RPA and the “robot” concept#
The other side of automation doesn’t use APIs at all – instead, it basically mimics exactly what you would be doing inside of an app or operating system. If you’ve ever recorded an Excel macro, this will sound familiar; there are basically a couple of steps:
- Open up your recording app (like UIPath)
- Click “record”
- Do the manual task
- Stop recording
- “Play” the task whenever you want
It’s sort of like recording a “how to” video, but for your computer – and it’s much different than using APIs, because anyone can do this. The hot acronym for this these days is Robotic Process Automation, or RPA.
(Recording an Excel macro)
If you’ve had a job before (sorry), you can probably think of one or two things that you’ve that really, probably, should be automated by now – and how you’d be able to do it with robots. At my first startup job, I was in charge of our monthly metrics report – I needed to run a basic SQL query, copy the results, paste them into an Excel spreadsheet, update a bunch of column names, and email the sheet to a few people on the leadership team. Would have been a great candidate for RPA. But alas, I was young and dumb – and now I’m just dumb.
This, in a nutshell, is most of what UIPath does – it’s a product (sorry, a platform) that helps teams robotically automate their tasks.
The UIPath product#
The main value proposition of UIPath – and their core product – is the ability to record your manual tasks, save them as a process, and replay them whenever you want, instantly. Here’s their app for recording tasks (UIPath Studio):
You can read a basic tutorial on how to do recordings here. And beyond recording, you can also build workflows visually by choosing available apps and options and chaining them together on certain conditions.
But where UIPath gets really interesting is the ability to add other shit (this is the technical term) to automations. A couple of big ones:
You can create basic web forms in UIPath and then work with that data in a process. There’s a good example on their site about how banks use it to automate KYC (know your customer) processes – they’ll make customers fill out a form, then automatically create templated documents for them to fill out and sign.
2) Business logic#
UIPath lets you loop through lists of things, set conditions based on the current date, and set processes to run on a schedule. So once you’ve got the basics of your automation done, you can fine tune the specifics easily.
3) Working with external data#
You can read and write data from external sources in UIPath: you can get emails from an inbox, read data from a database, and even make an HTTP request. And when you combine this with something like a form, that gives you the ability to create (admittedly rudimentary) UIs on top of data – which makes UIPath start to look more and more like an app builder.
UIPath lets you build workflows in two ways – via recording your actions, and visually through their UI workflow builder. For more custom work like logic, loops, and working with data, recording doesn’t quite cut it.
These days, I think it’s fair to say that UIPath ends up combining both types of automation – working with data and APIs, as well as RPA and good ol’ recording what you’re doing. And that’s their marketing pitch, pretty much – automate everything, no matter how or why. And like with every generic “builder” they face the core problem of people like me showing up to the website and wondering “what would I actually do with this” – which is why they put so much time into diving into use cases.
UIPath or AIPath?#
UIPath is making a pretty big bet on AI. At the last update, their entire homepage is about AI:
There's something called the "Agentic AI Summit" which I guess is their conference about UIPath and AI. Cutting through this extensive noise, there are two important products to focus on here.
The first is AgentPath, which is UIPath's take on AI agents. It's a graphical interface for building agents that do clerical work for you like processing invoices, researching a topic, or matching HR codes in a hiring process. One way to think about this is that UIPath's core product gave you the ability to record what you usually do, and then automate it exactly; agents are just a much more sophisticated version of this that can pull in data from other sources and make semi-independent decisions.
The second is called "Agentic Orchestration" and it's the more interesting one to me. It's a product that helps integrate agents into the normal automation workflows that some of these very large enterprises have. For example, UIPath agents now play nice with an industry standard "language" for (nerds) modeling business decisions and rules called DMN. It also lets you decide when to escalate things to humans (like when an agent is struggling with a decision).
Also some top quality cringe, for those interested.
For more depth on UIPath financials and comps, check out John’s newsletter.
