AI Agent Examples: 12 Ways Businesses Use Agents Today
Businesses use AI agents today to answer customer chats and phone calls, process insurance claims and freight orders, handle prior authorizations and HR questions, upgrade and migrate code, write database queries, triage security alerts and run bank operations. Each of the 12 examples below comes from a company announcement, a vendor case study or published research, with the result it reports and the date. In most of them, a person still makes the final or high-risk decision.
Vendor case studies are marketing material, and the figures in them are the publisher's own. We give each figure as the source states it, say who published it, and note where a company later changed course. All sources were read on October 5, 2026.
What counts as an AI agent?
An AI agent is software in which a language model decides which steps to take and which tools to use to finish a task. Anthropic's guide of December 19, 2024 puts it this way: "Agents are systems where LLMs dynamically direct their own processes and tool usage." A chatbot that only answers questions falls outside that definition. One that looks up an order and issues the refund fits it. Our article on agentic AI vs generative AI covers the distinction in more depth.
The label is used loosely. Gartner warned on June 25, 2025 of "agent washing", the rebranding of existing products "without substantial agentic capabilities", and estimated that "only about 130 of the thousands of agentic AI vendors are real."
Executives report wide adoption. In PwC's survey of 308 US executives, published May 15, 2025, 79% said AI agents were already being adopted in their companies. The functions where most were using or planning to use agents were customer service (57%), sales and marketing (54%) and IT and cybersecurity (53%).
Which businesses use AI agents, and what results do they report?
The table summarizes the 12 examples. Each is described below with its source.
| # | Use | Company | Reported result | Date |
|---|---|---|---|---|
| 1 | Customer service chat | Klarna | Two-thirds of chats in the first month | February 27, 2024 |
| 2 | Software support site | Salesforce | 83% of 32,000 weekly conversations resolved | January 16, 2025 |
| 3 | Utility phone line | PG&E | 67% of calls contained | May 25, 2026 |
| 4 | Insurance claims | Allianz | 80% less processing and settlement time | November 3, 2025 |
| 5 | Freight quotes and orders | C.H. Robinson | Orders accepted in under 90 seconds | April 16, 2025 |
| 6 | Prior authorizations | MUSC Health | 40% done without human involvement | March 7, 2025 |
| 7 | HR questions | IBM | 94% of common questions contained | August 21, 2025 |
| 8 | Code upgrades | Amazon | 4,500 developer-years saved | August 1, 2024 |
| 9 | Code migration | Nubank | 12× efficiency, 20× cost savings | November 14, 2024 |
| 10 | Data questions to SQL | Uber | Query writing from about 10 to 3 minutes | September 19, 2024 |
| 11 | Security alert triage | Microsoft | 6.5× true positives per analyst minute | November 17, 2025 |
| 12 | Bank operations | BNY | Contract review cut from four hours to one | December 12, 2025 |
1. Customer service chat (Klarna)
Klarna's assistant handles queries about refunds, returns, payments, cancellations and disputes. Klarna reported on February 27, 2024 "2.3 million conversations, two-thirds of Klarna's customer service chats" in its first month, work it equated to "700 full-time agents", with resolution in "less than 2 mins compared to 11 mins previously." In May 2025 the company started hiring human agents again. Its CEO told Bloomberg, as reported by Fortune on May 9, 2025, that "what you end up having is lower quality." Lyft said its own assistant, reported by TechCrunch on February 6, 2025, "reduced the average customer service resolution time by 87%."
2. Software support with escalation (Salesforce)
Salesforce runs its own agent on its help site. On January 16, 2025, it reported "about 32,000 customer conversations per week with a resolution rate of 83%", with "only 1% of customers needing to speak to a human." The agent handles routine tasks such as password resets and product questions, and escalations to people were cut in half.
3. Phone calls (PG&E)
The California utility uses a voice agent to authenticate callers, route calls, give outage updates and answer billing questions in English and Spanish. The vendor's case study, published May 25, 2026, reports "67% containment, 6% higher than the legacy IVR", "35,000 labor hours" saved and a "22% increase in CSAT for outage calls."
4. Insurance claims (Allianz)
Allianz's Project Nemo handles food spoilage claims under AUD 500 in Australia with seven agents for planning, security, coverage, weather, fraud, payout and audit. Allianz reported on November 3, 2025 an "80% reduction in claim processing and settlement time." A claims professional approves every payment: "By design, payout decisions are never automated."
5. Freight quotes and orders (C.H. Robinson)
The logistics company's agents read customer emails, quote prices, enter orders and book appointments. It announced on April 16, 2025 more than 3 million shipping tasks done by AI. Mark Albrecht, its VP for Artificial Intelligence, said: "Instead of waiting up to four hours for a person to get to their shipment in an email queue, over 5,200 customers are getting their loads accepted in under 90 seconds."
