Agents Will Eat $143B by 2027, and Most of It Comes Straight From RPA's Plate

Here is a number that should make every automation vendor nervous. IDC projects that spending on AI agent platforms will reach $143 billion by 2027, and it says a large chunk of that comes directly out of the budget that used to go to robotic process automation and the older business-process software around it. After a decade of being the default answer to "automate this", RPA is suddenly the incumbent getting eaten.

Before the obituaries, one honest caveat: RPA is not collapsing. IDC also expects RPA spending to more than double by 2028, to around $8.2 billion. Both things are true at once. The category keeps growing in absolute terms while agents claim the new money and the most interesting work. What is happening is not a death, it is a demotion, from the future of automation to a piece of plumbing. To see why, it helps to remember what RPA actually was.

What RPA was, and why it was always brittle

RPA, robotic process automation, is the technology of software "bots" that operate other software the way a person would: clicking buttons, copying fields, moving data between systems that were never designed to talk to each other. For fifteen years it was genuinely useful, because most business software has no clean way to connect, and a bot that mimics a human clicking through screens could bridge that gap.

The catch was always fragility. An RPA bot follows a fixed script of "click here, type there, read this box". It does exactly what it was told and nothing else. So the moment reality drifted, a vendor moved a button, a form added a field, a page loaded half a second slower, the bot broke. It did not adapt, it did not reason about what changed, it just failed, often silently, and a human had to go rebuild the script.

Anyone who has run RPA at scale knows the real cost is not building the bots. It is maintaining the army of them, each one a small liability that snaps whenever the world underneath it shifts. The bots automated the work but created a new full-time job: keeping the bots alive.

What agents do differently

An AI agent approaches the same task from the other end. Instead of memorizing a fixed sequence of clicks, it understands the goal and figures out the steps, and it can adjust when things are not exactly as expected.

That difference shows up in three ways that matter.

It handles ambiguity. Give an RPA bot an invoice in a slightly new layout and it chokes. An agent reads the invoice, finds the total wherever it moved to, and keeps going, because it understands what an invoice is rather than memorizing where the number sat last Tuesday.

It recovers from change. When a website redesigns or a form sprouts a new field, an agent adapts on the fly far more often than it breaks. The thing that used to trigger a maintenance ticket is now just a small detour.

It explains itself. A failed RPA bot leaves you a stack trace and a shrug. An agent can tell you what it was trying to do, what it ran into, and what it decided, in plain language. When something does go wrong, you get a colleague's account of it, not a cryptic error.

Put simply: RPA automated the keystrokes. Agents automate the judgment that decides which keystrokes to make. That is why the same task handed to an agent does not generate the maintenance tax that made RPA quietly expensive.

Why the incumbents are scrambling

You can read the whole shift in how the RPA leaders are repositioning. UiPath is redesigning its platform around agents that build and adapt automations rather than just run fixed ones. Automation Anywhere now pitches "Agentic RPA", where you describe a process in plain language and the platform spins up both a classic bot and a reasoning agent to supervise it. The category leaders are, in effect, bolting a brain onto the bot, because the bot alone is no longer the thing customers want.

That is the tell. When the incumbents start describing their product as the new thing rather than the old one, the new thing has already won the argument.

What this means if you still run RPA

If your business depends on RPA today, you do not need to rip anything out this quarter. The bots that work, keep working. But the questions worth asking have changed.

For any new automation, the default should flip: ask whether an agent can do it before you script another brittle bot, because the agent will not generate the same maintenance bill. For the bots you already have, the ones that break most often are the best early candidates to replace, since their fragility is exactly what agents fix. And when you evaluate any agent platform, look past the demo to the boring questions that decide success: does it have real memory of your processes, its own identity and permissions, and a clear log of what it did. Those are the things that turn a clever demo into something you can actually rely on.

The deeper point is that $143 billion does not move because of a buzzword. It moves because businesses spent a decade paying the hidden tax of fragile automation, and something finally arrived that does the same work without the brittleness. RPA taught a generation of companies that their software could run itself. Agents are about to teach them it can do so without breaking every time the world changes.

Want to test the most advanced AI employees? Try it here: https://Geta.Team