#78: BRF+, Stop the Agents, Vibes, Magic
In this issue:
Stuck on Manual in 2025?
Nobody likes doing the same stuff over and over. Humans are supposed to have a higher purpose than punching numbers into SAP GUI, right? While every company would agree with this on paper, how many actually provide their workers with tools to automate routine activities?
The tools have been available for a long time. For example, this post about accessing BAPIs from Excel via VBA is dated 2008. GuiXT existed since the ’90s and SAP GUI scripting has its own community page. The star of early 2000s Winshuttle might have been living in RPA’s shadow lately, but it used to be the envy of all business users - everybody loves Excel!
And speaking of RPA, there is no shortage of that either. SAP’s artist formerly known as iRPA, UiPath, MS Power Automate - you name it. Everyone who’s anyone has some RPA tool.
Then why, in 2025, do we not only lack flying cars but also have business users asking on Reddit how to automate their work? There are obvious reasons like licensing costs, setup complexity and access management. But some factors might not be so clear.
We seem to be perpetually waiting for the next great tool. You’re probably reading this and thinking, "Who needs RPA when AI agents are going to solve everything for us very soon?" But then nothing ever gets automated, not with RPA, not with AI. It’s as if "a bird in the hand" doesn’t apply to technology.
The impact on individuals is underrated. There is expectation for many users to be affected before anyone looks at automation because it looks impressive for the shareholders (X hours per Y users = profit!). But there are always small groups with tedious specialized tasks that are not less important. Automation doesn’t have to always be “go big or go home”. The tools suited for individual users can coexist with enterprise-wide solutions.
Most importantly, successful automation - of any size - requires IT and business to work together. It requires trust and a willingness to accept less-than-perfect outcomes (see MVP :) ). If you want any chance at automating routine work and improving UX, start the conversation. Preferably not on Reddit. JP
Chill Out On The Agents
I was going to write about Salesforce's Agentforce 2dx announcement. It was simultaneously hilarious and poignant, wry and wise. You would have loved to read my insightful commentary on things like how Marc Benioff is the biggest Copilot hater in the world, or how Salesforce's "Data Cloud" product makes SAP's choice of nomenclature "Business Data Cloud" all the more confusing and bland. I might've even tossed in a snarky Joule aside.
But I think "agents" and "agentic" are starting to be like that thing where if you repeat a word enough it starts to sound like nonsense. (Try it! Say your own name out loud a hundred times in a row.) What they do is amazing, what they will be able to do very soon will be even more amazing.
The endless announcements and hype confuse reality. Over the long term, if you take care to speak truthfully and act meaningfully, the stars will align. In the frontier AI space Anthropic does the best job of this. They do something worthwhile, say it clearly, and strip out hyperbole.
And then every time I go make fun of some agent announcement I'm just adding to the problem! So I'm going to shut my mouth on agents for a while. PM
BFFs With BRF+
For as long as computer systems have existed, users wanted to have some options to maintain business rules without relying on the moody IT department. Most of these rules look like “if document type is A, then do this, unless it’s 2 pm on Tuesday”. And with that usually you’d get a requirement “can we put that ‘A’ value where we can change it?”
Creating a custom table (or even several) to maintain such values is actually quite a reasonable solution. And be sure to check out this great blog post for a neat ready-to-use option.
For some time, Z tables were inexplicably considered “bad,” leading developers to reach for standard solutions that were somehow deemed better. Not sure who started spreading the ridiculous idea of using TVARVC table as a catch-all for random business rule values, but sadly, it spread through SAP systems like wildfire - with a similarly disastrous effect.
SAP wasn’t asleep at the wheel though. With NetWeaver 7.0 EHP1 release, the business rule framework called BRF+ was released. But while the framework itself is solid, it did receive a kiss of death from slow and clunky Web Dynpro UI. And as this excellent article points out, “BRF+ is an interesting tool, but it’s not the revolution SAP might have hoped it would be”.
