"Robotic" Process Automation?

A vision of the future we were told...Robot Process Automation - or RPA for short - conjured up all manner of images when the term first became popular. Humanoid mechanical contraptions doing the work of people was expected. Alas, the reality was somewhat less disruptive, and the "robots" were in fact scripted workflows on-rails that carried out repetitive copy/past/click-here tasks that used to be the job of what used to be known as "data entry clerks".

But we were promised flying cars and hoverboards?

RPA was founded on the idea of automating repetitive tasks by mimicking human actions and when it first broke onto the scene it generated plenty of hype. Nowhere near the tsunami of hype we're experiencing today when it comes to AI of course, but hype nonetheless. Companies like UiPath raised billions on the premise of enabling “fully automated enterprises", but the reality never quite lived up to the promise. The reason? RPA bots could really only handle rigid, predefined processes, and maintaining them was costly and cumbersome. It did the job, and the labour arbitrage cost-savings were possible, even if they weren't the 80% saving we all expected.

Operations work isn’t glamorous. It doesn’t grab headlines or feature in the glossy marketing pitches of big tech, but it’s the backbone of most organisations. All of that data we need to capture, then move, then migrate, then report on needs to be managed by someone (or something) and that used to mean someone in a contact centre or back-office function did the grunt-work.

I've seen and worked on the RPA business model first-hand. Why pay ￡XX per month/annum for a person to spend a mind-numbing day at work repeating the same data-entry or data manipulation task over and over again? Instead, why not get a "robot" to do the job instead for a massively reduced cost. Or even better, why not pay on a per-task basis, the true definition of an "outcome based" contract?

Now though, thanks to advancements in large language models (LLMs) and generative AI, it looks like the game is changing, and that we're finally at a point where true intelligent automation can deliver on the promises RPA could only hint at.

How AI is Changing the Game

As you'd expect, moving away from pre-defined workflows and scripted processes is now made possible with Gen-AI and LLMs. Instead of hard-coding bots to follow step-by-step instructions, AI agents can be given a goal - like transferring data from a document to a database then updating a record, by being equipped with the tools to figure things out for themselves. These agents adapt to different inputs, handle exceptions, and evolve as business processes change (allegedly - it's difficult to validate this in a real-world scenario right now given the newness of the technology, but I'm sure no-one in "the valley" would lie to us, would they?).

This flexibility makes them far easier to implement and maintain than traditional RPA. It also opens the door to automating tasks that were previously too nuanced for software to handle. With AI agents, the focus shifts from mimicking human behaviour to achieving business outcomes.

So 2025, we're told, is the year of the autonomous agent.

The evidence is there; companies like Anthropic are enabling AI models to interact with existing software via their "Computer Use" approach (https://www.anthropic.com/news/3-5-models-and-computer-use) and there's been an explosion in startups offering all sorts of automation using LLMs; everything from customer support to document extraction. OpenAI has also released their own agent framework - Swarm (https://github.com/openai/swarm) - and rumours abound that they're meeting with the US Government on 30th January 2025 to demo some earth-shattering new agent capabilities (let's see how this article dates, shall we?).

The Business Opportunity

The potential market for intelligent automation is enormous. For all the work that software has already streamlined, countless processes remain untouched, relying on paper forms, spreadsheets, and manual data entry. You and I know this, because you and I are dealing with these inefficiencies and archaic processes every day. Intelligent automation will, hopefully, address not just the labour costs of these tasks but also the inefficiencies they create...although everyone seems to be forgetting the layers upon layers of legacy architecture that most companies out there still have to contend with.

Question is, can the agents work with our architectural sins of the past?

A Vision for the Future

The promise of intelligent automation is exciting, but it’s not without challenges. Success will require businesses to build adaptable, user-friendly solutions that deliver measurable value. I personally also think that creating a "moat" is going to be harder than ever. That's not to say it will be easy to execute to the same level of effectiveness as the leading players in the agentic space, but it will probably be a lot easier for a lot of people to have a crack at it.

What are you doing about it?

Agents and automation are topics that I'm particularly interested in, and I'd love to hear how you're planning to take advantage of these technologies. Human to human of course. :-)