RPA RIP
I am a strong advocate for business automation, helping to streamline common tasks to ensure human intelligence can be applied to higher-value opportunities, including areas that require curiosity, intuition, etc.
However, it is important to note that not all automation is created equal.
Over the past decade, I have been vocal about my concerns regarding the overuse of traditional Robotic Process Automation (RPA), which frequently conflicts with recommendations from service integrators and consultants.
My concern is simple, traditional RPA is commonly built on 20+ year-old technology, reliant on screen scraping (simulated user interface) techniques. In many respects, it is similar in concept to Selenium, the browser-based regression automation suite, first developed in 2004.
The value of RPA is the ability to replace human interactions (at the user interface level), allowing businesses to reduce or redeploy headcount. This delivers a compelling short-term ROI, which is why it is favoured by service integrators and consultants, as they can demonstrate “quick wins” with no accountability for future ramifications.
Unfortunately, this approach is very short-sighted, it essentially perpetuates legacy business processes, which can have disastrous medium/long-term ROI implications and impact future innovation.
This issue is compounded by the fact that traditional RPA is very “brittle” as it is often coupled to the user interface of the application it is automating. Therefore, any change to the application immediately breaks the automation (there is no intelligence).
To make matters worse, traditional RPA replicates the user and therefore requires user-level authentication, permissions and licensing, which impacts security and operational costs.
Finally, traditional RPA does not scale effectively. Each automation (bot) commonly requires an individual Windows-based virtual machine (or client), including access to the software that needs to be automated. This can quickly become complex and costly to support, further impacting the medium/long-term ROI.
Thankfully, there is another option, leveraging modern techniques and architecture to deliver API-centric automation that is loosely coupled, delivering flexibility, security, scale and cost-effectiveness. It also presents the opportunity to re-think the business process itself, which in my opinion, should always be the starting point before any automation is considered.
The one downside is the requirement to have access to viable API endpoints. In modern software, these are usually natively available. However, legacy software may require an additional intervention, adding complexity to the initial implementation.
With the introduction of Generative AI, including projects such as LangChain, we started to see evidence that API-centric automation could be combined with Generative AI to create “intelligent agents”.
These agents, with specific grounding, have the ability to take action and learn, whilst remaining loosely coupled to protect flexibility, as well as maintaining modern security standards.
In November, the venture capital firm a16z released the blog post “RIP to RPA: The Rise of Intelligent Automation”, which provides a great overview of this emerging trend.
In addition, a16z published a short video discussing the topic.
I agreed with their premise and am pleased to see others promoting modern approaches to automation, whilst highlighting the drawbacks associated with traditional RPA.
This year, I expect “Agents” and “Agentic AI” to enter the mainstream, likely positioned as the next evolution of Generative AI.
As a result of this market shift, I expect many traditional RPA vendors to “rebrand”, distancing themselves from “RPA” terminology, and replacing it with “Agentic AI”. This approach is “putting lipstick on a pig”. Therefore, I recommend a thorough investigation of the underlying architecture (looking beyond the sales pitch and marketing) before any investment.
As an alternative, I recommend new engagements start with the leaders/innovators in Generative AI and/or open-source alternatives. For example, Project Mariner from Google DeepMind and Operator from OpenAI.
My team has been experimenting with Agentic AI for many months. We intend to deliver a robust AI automation framework for our business, with dynamic workflows that can support a wide range of common and complex business tasks.