Generative AI is ripping the rulebook on robotic process automation. Here’s how LLM-driven automation changes economics, risk and the vendor landscape.
Illustration by IMF Alpha editorial · Reviewed by Pedro Marini
The headline is simple: large language models are turning slow, brittle RPA into something much more dynamic. After a decade of rule-based bots that stitch screens together, models now let systems infer intent, rewrite workflows on the fly and surface exceptions — sometimes from a single prompt.
That reads like product copy, I know. But the shift is structural. RPA vendors used to sell playbooks. Now they need to sell judgment. When AI is embedded in orchestration platforms, companies pick up three practical advantages at once: faster rollouts, broader uptake by nontechnical staff, and the ability to cope with messy, unstructured work.
A quick sketch of the shift
Why this matters now
Two things converged this year. LLMs got cheap and reliable enough to be part of production pipelines. And enterprises grew tired of automation pilots stalling—projects that never graduated into steady operations. The result is a second wave of adoption, often driven by citizen developers and low-code tooling.
Think of it like moving from hand-coded sites to content management systems. Once nontechnical people could edit content without a developer, throughput jumped. The same dynamic is playing out with business processes, though with more compliance headaches and edge cases to manage.
Concrete examples
These are not pie-in-the-sky scenarios. Major automation vendors and cloud providers are already bundling generative capabilities into orchestration tools. Expect more partnerships and M&A as legacy RPA firms scramble to add AI expertise.
Risks and counterpoints
This is not a cure-all. Purely deterministic, high-volume tasks remain where classic RPA shines. But where processes involve text, judgment calls or frequent change, LLMs deliver outsized value.
Signals to watch next
Investor and operator takeaways
Automation is shifting from an IT project to a business capability. That’s where the real economic value sits: making automation usable, safe and measurable for the business, not just the data center.
Pedro Marini
Lightweight local models are enabling offline budgeting, privacy-preserving credit tools, and a new battleground for chips and banks.
As attackers weave large language models into phishing, malware obfuscation and supply-chain schemes, CISOs face a fast-moving threat and a market shift.
After months of cooling inflation and softer payrolls, the Fed is telegraphing a rate cut. Here’s who benefits, who gets squeezed, and how to position now.