Pop­u­lar­ized in recent years by the main­stream debut of Chat­G­PT, gen­er­a­tive AI capa­bil­i­ties con­tin­ue to attract the atten­tion and curios­i­ty of lead­ers in enter­prise technology. 

But there’s much more to dis­cuss when it comes to arti­fi­cial intel­li­gence, from the promise of agen­tic AI capa­bil­i­ties to the con­crete busi­ness val­ue that AI stands to unlock with­in the next decade. 

To dis­cuss these top­ics and more, ASUG recent­ly caught up with Andreas Welsch, the Founder & Chief AI Strate­gist at Intel­li­gence Brief­ing and the author of ​‘AI Lead­er­ship Hand­book: A Prac­ti­cal Guide to Turn­ing Tech­nol­o­gy Hype Into Busi­ness Out­comes.’ Welsch pre­vi­ous­ly spent 23 years with SAP in a vari­ety of lead­er­ship roles, most recent­ly advis­ing For­tune 500 lead­ers on real­iz­ing busi­ness val­ue from AI and accel­er­at­ing the inte­gra­tion of AI across SAP’s enter­prise applications. 

AI’s Slow Start 

Since the 1950s, researchers and indus­try lead­ers have been try­ing to build sys­tems that can mim­ic human behav­ior, human rea­son­ing, and dri­ve automa­tion. But it hasn’t been easy, with plen­ty of peri­ods of excite­ment and dis­ap­point­ment, explains Welsch. ​“We are now at a stage where we have lots of data avail­able, we have infra­struc­ture avail­able that’s scal­able and elas­tic, and we have bet­ter algo­rithms and mod­els in place that can, to a good extent, mim­ic human behav­ior and rea­son­ing,” he says. 

“About eight years ago, machine learn­ing (ML) and deep learn­ing emerged, enabling sys­tems to ana­lyze data, iden­ti­fy pat­terns, and draw con­clu­sions or infer­ences. Today, with GenAI, for the first time in the his­to­ry of com­put­ing, we’re able to not just make pre­dic­tions or give rec­om­men­da­tions, but to gen­er­ate infor­ma­tion and cre­ate some­thing new, whether it’s text or image or video or audio, and do that at a lev­el of qual­i­ty that is almost human-like,” says Welsch. ​“So, from an enter­prise per­spec­tive, we’ve come from sta­tis­ti­cal meth­ods and opti­miza­tion and fore­cast­ing to mak­ing pre­dic­tions, rec­om­men­da­tions, and clas­si­fy­ing infor­ma­tion, to now being able to gen­er­ate information.” 

Gen­er­a­tive AI Chal­lenges for CIOs 

Beyond the hype, it’s still ear­ly days for gen­er­a­tive AI, espe­cial­ly when it comes to adop­tion and assess­ing busi­ness impact. Prac­ti­cal issues such as man­ag­ing costs, inte­grat­ing data, and upskilling employ­ees are on the minds of many tech­nol­o­gy executives. 

“Lots of orga­ni­za­tions have start­ed pilot projects, and many orga­ni­za­tions have adopt­ed pro­duc­tiv­i­ty tools,” says Welsch. ​“But I think CIOs are well advised to col­lab­o­rate with their CHRO [Chief Human Resources Offi­cer] peers and their learn­ing and devel­op­ment teams to also help employ­ees learn how we actu­al­ly use these tools.” 

Welsch addi­tion­al­ly advis­es CIOs to keep an eye on costs for these solu­tions. They’ll need to fine-tune their data strate­gies to ensure busi­ness and cus­tomer data is accu­rate and clean. Anoth­er key chal­lenge is invest­ing in upskilling and training. 

“Skills need to be built on dif­fer­ent lev­els,” says Welsch. ​“But where I believe gen­er­a­tive AI does actu­al­ly low­er the bar­ri­er of entry is for devel­op­ers. You don’t have to be a data sci­en­tist to use and incor­po­rate gen­er­a­tive AI tools, unlike machine learn­ing, where you were build­ing mod­els based on your data. Now you can use a mod­el off the shelf from one of the main providers,” Welsch says. ​“From there, it’s about learn­ing how to inte­grate the data and write good prompts that cost-effec­tive­ly deliv­er the desired outputs.” 

From Genera­tive AI to Agen­tic AI 

Accord­ing to Welsch, we’re head­ing toward even more auton­o­my and high­er lev­els of automa­tion. Agen­tic AI, or AI agents, can take a user’s goal, inter­pret it, break it down into sub-goals and tasks, and work on those tasks inde­pen­dent­ly. These agents can also rea­son and review their work to decide if results meet user require­ments before shar­ing back with those users. 

With­in such advance­ment, Welsch empha­sizes the impor­tance of keep­ing humans in the loop. Peo­ple will need to review the infor­ma­tion that these agents cre­ate, and there will still be tasks that humans are bet­ter equipped to handle. 

Agen­tic AI is mov­ing fast. Last fall, SAP – along with oth­er major ven­dors like Sales­force, Microsoft, and sev­er­al oth­ers announced or launched their first AI agent offer­ing. Some open-source frame­works and star­tups are devel­op­ing agent frame­works in this space. Welsch pre­dicts we’ll see an exper­i­men­ta­tion phase for AI agents akin to enter­prise tech­nol­o­gists’ ini­tial work with large lan­guage mod­els (LLMs) like Chat­G­PT 18 – 24 months ago. 

