10 agentic AI examples and use cases

Agentic AI marks a shift from simpler prompt-response systems to active autonomous collaborators. These systems can reason over multiple steps, plan actions, call external tools or APIs, and adapt based on feedback. They can handle complex, multi-step tasks with minimal human input.

Top companies across various industries are already deploying agentic AI for real impact. Financial institutions use AI agents to automate transaction analysis and compliance checks, e-commerce companies employ them for dynamic recommendations, and engineering teams rely on multi-agent systems for code review and test generation. In this blog, we will explore ten agentic AI examples and use cases in the real world. 

🚛 Delivery Hero’s data analyst

Delivery Hero built QueryAnswerBird (QAB), an AI-powered data analyst assistant, to enable employees to query, visualize, and discover business data without code. The solution consists of two components:

• The Text-to-SQL feature is designed to speed data access. It combines an LLM with retrieval-augmented generation (RAG) to access the company’s internal metadata, documentation, and SQL schemas. Key features include natural-language to SQL conversion, query syntax validation, and table/column usage guides. 
• The Data Discovery feature helps users explore data to derive business insights. It utilizes a vector store to manage data from the company’s Data Discovery Platform and the Log Checker, which manages user and event data from apps and websites at the log level. The feature includes information acquisition functions that help improve employees’ data literacy by explaining queries and tables and providing guidance on utilizing log data.

Here’s how it works: a user asks a question in Slack, QAB retrieves relevant internal data definitions and examples via vector search, crafts a compliant SQL query tailored to business logic, validates it, and returns the result.

🛍 eBay’s agentic platform for RecSys

Mercury is eBay’s internal agentic AI platform. It powers LLM-driven recommendation experiences on the marketplace and lets teams efficiently build and scale autonomous, goal-oriented AI workflows:

• The platform integrates RAG to combine LLM’s outputs with real-time, domain-specific data, ensuring recommendations stay accurate and current. 
• The Listing Matching Engine bridges the gap between LLM text outputs and eBay’s two-billion-item inventory by converting textual suggestions into relevant live listings. 

Additionally, the platform includes internal models to detect and prevent prompt injection attempts by malicious actors.

🚘 Uber’s enhanced Agentic RAG

Uber uses enhanced Agentic RAG (EAg-RAG) to improve the answer quality of its on-call copilot, Genie. To ensure near-human answer precision of the copilot, Uber adds AI agents at multiple stages: 

• Query Optimizer reformulates or breaks down ambiguous queries.
• Source Identifier narrows down the document set. 
• Post-Processor Agent deduplicates and orders the retrieved context. 

As a result, the share of acceptable answers increased by 27%, and incorrect advice was reduced by 60% compared to traditional RAG architecture.

🎙 Salesforce’s AI agent for event management 

Salesforce built the Ask Astro agent, which is integrated into the Salesforce Events mobile app. The event agent enables attendees to ask natural-language questions, get session recommendations, and manage their personal schedules. 

​​Ask Astro ingests structured event data, like session schedules, and unstructured FAQs, indexed using a hybrid search architecture to improve accuracy. For example, when a user asks, “Which sessions is Marc Benioff doing on Tuesday?” the agent identifies the topic, retrieves sessions, filters by time and speaker, and returns grounded answers.

💻 Google’s coding agent

Google developed Jules, a massively parallel asynchronous AI coding agent. It assists developers by autonomously performing common coding tasks – for example, it can fix bugs, write tests, and update dependencies. Jules runs tasks in the background: it creates a plan, executes it, runs tests, and reports back with completed pull requests. 

It can run multiple variations of tasks in parallel (e.g., testing different frameworks) and allows developers to review or merge the best outcomes. 

🔍 Anthropic’s agent for research

Anthropic’s Research feature uses multiple Claude agents to explore complex topics. The system is designed for open-ended, complex queries that benefit from multiple parallel explorations rather than a single linear pipeline. Here’s how it works:

• A lead agent takes a user’s research request, builds a strategy, and then spawns multiple sub-agents in parallel, each assigned a portion of the research task. 
• These sub-agents independently use tools – like web search and document analysis – and iterative “thinking” to explore different trajectories simultaneously. 
• The lead agent then collects, synthesizes, and filters the results.
• A final “citation agent” attaches proper source references. 

This approach showed performance gains compared to a single agent in information-heavy searches.

💼 Netguru’s sales assistant

Omega is Netguru’s internal AI agent that assists salesmen by automating repetitive tasks and guiding sales workflows. It integrates into tools the team already uses – like Slack or CRM – and helps with daily activities, including preparing call agendas, summarising transcripts, generating feature lists, tracking deal momentum, and reminding about follow-ups. 

Omega is built as a modular multi-agent system in which different roles collaborate: The Sales Agent analyses input, the Primary Agent executes tasks, and the Critic Agent reviews the output and provides feedback.

💳 Ramp’s agent for matching transactions

A fintech company, Ramp, built an AI agent that handles merchant classification corrections. The agent resolves classification requests in under ten seconds, saving hours of human work. Here’s how it works: 

• A user flags a misclassified transaction and supplies a corrected merchant name, website, and category.
• The system gathers rich context: the raw card-acceptor name, the merchant category code, receipt line-items, user memos, and existing merchant records. 
• The retrieval-augmented generation (RAG) system selects relevant existing merchant records.
• The agent “decides” whether to update an existing merchant, create a new one, or reassign the transaction.

✅ Airtable’s Field Agents

The Field Agents are AI-powered fields inside an Airtable base that can autonomously gather insights and generate content based on users’ data. Field Agents are built as an asynchronous event-driven state machine and consist of three main components: 

• Context Manager tracks guidelines, sessions, and event context.
• Tool Dispatcher invokes predefined actions. 
• LLM-powered Decision Engine reasons, plans, and chooses what to do next.

For example, when a user requests, “Summarize this project status,” the agent processes the event, uses the context to decide whether to call tools or respond directly, and then executes until it arrives at a final output.

📄 Moveworks’ document assistant

Moveworks built Brief Me, an AI-powered document assistant integrated into the Moveworks Copilot. It lets users upload files – like PDFs, Word docs, or URLs – and interact with them conversationally.

Brief Me helps with tasks like summarising long reports, comparing documents, extracting action items or trends, and generating new content based on the user’s data.

Evaluate AI agents with Evidently

The examples covered here show that agentic AI no longer exists only as demos, but as production deployments delivering measurable business value.

If you are building complex systems like AI agents, you need evaluations to make sure they work as expected – both during development and in production. That’s why we built Evidently. Our open-source library, with over 30 million downloads, makes it easy to test and evaluate LLM-powered applications, including AI agents.

We also provide Evidently Cloud, a no-code workspace for teams to collaborate on AI quality, testing, and monitoring, and run complex evaluation workflows. You can generate synthetic data, create evaluation scenarios, run adversarial tests, and track performance – all in one place.

Ready to test your AI agent? Sign up for free or schedule a demo to see Evidently Cloud in action. We're here to help you build with confidence!