Curious how the new CUA (Computer Using Agent) model can be trained. Would love to automate some accounting processes in SAP by using the CUA model API when it becomes available to devs.
Pretty much the same way like operator was build I guess.
Most probably some guys clicked through the internet all day and they recorded it.
And then from that video stream find a threshold in image change and assume something happened… The dom was then linked to it and scraped…
In SAP depending what you are using it might be similar… sap/ui5 would work easily…
But did you know you can easily build apis for SAP as well? Just get an intern… they can build that…
hmm operator doesn’t work on ui5 stuff?
Excuse my old SAP knowledge… been some years since I worked with it last time.
All good! After researching further seems if we just created an AI Agent specific to our needs using an OpenAI model for use in SAP that will suffice. We’re looking to process invoices and have the agent make journal entries into SAP based upon the invoice content. Seems to be a more scalable and faster approach to automation vs the front end CUA model.
My question is, why are AI agents just now in 2025 becoming a legit use case. Is it because the AI models have finally become “good” enough?
Thanks and appreciate your input!
You want to process invoices with GPT? Nah… lol…
That’s not what it is meant for.
It can give you an approximation of the content of a document.
Would it be ok when you extract 90% of the line items 95% of the time and sometimes all and sometimes only one price is missing or a number?
Ha…well shoot. I guess there isn’t a solution right now to build an AI agent that can intake invoices and enter them as journal entries into SAP??