The Latest Agentic AI Models and Use Cases | 2026
Agentic AI Models: What They Are and How Businesses are Leveraging Them?
If you've been hearing the term "agentic AI" and wondering what it actually means, you're not alone. It sounds technical, but the idea is surprisingly straightforward. And once you understand it, you'll see why companies like JPMorgan Chase, Klarna, and LinkedIn are betting billions on it.
This article breaks down what agentic AI models are, which models power them, real examples of companies using them today, and most importantly, how your business can put them to work.
What Is Agentic AI?
Most AI tools you've used work like a very smart search box. You type something in, it gives you an answer, and that's the end of it. Every time you want something new, you have to ask again.
Agentic AI is different. Instead of just answering one question, an agentic AI system is given a goal and then figures out the steps needed to reach it, takes action, checks its own work, and adapts along the way. It doesn't wait to be told what to do next.
Think of it like the difference between a calculator and an accountant. A calculator gives you results when you push buttons. An accountant understands your goal (pay less tax, grow profit), gathers the right information, runs the numbers, and comes back with a plan.
Four things define an agentic AI system:
- Goal-setting: You tell it what you want, not how to do it
- Multi-step reasoning: It breaks the goal into tasks and works through them
- Tool use: It can search the web, read documents, call APIs, send emails, write code, and more
- Autonomous action: It executes, checks results, and adjusts without waiting for you
This shift, from AI that answers to AI that acts, is why 2024 and 2025 have been called the beginning of the "agentic era."
The Agentic AI Models
Not every AI model is built for agentic work. Agentic tasks require strong reasoning, reliable tool use, and the ability to stay focused over many steps without drifting off course. Here are the models leading the field today.
OpenAI's o1 and its successor GPT-5 are among the strongest models for deep, structured reasoning. They excel at breaking complex problems into steps, following multi-step instructions, and using tools reliably. Investment banks and law firms are heavy users because of their precision on technical tasks. GPT-5 is described as a "unified system" that routes tasks to the right internal model in real time, making it one of the most versatile options for enterprise agentic work.
Anthropic's Claude models, particularly Claude Sonnet 4.6 and Claude Opus, are specifically optimized for agentic workflows. Anthropic describes Claude as capable of working "autonomously for hours" on complex tasks. Claude is the model behind several real-world deployments covered later in this article, including Intercom's customer support agent and DoorDash's voice assistant. It is particularly strong at following nuanced instructions and handling tasks that require careful judgment.
Google's Gemini 2.5 is a "thinking model." It spends time reasoning through a problem before responding, rather than answering immediately. It's especially strong at handling multimodal tasks (text, images, audio, and video together) and at processing very long documents. Teams that need agents to work across different types of content, say, analyzing both a PDF contract and a data spreadsheet, often reach for Gemini.
DeepSeek, developed in China, emerged as a major surprise in 2025. Its V3.2 model delivers near-frontier reasoning quality at significantly lower cost, making it attractive for teams that want powerful agentic capabilities without the price tag of the big US labs. It is open-source, meaning businesses can self-host it for full data privacy, a meaningful advantage for industries with strict compliance requirements.
Kimi K2 is an open-source model that topped the agentic benchmark Tau2-Bench Telecom, which specifically measures how well an AI handles customer service scenarios involving tools and APIs. It is the highest-scoring open-source model for action completion and tool selection, making it a strong candidate for businesses building customer-facing agents.
The Platforms That Make Agentic AI Models Work
Models are the brains, but they need infrastructure to act in the real world. These are the main platforms businesses use to build and deploy agentic AI.
- Salesforce Agentforce is one of the most widely adopted enterprise platforms, built around its Atlas Reasoning Engine. It lets companies deploy agents that work within Salesforce's ecosystem, connecting to CRM data, customer records, and communication tools. Priced at $2 per conversation, it has been adopted by over 12,000 companies.