6. Prior authorizations (MUSC Health)
The South Carolina health system uses agents to request insurer approval for treatments and to manage patient check-ins. Becker's Hospital Review reported on March 7, 2025 that the agents "execute 40% of prior authorizations without human involvement", and that work taking "about 30 minutes" manually now takes about one. Denials go to a person.
7. HR questions (IBM)
IBM's AskHR agent handles routine HR requests, with human advisors taking complex cases. IBM's case study of August 21, 2025 reports a "94% containment rate of common questions", "more than 11.5 million employee interactions in 2024 alone" and a "40% reduction in the HR team's operational costs over the past four years."
8. Code upgrades (Amazon)
Amazon used an AI agent to move internal applications to a newer Java version. AWS wrote on August 1, 2024 that Amazon migrated "tens of thousands of production applications from Java 8 or 11 to Java 17", saving "over 4,500 years of development work" against manual upgrades, with performance gains "worth $260 million dollars in annual cost savings."
9. Code migration (Nubank)
The Brazilian bank delegated a large refactor of an eight-year-old, multi-million-line codebase to a coding agent, with engineers reviewing each change. The vendor's case study of November 14, 2024 reports "a 12x efficiency improvement in terms of engineering hours saved, and over 20x cost savings", and explains how each figure was calculated.
10. Data questions to SQL (Uber)
Uber's QueryGPT turns a question in plain English into a database query, using separate agents to pick the business domain, the tables and the relevant columns. Uber's engineering blog reported on September 19, 2024 that writing a query took about 3 minutes instead of about 10, and that in a limited release 78% of users said it saved them time.
11. Security alert triage (Microsoft)
Security teams use agents to sort user-reported phishing emails. A randomized controlled trial of Microsoft's Security Copilot Phishing Triage Agent, posted on November 17, 2025, found analysts with the agent achieved "up to 6.5 times as many true positives per analyst minute and a 77% improvement in verdict accuracy", and were "not prone to rubber-stamping the agent's malicious verdicts."
12. Bank operations and contract review (BNY)
BNY runs agents it calls "digital employees", which "handle everything from payment instruction validation to code security enhancements", according to an OpenAI case study of December 12, 2025. A contract review agent "Reduces legal review time by 75%, from four hours to one, across 3,000+ annual vendor agreements each year."
What do the examples that work have in common?
They share four traits: a narrow, high-volume task, a clear measure of success, a person on the high-risk decision, and a defined exit to a human. Those traits also make a sound checklist for choosing a first agent.
- Narrow tasks. Allianz automates claims under AUD 500. C.H. Robinson started with email orders from its largest customers.
- A measure fixed in advance. Containment, resolution time, hours saved and accuracy are all countable, and each source says what it counted.
- People on the decisions that matter. Allianz never automates payouts. MUSC sends denials to staff. Nubank's engineers review each change.
- A way out. Salesforce hands unresolved cases to people, and IBM runs a "two-tier support model."
Where agents disappoint, the sources point to the same gaps. Klarna's CEO tied its reversal to cost being "a too predominant evaluation factor". Stanford HAI's 2026 AI Index reports that agents "still fail roughly 1 in 3 attempts on structured benchmarks." And Gartner predicted on June 25, 2025 that "Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls."
What has Easital built in these categories?
Easital has built AI systems in four of these areas: phone calls, coding, store operations and research reporting. The descriptions below follow the published project pages, and no client is named.
- Phone calls. Calldone, built by Easital for a client in the United States, runs AI voice agents that answer business phone calls, book appointments, answer questions and qualify leads.
- Coding. Manob.ai, Easital's own product, includes an agentic code editor, a cloud sandbox with live preview and one-click deployment.
- Store operations. mAutomate, also Easital's own, builds an online store and drafts marketing and support replies. The AI proposes the work and the owner approves it.
- Research reporting. For a market-research agency in the Caribbean, Easital built an internal system that manages field interviews, records and transcribes them, analyzes the answers with AI and produces the report.
Key takeaways
- Sourced agent examples cluster in customer service, phone calls, claims, logistics, healthcare administration, HR, software engineering, data analysis, security and banking operations.
- Reported results include two-thirds of chats handled (Klarna), 80% faster claims (Allianz), 94% of HR questions contained (IBM) and 4,500 developer-years saved (Amazon).
- Most examples keep a person on payments, denials or code merges, and give customers a route to a human.
- Vendor case studies report their own results; read how each figure was counted.
- Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, so start narrow and measure.
Frequently asked questions
What is an example of an AI agent in business?
