Search for agentic AI examples and you mostly find hypotheticals: an agent that could plan your vacation, one that might run a warehouse. That makes it hard to tell what is working today from what is still a slide in a pitch deck.
This guide sticks to real deployments. Each of the 13 examples below links to the company's own announcement, case study or a named report, and states what the agent does, how much it decides on its own, where a person stays in control and what results were reported. Two of them did not go as planned, and those are the most useful lessons. If you need the basics first, read our agentic AI vs generative AI explainer.
What counts as an agentic AI example?
There is no single agreed definition. MIT Sloan (opens in a new tab) cites a research paper that describes AI agents as "autonomous software systems that perceive, reason, and act in digital environments," and contrasts them with chatbots, which answer questions rather than integrating with other software to complete tasks. For this list, a system had to pass four tests:
- It acts, not just answers. It reads from or writes to other systems, such as a CRM, an HR platform, a code repository or a phone line.
- It works in steps. It completes a multi-step task, deciding or following the next step based on what it found.
- It is in real use. It is deployed with customers or employees, or generally available, not only a demo.
- It is documented. A company announcement, case study or named report describes it.
Agentic AI examples at a glance
| Industry | Example | What the agent does | Human checkpoint |
|---|---|---|---|
| Customer service | Salesforce Help (Agentforce) | Answers and resolves support questions | Hands off about 4% of chats to engineers |
| Customer service | Klarna AI assistant | Handles customer service chats in 35+ languages | Human support kept available (2025 change) |
| HR | IBM AskHR | Answers HR questions and completes 80+ HR tasks | Advisors handle complex cases |
| HR / recruiting | LinkedIn Hiring Assistant | Sources candidates for recruiters | Recruiter gives feedback and decides |
| Software | GitHub Copilot coding agent | Turns an issue into a pull request | Developer reviews the PR |
| Cybersecurity | Microsoft Security Copilot agents | Triages phishing and data-loss alerts | Admins review and give feedback |
| Manufacturing | Siemens Eigen Engineering Agent | Writes and validates PLC control code | Engineers own system decisions |
| Banking | FIS financial crimes agent (with Anthropic) | Gathers evidence for anti-money-laundering cases | Investigator makes every decision |
| Healthcare | Tampa General Hospital (Hyro) | Handles contact-center calls and scheduling | Transfers to staff |
| Healthcare | WellSpan Health (Hippocratic AI) | Makes and takes patient calls | Clinical staff behind it |
| Science | Google AI co-scientist | Generates and ranks research hypotheses | Scientists test the ideas |
| Consumer | OpenAI ChatGPT agent | Completes browser tasks on a virtual computer | Asks permission before consequential actions |
| Retail | ChatGPT Instant Checkout | Bought products inside the chat | Scaled back in March 2026 |
Agentic AI in customer service
1. Salesforce: an agent on its own help site
Salesforce put its Agentforce agent on the Salesforce Help site in October 2024 and reported in April 2025 (opens in a new tab) that it had handled more than 500,000 conversations, resolving more than 84% of the questions that came through it. About 4% of conversations were handed to a human support engineer, a rate Salesforce says it keeps on purpose. A July 2025 update put the total above 1 million conversations.
The interesting details are in the design choices. Renewal questions are routed straight to a person, and the team studies every handoff to find gaps in what the agent knows. Salesforce credits clean data, nearly 740,000 pieces of help content, as the biggest factor in its results. That is a RAG problem as much as an agent problem.
2. Klarna: big early numbers, then a course correction
In February 2024 Klarna announced (opens in a new tab) that its OpenAI-powered assistant had held 2.3 million conversations in its first month, two-thirds of its customer service chats, doing "the equivalent work of 700 full-time agents." Klarna said resolution time fell from 11 minutes to under 2, repeat inquiries dropped 25%, and it estimated a $40 million profit improvement for 2024.