- Crescendo is a customer experience (CX) platform that puts agentic AI to work
specifically for customer service operations. It uses all above agentic AI models and advanced Large Language Models (LLMs) for providing fast, context-aware customer service and analyzing the customer interaction data.
Rather than requiring a company to build an agent from scratch, Crescendo handles the integration work, connecting to a business's existing apps and tools, and wraps its AI agents with a human-in-the-loop layer, meaning a real person can step in when the AI hits the edge of its confidence.
Its customers span retail, e-commerce, healthcare, connected devices, and SaaS. Multiple case studies (covered in the next section) show consistent results: 50 to 99%+ instant resolution rates, high CSAT scores, and meaningful cost savings.
- Microsoft AutoGen is an open-source framework for building multi-agent systems, where multiple AI agents collaborate with one orchestrating the others. It has deep integration with Microsoft Azure and has become one of the most widely used frameworks among developers building custom agentic workflows.
- Amazon Bedrock is AWS's managed platform for running AI models (including Anthropic's Claude) in production. It handles the infrastructure so companies can focus on building the agent logic. DoorDash, for example, uses Bedrock to power its voice support system.
- LangGraph is a developer framework that lets teams build agents as structured workflows, giving them fine-grained control over how the agent reasons and makes decisions. This is important for regulated industries where predictability matters.
8 Real Case Studies: Companies Using Agentic AI Models Right Now
These are not hypothetical examples. These are real deployments with real results and real lessons.
1. JPMorgan Chase: 450+ Use Cases Across an Entire Bank
Source: McKinsey Interview, October 2025 | AI News, December 2025
JPMorgan Chase is executing what may be the most ambitious agentic AI deployment in corporate history. With an $18 billion annual technology budget and over 450 AI use cases in production, the bank has built a proprietary platform called LLM Suite, used daily by more than 200,000 of its employees.
Agentic AI agents at JPMorgan now create investment banking presentations in 30 seconds (work that previously took junior analysts hours), draft M&A memos, automate trade settlement, detect fraud in real time, and equip call center agents with instant context-aware answers. Portfolio managers have cut research time by up to 83%. The bank estimates $1.5 to $2 billion in annual business value from its AI programs, and it holds the number one global AI maturity ranking among banks according to the Evident AI Index 2025.
CEO Jamie Dimon has said: "It affects everything, risk, fraud, marketing, idea generation, customer service. And this is the tip of the iceberg."
2. Rachio: Scaling Support for 1 Million Users with Crescendo
Source: Crescendo Case Study: Rachio
Rachio, a smart home irrigation company serving over one million customers, was struggling to maintain support quality as its user base grew. Its Head of Customer Support Operations described their previous AI accuracy as sitting at just 20%. After deploying Crescendo's agentic AI platform, Rachio reached 95% to 99% AI accuracy within weeks, achieved an overall CSAT of 95%, and now handles over one million annual tickets, all managed by a single customer service leader. The Crescendo system integrates agentic AI for routine resolution with seamless human escalation for edge cases.
3. LinkedIn: Automating the Hiring Workflow
Source: LinkedIn Engineering Blog
LinkedIn's Hiring Assistant is a multi-agent system designed to help corporate recruiters work faster. One "supervisory" agent orchestrates a set of specialized sub-agents, each responsible for a different piece of the recruitment process: writing job descriptions, sourcing candidates, drafting outreach messages, generating screening questions, and ranking applicants. The orchestrator breaks down a recruiter's request into tasks, assigns them to the right sub-agent, and assembles the results.
Each recruiter gets their own dedicated agent instance, meaning the system learns the preferences and patterns of the individual it works with over time. This is a real example of a multi-agent architecture being used in production at enterprise scale.
4. DoorDash: Voice Support for Delivery Workers
Source: AWS Case Study, 2024-2025
DoorDash uses Amazon Bedrock with Anthropic's Claude to power a voice AI agent that handles hundreds of thousands of support calls daily from Dashers (its delivery workers). The agent maintains conversational latency at or below 2.5 seconds, fast enough to feel like a real phone call, and has reduced escalations to human agents by several thousand per day. For a company running a gig economy at scale, with support needs that spike unpredictably, having an agent that can absorb volume without delay is operationally critical.