Allianz's Project Nemo is one: seven agents check coverage, weather, fraud and payout for small food spoilage claims, and a claims professional approves each payment. Allianz reported an 80% reduction in processing and settlement time on November 3, 2025.
What is the difference between a chatbot and an AI agent?
A chatbot answers questions. An agent decides which steps to take and uses tools to act, such as looking up an order, filing a claim or changing a booking. Many customer service bots now do both.
Which business functions use AI agents most?
In PwC's May 2025 survey, the functions where most companies were using or planning to use agents were customer service (57%), sales and marketing (54%) and IT and cybersecurity (53%).
Do AI agents replace employees?
The sources differ. Klarna equated its assistant to 700 full-time agents and later hired people again over quality. IBM and Salesforce describe agents handling routine requests while people take complex ones, and C.H. Robinson described its results as "3 million manual tasks our people didn't have to do."
How much does it cost to build an AI agent?
Clutch reports an average AI development project of $120,594.55 as of September 21, 2026, with $10,000 to $49,999 the most common band. The monthly model bill depends on volume and model choice. Our article on AI agent development cost gives the figures and a worked estimate.
Easital Technologies Ltd. designs, builds and runs AI agents, from a scoped first use case to a monitored production system. See AI agent development, AI automation services, voice AI agent development, AI chatbot development and our work.
Sources
All sources were opened and checked on October 5, 2026.
- Anthropic, "Building effective agents", December 19, 2024. https://www.anthropic.com/engineering/building-effective-agents
- Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", press release, June 25, 2025 (read through a reader proxy). https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
- PwC, "PwC's AI Agent Survey", May 15, 2025 (read through a reader proxy). https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html
- Klarna, "Klarna AI assistant handles two-thirds of customer service chats in its first month", press release, February 27, 2024. https://www.klarna.com/international/press/klarna-ai-assistant-handles-two-thirds-of-customer-service-chats-in-its-first-month/
- Fortune, report on Klarna's return to human customer service, citing Bloomberg, May 9, 2025. https://www.fortune.com/2025/05/09/klarna-ai-humans-return-on-investment
- TechCrunch, "Lyft's new AI customer assistant is powered by Anthropic's Claude", February 6, 2025. https://techcrunch.com/2025/02/06/lyfts-new-ai-customer-assistant-is-powered-by-anthropics-claude
- Salesforce, "Agentforce customer zero", January 16, 2025. https://www.salesforce.com/news/stories/agentforce-customer-zero/
- PolyAI, "How Pacific Gas and Electric saved 35,000 labor hours with PolyAI", May 25, 2026. https://poly.ai/customers/pge
- Allianz, "When the storm clears, so should the claim queue", November 3, 2025. https://www.allianz.com/en/mediacenter/news/articles/251103-when-the-storm-clears-so-should-the-claim-queue.html
- C.H. Robinson, "At C.H. Robinson, Artificial Intelligence Has Now Performed Over 3 Million Shipping Tasks", press release, April 16, 2025. https://www.chrobinson.com/en-us/about-us/newsroom/press-releases/2025/ai-performs-over-three-million-shipping-tasks/
- Becker's Hospital Review, "Prior authorizations, patient check-ins: MUSC Health's AI agents", March 7, 2025 (read through a reader proxy). https://www.beckershospitalreview.com/healthcare-information-technology/ai/prior-authorizations-patient-check-ins-musc-healths-ai-agents/
- IBM, "IBM AskHR" case study, August 21, 2025. https://www.ibm.com/case-studies/ibm-askhr
- AWS DevOps Blog, "Amazon Q Developer just reached a $260 million dollar milestone", Aytul Arisoy Cholkar, August 1, 2024. https://aws.amazon.com/blogs/devops/amazon-q-developer-just-reached-a-260-million-dollar-milestone/
- Cognition, "How Nubank refactors millions of lines of code to improve engineering efficiency with Devin", November 14, 2024. https://devin.ai/customers/nubank
- Uber Engineering, "QueryGPT: Natural Language to SQL Using Generative AI", September 19, 2024. https://www.uber.com/blog/query-gpt/
- James Bono, "Randomized Controlled Trials for Phishing Triage Agent", arXiv:2511.13860, November 17, 2025. https://arxiv.org/abs/2511.13860
- OpenAI, "BNY builds 'AI for everyone, everywhere' with OpenAI", December 12, 2025. https://openai.com/index/bny/
- Stanford HAI, "The 2026 AI Index Report", 2026. https://hai.stanford.edu/ai-index/2026-ai-index-report
- Clutch, "AI Pricing Guide", Anna Peck, updated September 21, 2026. https://clutch.co/developers/artificial-intelligence/pricing