Of course, with push to the cloud, there is even more rule functionality (see the diagram in this video). But something tells me Z_ PUT_ALL_CODES_HERE will continue to exist in many SAP systems. JP
Collaborative Vibes
Through AI nerd influencer Andrej Karpathy, the term "vibe coding" has made a splash. The gist: "the LLMs are getting too good. I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works". That's a big, warm hug for AI from an industry luminary.
I see that happening, too. I had a request the other day to help someone whip up a quick Python tool to handle a weird one-off scenario. I prepared my prompt, smashed it through Claude with extended thinking, did a follow-up after something didn't go how I wanted, et voilà! The thing worked. I shipped it off to my colleague, they used it for a few days, and it's now gathering dust on a hard drive somewhere.
But there's a distinction. I just started working at Nova AI, where we're building agents to create enterprise (SAP) applications. Nothing will convince you that there is still a world outside of vibe coding like working with enterprise applications. There are extra technical pieces: a stunning amount of custom tooling you need to provide the AI, showing the AI the right stuff at the right time, and working your way around context limits. There's extra knowledge work: LLMs know less about SAP than about popular Python libraries, so you better up your prompting game and find, test, and patch up those holes. There is SO MUCH stuff to do in enterprise applications that it kinda boggles the mind. A demo of autonomous agents building SAP apps will knock your socks off - but so will the tooling that's been developed to support such a demo. Claude, with only its turbo-AI wits, just can't go the whole way. (Yet.)
In the future, collaborative vibes are how I see this working. I imagine an awesome Zoom call where 4 humans and an AI monkey around in real time on a new enterprise application. The humans bounce ideas off each other and fill in context gaps on business needs, and the AI continually revises an application while asking clarifying questions along the way. It'll be fast enough that the meatbags can vibe along with the AI, together. Until Claude can memorize your entire business and dream up new things by itself. PM
Some Kind of Magic
Low code is so yesterday. If you want to build an SAP Fiori app these days, just draw an idea on a paper napkin, upload it to the SAP Fiori Project Accelerator, and bask in the rays of glory. Who needs those pesky, expensive developers?
This is the impression you’d get from reading SAP marketing posts and some LinkedIn influencers. But as usual, here comes Reality Debbie Downer.
It is indeed possible to upload a sketch and have Fiori tools generate something from it. (Although we’ve already seen that Joule can also generate an app from just a description, so making a drawing feels like “slavery with extra steps.”) But that something is not going to be a fully functional app. And, ironically, the part that you can’t get from a drawing is usually the tough one.
There are also some prerequisites, and SAP Help doesn’t offer a simple answer to what exactly is needed to transform the prototype into a productive app. This kind of “magic” reminds me of Lioz Shem Tov’s act - brilliant and hilarious, but clearly not at all magical. JP
P.S. Check out my longer comment on this subject on LI.
Take A Breath, Paul
Ethan Mollick, on LinkedIn, has a "sense we are getting to the final days of sites like this being a viable way for humans to communicate." The conclusion he reaches: "the public-facing discussion forums are not really viable, at least in the format in which they used to operate." For him, the core issue is LLMs leaking across everything in a way that's obvious.
I think he's right. We don't need this. I'm pretty sure LLMs don't get anything out of it either. I sort of hate posting on LinkedIn. Every time I write something, I think "is this just another shitty post in a sea of shitty posts?" When I say something on social media, I mean it - but I almost never like saying it there.
Why does this fit in a newsletter about enterprise software? I guess it's more the "enterprise" than the "software". A professional career involving large organizations carries with it a little "I don't actually care about some things, but we all acknowledge that we pretend to care more than we do", so it can be pretty awesome to go somewhere to see where old friends and colleagues update what they're doing. Real human connection that breaks the surface of meaningless stuff! But Ethan is right: LLMs have pushed LinkedIn to a breaking point. It's time for something else. PM
Watch Nerdletter Talk on our YouTube channel for insightful commentary and bonus content.
In the new Show&Tell video our guest is SAP SD expert and book author James Olcott.
Enjoyed reading / watching? Support the Boringverse by buying us a cup of coffee !
This fun video came out after this Nerdletter issue and goes nicely to illustrate the vibe coding story. :) https://www.youtube.com/watch?v=JeNS1ZNHQs8