5. eBay: Smarter Product Recommendations
Source: eBay Engineering: Mercury Platform
eBay built an internal agentic AI platform called Mercury to power its product recommendation engine across a marketplace of two billion items. The platform uses retrieval-augmented generation (RAG) to combine the AI's output with real-time inventory data, so recommendations aren't just contextually smart, they're actually available to buy. A Listing Matching Engine converts the AI's text-based suggestions into live listings. The platform also includes internal models to detect and block prompt injection attacks, a sign of how seriously production-grade agentic systems need to address security.
6. John Deere: AI in the Field
Source: John Deere See and Spray Technology
John Deere's See and Spray technology uses computer vision and agentic workflows to distinguish crop plants from weeds in real time, applying herbicide only where needed. The result is a reported 70% reduction in chemical usage. This is one of the clearest examples of agentic AI working outside of software, acting autonomously in the physical world with real environmental and cost consequences.
7. EVPassport: 70% AI Resolution for EV Charging Support
Source: Crescendo Case Study: EVPassport
EVPassport, a fast-growing EV charging network, needed to deliver 24/7 customer support across a geographically distributed infrastructure. After deploying Crescendo's agentic AI platform, 70% of customer inquiries are now handled entirely by AI, with human agents responding in under 30 seconds for escalated cases. EVPassport's AI Lead summed it up: "We are the most reliable EV charging network in the United States and the most advanced in using AI to deliver support, and Crescendo is making that possible."
8. Klarna: Customer Service at Scale (and What They Learned)
Source: Klarna Press Release, 2024 | Case Analysis
In January 2024, Klarna deployed an AI assistant built on OpenAI across 23 markets and 35 languages. In its first month alone, the agent handled 2.3 million customer conversations, doing the equivalent work of 700 full-time employees. Average resolution time dropped from 11 minutes to under 2 minutes, repeat inquiries fell by 25%, and customer satisfaction scores matched those of human agents. Klarna projected $40 million in profit improvement for 2024.
But the story has an important second chapter. By mid-2025, Klarna's CEO admitted that the push to go AI-only had compromised the quality of customer experience for complex or emotional cases. The company reversed course, not by abandoning AI, but by adopting a hybrid model. AI handles routine, high-volume requests while human agents manage nuanced situations. Today the system supports the equivalent of 800 full-time employees' worth of work, more than when they tried to go fully automated.
The lesson: Agentic AI works best as a partner to humans, not a replacement for human judgment.
How Businesses Can Use Agentic AI Models to Ease Operations
You don't need to be JPMorgan or eBay to benefit from agentic AI. The technology is increasingly accessible through platforms that handle the complexity for you. Here's where it makes the most practical difference across common business functions.
1. Customer Support
This is currently the most mature and proven use case. Agentic AI can handle order tracking, refund requests, account questions, password resets, and product troubleshooting, 24 hours a day, in multiple languages, without a queue. The key is pairing it with a clean escalation path so that emotionally sensitive or complex cases reach a human quickly.
Platforms like Crescendo are built specifically for this, handling the integration with your existing tools (CRM, order management, ticketing) and maintaining human oversight so the AI operates within boundaries you define. The case studies above show consistent results across industries: 50 to 99% instant resolution, CSAT scores in the 85 to 99% range, and lean support teams that punch well above their weight.
For larger enterprises building custom solutions, Salesforce Agentforce and Amazon Bedrock with Claude are well-tested options. The Klarna case shows both the upside and the limits: AI can handle enormous volume efficiently, but hybrid models consistently outperform fully automated ones for customer satisfaction.
2. Sales and CRM
AI agents can monitor your CRM, identify leads that are going cold, draft personalized outreach emails, summarize account history before a sales call, and flag upsell opportunities based on usage data. JPMorgan's wealth management agents helped advisors respond to client inquiries during market volatility with personalized, portfolio-specific messages, contributing to a 20% increase in gross sales.