In May 2025, CEO Sebastian Siemiatkowski told Bloomberg, as reported by Fortune (opens in a new tab), that Klarna was recruiting human agents again so customers could always reach a person, saying cost had been "a too predominant evaluation factor" and that quality suffered. The lesson is not that the agent failed. It is that containment rate is not the only metric that matters to customers.
Agentic AI use cases in HR
3. IBM AskHR: an HR help desk that completes tasks
IBM's internal AskHR (opens in a new tab) goes past answering policy questions. It generates employment verification letters, processes vacation requests and helps managers start employee transfers and update org structures in SAP SuccessFactors. In 2025 IBM integrated it with watsonx Orchestrate, where a model classifies each request, routes it to a domain such as payroll or benefits, then triggers a task or writes an answer.
IBM reports a 94% containment rate on common questions, more than 11.5 million employee interactions in 2024, automation of more than 80 HR tasks and a 40% cut in HR operating costs over four years. The model is explicitly two-tier: AI handles routine requests, and human advisors take the complex ones.
4. LinkedIn Hiring Assistant: an agent for recruiters
LinkedIn calls Hiring Assistant its first AI agent. It takes on candidate sourcing, the slow part of a recruiter's day, and improves with each round of recruiter feedback. LinkedIn announced global availability in English (opens in a new tab) in September 2025 and said early users were saving more than 4 hours per role, reviewing 62% fewer profiles and seeing a 69% improvement in InMail acceptance rates. The recruiter still decides who moves forward.
Agentic AI in software development and IT security
5. GitHub Copilot coding agent
This is one of the clearest agentic AI tools to try yourself. You assign a GitHub issue to Copilot, and the coding agent (opens in a new tab) works in the background in its own development environment powered by GitHub Actions, then opens a pull request and asks you to review it. It handles bug fixes, new features, test coverage and documentation. It became generally available to paid Copilot subscribers in September 2025. The human checkpoint is the code review, which is the same control teams already use for human changes.
6. Microsoft Security Copilot agents
Security teams drown in alerts, so triage is a natural agent job. In March 2025 Microsoft announced six Security Copilot agents (opens in a new tab), with previews from April 2025. The Phishing Triage Agent in Microsoft Defender sorts real phishing threats from false alarms and explains its reasoning in plain language. Others triage data loss alerts, flag identity policy gaps and prioritize patches, which ship "with admin approval." Microsoft says it detected more than 30 billion phishing emails targeting customers in 2024, which shows the scale of the problem.
Agentic AI use cases in manufacturing
7. Siemens Eigen Engineering Agent
Most manufacturing examples online are still predictions. This one is shipping. Siemens' Eigen Engineering Agent (opens in a new tab), launched in April 2026, "plans, executes and validates" automation engineering tasks: it writes PLC control code, builds HMI screens and configures devices alongside Siemens' TIA Portal. A June 2026 update added reading electrical design files and generating whole TIA Portal projects from a plain-language machine description.
Siemens says more than 100 companies in 19 countries use it, including ANDRITZ Metals and Prism Systems, and reports 2 to 5 times faster execution than manual workflows. The agent refines its output against quality benchmarks and flags inconsistencies it cannot resolve, leaving engineers to make system-level decisions.
Agentic AI in banking and financial services
8. FIS and Anthropic: anti-money-laundering investigations
In May 2026 FIS announced (opens in a new tab) an agent built with Anthropic that gathers evidence across a bank's core systems, checks activity against known money-laundering patterns and surfaces the highest-risk cases. FIS says it will compress investigations "from hours to minutes" and cut false positives. BMO and Amalgamated Bank are the first banks building with it, and general availability is planned for the second half of 2026, so treat this as an early example.
The design is a good template for regulated work: "Investigators will remain in control of every decision," and every conclusion is meant to trace back to its source data.
Agentic AI use cases in healthcare
Healthcare adoption is concentrated in patient access, the phone lines and scheduling queues, rather than diagnosis. A July 2026 Becker's Hospital Review roundup (opens in a new tab) of health systems expanding agentic AI shows the pattern.
