We explained what RPA was back in late 2024 — bots that mimic clicks and keystrokes to automate rule-based, repetitive work without touching the underlying systems. Two years and one AI cycle later, the question isn’t what RPA is anymore. It’s whether it still matters, now that every vendor is talking about agents instead of bots. The 2026 numbers say RPA hasn’t gone anywhere — but it isn’t the thing generating the excitement anymore either, and that distinction matters if you’re deciding what to invest in next.
The Market Didn't Collapse. It Slowed.
Gartner's most recent market-share analysis puts the global RPA software market at $3.6 billion in 2024, up 14.5% from the year before — still growing, but Gartner's own analysts point to why the rate is easing: generative AI, computer-use tools, and agentic automation are starting to pull budget and attention that used to go straight into classic RPA licenses. Zoom out and the bigger category RPA sits inside — hyperautomation enablement software — is still projected to reach $1.07 trillion by 2028 at a 13.9% CAGR. The money isn't leaving process automation. It's migrating toward the "agentic" label sitting on top of it.
The Vendors Are Rewriting Their Own Pitch
UiPath, still one of the three dominant RPA vendors alongside Microsoft and Automation Anywhere, closed its fiscal 2026 (ended January 31, 2026) with $1.611 billion in revenue, up 13% year over year, and $1.853 billion in annual recurring revenue, up 11%. Those are healthy numbers for a public software company, but they're a step down from the growth rates RPA vendors posted earlier in the category's life — and UiPath's own language reflects the shift: the company now describes its platform as combining "deterministic automation, agentic AI, and enterprise-grade orchestration," not RPA on its own. When the market leader stops leading with the term that made it famous, that's a signal worth reading.
Enterprises Are Buying the Vision. They Aren't Running It in Production Yet.
A Coleman Parkes survey commissioned by Camunda — 1,150 senior IT and business decision-makers at companies with 1,000+ employees across the US, UK, France, and Germany, fielded September–October 2025 — found 71% of organizations already report using AI agents. Only 11% of agentic AI use cases actually reached production in the past year. Seventy-three percent admit a real gap between their agentic AI vision and what's actually running. That's not a rejection of automation broadly: 95% of the same respondents said process automation increased business growth over the past year, up from 87% the year before, and the average organization has automated 48% of its processes with a self-assessed potential of 64%. The appetite for automation is intact. What's stalling is the leap from RPA-style deterministic workflows to agents making judgment calls.
Where the Stall Actually Comes From
Deloitte's survey of 501 senior manager-to-C-suite respondents at organizations actively piloting agentic AI (fielded April–June 2026) found the same pattern from the inside: 42% have tested or deployed AI agents, but only 15% have reached scaled, orchestrated multi-agent adoption, and just 21% say their business processes are actually "prepared" for it. Ask why, and the answers point back to infrastructure, not ambition — 72% say they lack unified, accessible data, 70% say they can't adequately trust and govern the agents they'd be deploying, and 67% cite integration complexity as a real barrier. Those are the exact problems RPA was built to sidestep: it doesn't need clean, unified data or a governance framework for autonomous judgment, because it isn't making judgment calls. It's following a script.
Our Verdict
RPA in 2026 isn't obsolete and it isn't the future either — it's infrastructure. The rule-based, high-volume, low-ambiguity work RPA was built for (data entry, reconciliations, system-to-system handoffs) still runs on it, still works, and still doesn't need an agent's judgment to get done reliably. The place agentic AI earns its premium is the work that was never a good fit for scripted bots in the first place: judgment calls, exceptions, and unstructured decisions. The mistake we're seeing clients make isn't sticking with RPA too long — it's ripping out working deterministic automation to chase an agent pilot before the data foundation and governance controls exist to run one in production safely. Keep the bots doing what bots do well. Build the data and governance layer Deloitte's respondents are missing before you hand judgment calls to something that can improvise.
If you're trying to figure out which of your automated processes actually belong on an agent and which should stay exactly as they are, that's a conversation we have with clients regularly.