#102: ABAP+AI, Example-Driven Development, Tech Radar

In this issue:

• ABAP + AI: A Practical Use Case

• Example-Driven Development

• SAP Tech Radar: Beep, Beep

• Yes But Also No

• Computer Science in 10 Minutes

ABAP + AI: A Practical Use Case

Alexander Tsybulsky’s “Adventures in Offline ABAP Development” was a breath of fresh air in my LI feed swamp. It would already be a solid article if it had been written by a professional developer, but considering Alexander is a FICO consultant, it’s especially impressive. (And I wonder whether that actually contributed to the choice of good use cases. It also makes me think that functional consultants equipped with AI might be a bigger threat to ABAP jobs than anything else. But that’s another story.)

This article is about practical experience and lessons learned from an AI adventure rooted in ABAP reality. Key lessons Alexander highlighted:

• “AI does not deliver faster than human but helps focus on better product design and maintainability.” This sounds surprising at first because AI tools do write code much faster than humans by definition. However, I believe this refers not just to the speed of cranking out an alpha version of the code, but rather to delivering a complete solution, which includes some back-and-forth.

• “Planning the agent work stages use vertical task split.” I haven’t heard of this approach, at least not under this name. Typically, developer tasks line up “horizontally” (select data, do stuff with data, present data), and that’s how I would explain tasks to an inexperienced developer, for example. Also, for me personally, working on every part of MVC at the same time would cause additional mental load. But AI tools don’t have that disadvantage, so giving them tasks in a different way could be advantageous.

• “Auto tech documentation can be a nice bonus.” 100%. Say what you want about AI code quality, but any AI tool (even free tier of ChatGPT) can produce very decent technical documentation. Just make sure to specify the style and level of detail needed. I’ve had odd cases when Gemini went either too salesy or started transcribing almost every line of code. Providing a good example of what you’re looking for works best, in my experience.

It looks like Alexander’s offline adventures are going to continue, so make sure to follow him to read the next installment. JP

P.S. As this issue was going to press, another good post from Philipp Dölker surfaced with more helpful links on the subject.

Example-Driven Development

It’s really hard to communicate new things to people. In my day job, the thing I think hardest about is “How do I best show what our product is for, and how to use it?” And I’m usually stumped for quite a while!

SAP RPT is getting an upgrade in H2 of this year to a 1.5 version. (Be careful if you go to the RPT page, because the page talks about 1.5…but I am not sure that the RPT trial yet points to that version! Maybe it does.) The idea of RPT is very cool: a general-purpose cell data prediction program for relational data. If you’ve got blank spots in your structured data, ask RPT to fill in those blanks based on the surrounding info. Honestly, a pretty slick concept.

But if you go try out the examples, you’ll quickly bump up against some not-quite-curated data sets. As Jelena points out in our Nerdletter Talk video, with SAP-like data there are many fields where it really wouldn’t make sense to predict a value. If there’s no sales org on a sales order, you have a much different problem than predicting the value! Other columns or features might be more amenable to prediction, like adjusting priorities on tickets.

This is not a harsh judgment on RPT. I’m impressed, and I bet I’ll be more impressed when further updates come. It means that as a product you need to put in the hard work to curate examples where the usefulness slaps customers in the face. This is almost as difficult as creating the product in the first place. PM

SAP Tech Radar: Beep, Beep

It’s been a while since Tobias Hofmann first published this SAP Development Tech Radar. Initially, some folks argued that it is more of a “Tobias’s Tech Radar” and not necessarily representative of the wide ecosystem. Well, “haters gonna hate”, but radar lives on and Tobias just published another update. New on the radar: AIF, ABAP Doc, and Knowledge Transfer Document (KTD).

AIF is worth keeping an eye on, but ABAP Doc is an odd addition IMHO. It got a mention in my ABAP book and I was more optimistic about its future in 2020. Now I’m not so sure. It feels like it never lived up to the promised potential and might as well be killed by AI soon. But I’ve heard some folks argue just the opposite: that it somehow will become more useful than never.

I had to look up what KTD meant in SAP context: apparently, it’s the thing. It probably fits in the same bucket as ABAP Doc: either is going to be very useful for AI or will be killed by it. Keep your radars finely tuned! JP

Yes But Also No

Dr. Autumn Krauss, chief scientist at SuccessFactors, dropped the rare useful and informative piece on the SAP news site: an action plan for early talent hiring and development. I have a reflection and a related complaint.

Reflection: I find the recommended step 3 (“reframe the business case for early talent”) compelling. Nothing could be more clear than these days early career folks face an environment full of quicksand and R.O.U.S. that us old-timers didn’t. A college degree isn’t nearly as useful. College debt is crippling. Employers are less willing to train and provide fewer incentives to stay for the long haul.

The people holding the purse strings need to see the benefit of investing in young folks. I’m at a stage of my career where I work with lots of people younger than me, and holy shit are they awesome. Certainly more so than I was at their age.

Complaint: My company is hiring, and I spend time reviewing applications and resumes. If you’re going to bot-spam, do at least one proofreading pass of your slop salad. I regularly see resumes claiming to have implemented for my company the very thing we’re trying to hire for. Trust me, if you’d built that for us already, we would know. Cease fire with the resume slop cannons already. PM

Computer Science in 10 minutes

Fireship is back with a banging video covering the whole century of Computer Science papers. (Yes, it’s been THAT long! Can you imagine?) In 10 minutes, you can learn enough about the subject to confidently win any computer-related discussion with “Have you even read Marvin Minsky and Seymour Papert?!” And if 10 minutes is too long for you, there is TLDR at the end. JP

• Watch our Nerdletter Talk 102 for a conversation about these stories.

• Are you interested in sponsoring the Nerdletter and featuring your amazing product? Have a story to share or feedback to give? Get in touch!

• This newsletter is always free and proudly written by humans and for humans. Show your support before it’s too late!