For smaller businesses, Salesforce Agentforce connects directly to Salesforce CRM data. HubSpot has also been building agentic capabilities into its platform, making this increasingly accessible without a large technical team.
3. HR and Recruiting
As LinkedIn's Hiring Assistant demonstrates, agentic AI can handle significant portions of the recruiting workflow: sourcing candidates, drafting job descriptions, screening applications, and scheduling interviews. For HR teams stretched thin, this means faster time-to-hire and more consistent candidate experiences.
Beyond recruiting, AI agents can answer employee questions about benefits, policies, and onboarding materials. Equinix, using Moveworks, achieved 68% deflection of employee IT and HR requests, meaning more than two-thirds of questions were answered without a ticket ever being created.
4. Finance and Compliance
Agentic AI is particularly powerful in finance because it can continuously monitor data rather than checking it periodically. Use cases include automated invoice processing, expense reconciliation, flagging anomalies in transactions, running Know Your Customer (KYC) checks, and monitoring for regulatory compliance in real time.
Mastercard's Agent Pay framework, launched in 2025, uses AI agents to track transactions and combat online fraud, binding credentials to their origin in real time to address a $750 million fraud problem. For smaller finance teams, the same principles apply at a smaller scale: agents that monitor and flag, with humans reviewing decisions.
5. Marketing
Marketing teams can use agentic AI to research competitors, draft content, personalize email campaigns, analyze campaign performance, and suggest optimizations, all without someone manually pulling reports and writing briefs. An agent can pull the latest data from your analytics platform every week, compare it to the prior period, and prepare a plain-English summary for your team.
Generative agents can also maintain brand tone consistency across high-volume content production, a growing challenge as content demands increase.
6. IT and Software Development
Coding agents are one of the most validated use cases today. Amazon reportedly saved the equivalent of 4,500 developer-years of work by using AI coding agents for routine software upgrades. Goldman Sachs added AI coding agents to its engineering team. JPMorgan's developers use AI to automate 40% of research tasks.
For businesses of any size, tools like GitHub Copilot, Cursor, and Claude Code can give development teams a significant productivity multiplier, writing boilerplate code, reviewing pull requests, catching bugs, and generating tests. This is arguably the fastest way to see a return on agentic AI investment today.
7. Supply Chain and Operations
Agentic AI can monitor supplier data, track shipment status, flag delays before they cascade, and recommend rerouting or restocking decisions in real time. John Deere's field-level AI is an extreme example, but the same logic applies in warehouses and logistics operations: agents that observe, decide, and act faster than any human dashboard review cycle.
A Note on Getting Started
The companies seeing the best results aren't the ones that deployed AI everywhere at once. They started with one high-volume, well-defined problem, usually customer support or an internal workflow, measured the results carefully, and expanded from there.
A few principles that consistently show up in the successful deployments above:
Start with a clear boundary. Define exactly what the agent is responsible for and what should always go to a human. Klarna's costly lesson was removing that boundary too aggressively.
Choose the right model for the job. Claude tends to excel at nuanced customer interactions and long-horizon tasks. GPT-5 is strong for structured analytical work. Gemini handles multimodal tasks well. Open-source options like DeepSeek are worth considering if data privacy or cost is a constraint.
Use platforms, not just models. Building an agent from scratch requires engineering resources. Platforms like Crescendo, Salesforce Agentforce, and Amazon Bedrock let you focus on the business problem rather than the infrastructure.
Keep humans in the loop, at least at first. The most effective agentic systems today are hybrid ones. Human oversight isn't a limitation; it's what makes the system trustworthy enough to scale.
Agentic AI is no longer a future technology. It's running in production at some of the world's largest companies, and it's increasingly accessible to businesses of every size. The question is no longer whether to use it; it's where to start.
Resolve 90% of Support Tickets Automatically with AI Agents
Keep Reading