9. Tampa General Hospital: contact-center voice agents
Tampa General uses Hyro voice agents in its contact center. Within weeks of launch, it reported call abandonment falling from 34% to 14.9%, average wait times dropping from 6.2 to 2.4 minutes and scheduled appointments rising 21%. It is expanding the agents to end-to-end scheduling for more specialties.
10. WellSpan Health: inbound and outbound patient calls
WellSpan's Hippocratic AI voice agent, called Ana, handles more than 160,000 patient calls a month and is expanding across inbound, outbound, outpatient and inpatient care, with planned outreach for missed imaging, post-discharge follow-ups and chronic disease check-ins. Like most figures in this guide, these are vendor or health system claims, not independent audits.
11. Google AI co-scientist: agents for research
On the research side, Google's AI co-scientist (opens in a new tab), announced in February 2025, is a multi-agent system built on Gemini 2.0. A supervisor agent splits a research goal into tasks for six specialized agents that generate, critique, rank and refine hypotheses. In lab follow-ups, drugs it suggested for acute myeloid leukemia inhibited tumor viability in multiple cell lines. Scientists choose which ideas to test, so it is a collaborator, not an autonomous lab.
Agentic AI tools for consumers and commerce
12. ChatGPT agent
OpenAI launched ChatGPT agent (opens in a new tab) in July 2025. It works on its own virtual computer with a visual browser, a text browser, a terminal and connectors such as Gmail and GitHub, so it can do things like research competitors and build a slide deck, or plan and order ingredients for a meal. It asks for explicit permission before consequential actions such as purchases, and OpenAI says it is trained to refuse high-risk tasks such as bank transfers.
13. ChatGPT Instant Checkout: an example that was scaled back
In September 2025 OpenAI launched Instant Checkout, which let people buy products without leaving ChatGPT, starting with Etsy sellers. By March 2026 it had changed course. According to Modern Retail (opens in a new tab), people researched products in ChatGPT but did not buy there, Walmart reported conversion three times lower in-chat than when shoppers clicked through, and retailers wanted control of inventory, promotions and loyalty. OpenAI now lets merchants use their own checkout and focuses ChatGPT on discovery.
The technology worked. The use case did not fit how people wanted to shop. That is a useful reminder that "can an agent do this?" and "do customers want an agent to do this?" are different questions.
What the successful examples have in common
Look across these deployments and the same traits keep showing up. They are a practical checklist for choosing your own first use case:
- High volume, repetitive requests. Support chats, HR tickets, security alerts and patient calls all arrive by the thousands, so small time savings add up.
- Clear rules and good data. Salesforce credits its help content; IBM ties AskHR to payroll and local HR rules. An agent is only as good as what it can look up.
- A defined handoff to a person. Code review, investigator sign-off, admin approval or a transfer to staff. None of the 13 runs without one.
- Narrow tools and permissions. Each agent works inside one system or a few. The Model Context Protocol is one common way to wire those connections in a controlled way.
- A structured workflow underneath. Most are closer to the patterns in our AI agentic workflows guide, such as routing and evaluator loops, than to a free-roaming agent.
The reality check
For every documented success there are many projects that stall. In June 2025 Gartner predicted (opens in a new tab) that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. It also warned about "agent washing": rebranding AI assistants, robotic process automation and chatbots as agents "without substantial agentic capabilities."
Gartner estimates only about 130 of the thousands of agentic AI vendors are real.
The same release forecasts that at least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from 0% in 2024. Both things can be true: the technology is real, and much of what is sold under the label is not. When a vendor shows you an agent, ask the four questions from the start of this guide, and ask for the handoff rate, not just the automation rate.
If you would rather build than buy, our guide on how to build an AI agent walks through a no-code route and a tested Python agent loop.
The bottom line
The strongest agentic AI examples today are not general-purpose digital employees. They are focused agents doing high-volume, rule-bound work, such as support, HR requests, alert triage, code changes, control software and patient calls, with a clear point where a person takes over. Start there, measure quality as well as cost, and treat any vendor claim as a claim until you see it in your own data. For more on agents, models and tools, browse our AI guides.