back this up. Stanford reports agent success on Terminal-Bench, a test of real command-line tasks, climbed from 20% to 77.3% in a year (source 3).
Vendors are building for hand-off, not just help. GitHub's cloud agent takes a task, works in its own cloud environment, runs tests and linters, and can open a pull request for review (source 5). Claude Code can be set to run on a schedule, such as reviewing pull requests every morning, or take bug reports straight from Slack (source 6).
2. Customer service
Support is the top agent job in LangChain's survey, named by 26.5% of respondents (source 1). It is a natural fit: lots of text, lots of small decisions, and a clear escalation path to a human.
Anthropic's previous report also found customer service tasks such as billing questions were common on its API (source 2).
3. Research and data analysis
Close behind is research and data analysis, at 24.4% in LangChain's survey (source 1). Think of an agent that reads twenty supplier price lists and builds a comparison, or one that pulls last month's sales and writes up what changed and why.
At large firms with 10,000 or more staff, internal productivity tops the list at 26.8% (source 1). Examples include finding answers in company documents or preparing meeting briefs.
4. Sales outreach
This is the fastest riser in Anthropic's data. Sales and outreach automation, covering lead research, enriching customer records and drafting cold emails, was one of two API uses whose share at least doubled in three months (source 2). The other was market and trading monitoring.
For a small business, this is often the easiest win: an agent that researches each new lead and drafts a first message for a person to check and send.
5. Security work
Security is a growing niche. Stanford reports AI agents solved cybersecurity challenges 93% of the time, up from 15% in 2024 (source 3). Vendors are responding. Google positions Gemini 4 Carbon for cybersecurity defence among other hard tasks (source 4).
6. Health and admin paperwork
Outside tech, note-taking and paperwork stand out. Stanford cites some AI note tools that cut doctors' note-writing time by up to 83% (source 3). The same pattern, an agent that listens, drafts and files, applies to many service businesses.
What is holding agents back
Adoption is real but uneven. In LangChain's survey, 57.3% had agents in production. That rose to 67% at the largest firms and fell to 50% at firms under 100 people (source 1).
The main blocker is quality, cited by about a third of respondents, ahead of speed at 20%. Cost was mentioned less than in past years (source 1). Teams that ship agents also tend to watch them closely: 94% of those in production have some form of monitoring (source 1).
The jobs question
There is a human side to the coding trend. Stanford reports that employment among software developers aged 22 to 25 has fallen by nearly 20% since 2024, while older colleagues' headcount grew (source 3). Agents are taking on the routine tasks that junior staff used to learn on.
What this means for you
Stanford also found that generative AI reached 53% of the population within three years (source 3). Your customers and staff are already using these tools. The question is whether you put them to work on purpose, with checks in place.
- If you write code, a coding agent is the most proven place to start.
- If you run support, start with drafting replies for a person to approve, not sending them.
- If you sell, try lead research and first drafts before anything that contacts customers on its own.
- Whatever you try, measure quality first. It is what sinks most projects.
Sources
Written by Agents Lookup in our own words. Facts come from these pages:
- 1. LangChain: State of Agent Engineering: www.langchain.com/state-of-agent-engineering
- 2. Anthropic: Economic Index report, Learning curves (March 2026): www.anthropic.com/research/economic-index-march-2026-report
- 3. Stanford HAI: Inside the AI Index, 12 takeaways from the 2026 report: hai.stanford.edu/news/inside-the-ai-index-12-takeaways-from-the-2026-report
- 4. Google AI Studio: Gemini models: aistudio.google.com/docs/models
- 5. GitHub Docs: About Copilot cloud agent: docs.github.com/en/copilot/concepts/agents/coding-agent/about-coding-agent
- 6. Claude Code Docs: Overview: code.claude.com/docs/en/overview
- 7. Stanford HAI: The 2026 AI Index Report: hai.stanford.edu/ai-index/2026-ai-index-report