Don't Feed Your AI Agent the ERP—Let RPA Do the Typing

Everyone's telling you to replace your RPA bots with AI agents. They're wrong. I've seen too many teams rip out their deterministic, rule-following bots and replace them with a chatty agent that can't even open the right screen in your ERP without hallucinating a field name. The truth is, agentic AI is brilliant at thinking, but it's terrible at doing — and doing is exactly what RPA was built for.

Let me walk you through a realistic scenario. Imagine you're an operations manager at a mid-sized insurance firm. You've got a claims process that involves pulling data from a legacy claims system, cross-referencing it with a policy database, and then updating a customer relationship management (CRM) tool. Today, that's done by a team of five people, manually, every day. You've heard about AI agents and you're tempted to let one take over the whole flow. But here's the catch: your claims system is ancient, with no API, and the only way to get data in or out is by mimicking a human clicking through the screens. That's RPA territory, not agent territory.

The Misconception: Agents Can Do Everything

Here's the thing about AI agents: they're designed to understand natural language, reason, plan tasks, and call tools to achieve a goal (IBM). But when it comes to actually interacting with a legacy system that has no API, they're lost. They can't click a button or type into a field the way a human does. That's where RPA shines. RPA follows predefined, rule-based scripts to perform repetitive tasks like copying data between applications and filling forms (Wikipedia). So, the moment your automation needs to touch a system that isn't API-friendly, you need RPA in the mix.

The Hybrid Stack: Agents as the Brain, RPA as the Hands

The dominant emerging architecture layers AI agents for reasoning over an RPA execution layer, so agents drive RPA rather than replace it (Wikipedia). In practice, that means the agent decides what needs to be done — say, "check this claim, calculate the payout, and update the CRM" — and then it hands off each step to an RPA bot that actually does the work. UiPath calls this 'Agentic Automation,' where RPA serves as the execution layer that turns the plans and reasoning of AI agents into tangible actions (UiPath). It's a beautiful division of labor: the agent thinks, the bot acts.

Walking Through the Claims Scenario Step by Step

Let's see how this plays out in your insurance firm. You set up an AI agent that receives an email with a new claim. The agent reads the email, extracts the claim number, and decides the first step is to pull the claim details from the legacy system. It then calls an RPA bot, which logs into the legacy system, navigates to the claim screen, and copies the data — a task it can do 24/7 without error (UiPath). The bot returns the data to the agent, which now has to calculate the payout. The agent might call a separate tool, like a pricing API, or even ask another agent to handle a complex valuation. Once the payout is calculated, the agent instructs another RPA bot to update the CRM and send a confirmation email. Throughout this, the agent is using feedback loops — including human-in-the-loop checkpoints — to ensure accuracy (IBM). If a claim looks suspicious, the agent can flag it for a human to review before proceeding.

Why RPA Isn't Going Anywhere

This hybrid approach isn't just a nice-to-have; it's the only way to scale. The global RPA market was estimated at $4.68 billion in 2025 and is projected to reach $35.84 billion by 2033 (Grand View Research). That's not a dying market; that's a market that's growing because RPA is the execution layer for all this new AI. And it's not just insurance — RPA is widely used across finance, healthcare, and manufacturing, and in functions like HR, customer service, and IT (UiPath). The BFSI sector alone accounted for about 37% of the RPA market in 2025 (Grand View Research). So if you're in banking or insurance, you're in the sweet spot.

The Real Risk: Overcomplicating Your Stack

Now, I'm not saying you should ditch agents altogether. Agents bring something RPA can't: the ability to handle variability and make decisions. But the risk is overcomplicating your architecture. You don't need an agent to do every single task. A lot of what you automate is high-volume, repetitive, rule-based work — and RPA excels at that (UiPath). The mistake I see is teams trying to make an agent do everything, including the parts that are pure execution. That's a recipe for brittle, unpredictable automation. Instead, you should use patterns like sequential task chains, parallel agent swarms, and human-in-the-loop checkpoints (Wikipedia) to build a workflow that's both smart and reliable.

My Recommendation: Start with an Agent That Calls RPA

If you're just starting your AI automation journey, don't try to replace your RPA bots with agents. Instead, put an agent on top of your existing bots. Let the agent handle the thinking — triaging requests, planning the sequence of steps, and deciding what needs human input. Let the RPA bots handle the doing — logging into systems, copying data, filling forms. This way, you get the best of both worlds: the adaptability of AI and the reliability of RPA. And remember, the biggest barriers to adoption aren't technological; they're organizational change and human oversight (Wikipedia). So invest in training your team and setting up clear human-in-the-loop checkpoints, and you'll be well on your way to a future where agents and bots work together, not against each other.

Bottom Line

The single best move you can make is to layer an AI agent on top of your existing RPA bots — let the agent reason and plan, and let the RPA bot execute. That's the hybrid stack that's winning in the real world.

Sources

• Robotic process automation (Wikipedia) - https://en.wikipedia.org/wiki/Robotic_process_automation
• UiPath (RPA) - https://www.uipath.com/rpa/robotic-process-automation
• IBM (AI agents) - https://www.ibm.com/think/topics/ai-agents
• Grand View Research (RPA market) - https://www.grandviewresearch.com/industry-analysis/robotic-process-automation-rpa-market