Welsch advis­es senior IT lead­ers to have their teams dig into this tech­nol­o­gy, which could auto­mate tasks more com­plex than those robot­ic process automa­tion (RPA) has been his­tor­i­cal­ly able to han­dle. Welsch sug­gests that now is the right time to start talk­ing with enter­prise soft­ware ven­dors and test agents in a con­trolled envi­ron­ment to under­stand how they work and their lim­i­ta­tions. Build­ing trust is cru­cial — both with­in orga­ni­za­tions and with exter­nal audit teams and end users. Risk mit­i­ga­tion, gov­er­nance, and review process­es will sim­i­lar­ly become increas­ing­ly impor­tant for IT organizations. 

Still, Welsch stress­es there is great poten­tial for AI agents to take on tasks that tra­di­tion­al­ly require human effort. Under­stand­ing, cre­at­ing, and shar­ing text in areas such as cus­tomer ser­vice oper­a­tions and IT ser­vice desks could be one such task. ​“If you can train an AI agent to address basic requests, that means your human agents can spend more time on the real­ly com­plex tasks with­out hav­ing to deal with and respond to each of the man­u­al and repet­i­tive tasks like reset­ting my pass­words,” he says. 

AI agents could also assist in review­ing cus­tomer ser­vice requests — iden­ti­fy­ing inquires, cat­e­go­riz­ing them, find­ing rel­e­vant infor­ma­tion, check­ing for past respons­es to sim­i­lar queries, and draft­ing respons­es. The drafts could then be sent to a human cus­tomer ser­vice agent for review and approval before being sent out. Anoth­er area ripe for AI agents is with­in mar­ket­ing func­tions, where they could help draft strate­gies, cre­ate mar­ket­ing briefs, refine mes­sag­ing, and write copy. While these tasks will still require human review, AI agents will enable mar­keters to work at a much faster scale, Welsch predicts. 

AI’s Trans­for­ma­tive Future 

While a lot can change in just six months, and pre­dic­tions sev­er­al years out can feel like a shot in the dark, Welsch believes that over the next few years, we’re like­ly to see the rise of intel­li­gent, col­lab­o­ra­tive, and autonomous mul­ti-agent systems. 

“Rough­ly 10 years ago, when I was work­ing for SAP’s chief tech­nol­o­gy offi­cer (CTO), we were putting togeth­er a vision,” Welsch recalls. ​“At the time, machine learn­ing was just emerg­ing. IoT was a big thing, and big data ana­lyt­ics was the bil­lion-dol­lar sto­ry. And we cre­at­ed this vision­ary video about a com­pa­ny using a mul­ti-agent sys­tem. Now, we’re at this point where there are fea­si­ble and viable frame­works that let you do that.” 

In Octo­ber 2024, SAP intro­duced col­lab­o­ra­tive AI agents with cus­tom skills to com­plete com­plex cross-dis­ci­pli­nary tasks with­in Joule, its gen­er­a­tive-AI copi­lot. Oth­er ven­dors are also inte­grat­ing sim­i­lar capa­bil­i­ties into their plat­forms. Ulti­mate­ly, this will enable spe­cial­ized vir­tu­al agents to oper­ate autonomous­ly and col­lab­o­ra­tive­ly – with­in teams, across depart­ments, even between com­pa­nies. For exam­ple, a vir­tu­al pro­cure­ment agent could nego­ti­ate a deal with a supplier’s vir­tu­al sales agent and then coor­di­nate with its company’s vir­tu­al agents han­dling man­u­fac­tur­ing, sup­ply chains, and logistics. 

Welsch says it will take time for this inno­va­tion to per­me­ate orga­ni­za­tions. ​“We’ve seen this time and again, and I don’t think that’s going to change. But in the next five years, I’m excit­ed about see­ing orga­ni­za­tions adopt this tech­nol­o­gy and become ever more capa­ble with the help of AI,” he says. ​“I think there will be a lot more automa­tion, and we’ll be see­ing [these agents] move towards auton­o­my between com­pa­nies. That’s where I think the big poten­tial is.” 

Addi­tion­al Resources from Andreas Welsch: Andreas has recent­ly pub­lished via LinkedIn Learn­ing. Indi­vid­u­als in orga­ni­za­tions with a LinkedIn Learn­ing sub­scrip­tion can get a 20-minute overview of AI agents and what to look out for at no addi­tion­al cost.

• AI Agents: Prepar­ing Your Orga­ni­za­tion for Change as a Busi­ness Leader 
• Agen­tic AI: Chal­lenges and Oppor­tu­ni­ties for Leadership 

FAQ Sec­tion:

Q: What is SAP Joule AI?

A: SAP Joule AI is SAP’s gen­er­a­tive AI assis­tant designed to enhance user pro­duc­tiv­i­ty and deci­sion-mak­ing across SAP applications.

Q: How does agen­tic AI dif­fer from tra­di­tion­al AI in SAP?

A: Agen­tic AI in SAP refers to AI sys­tems that can autonomous­ly per­form tasks, make deci­sions, and learn from inter­ac­tions, offer­ing more dynam­ic solu­tions com­pared to tra­di­tion­al AI.

Q: What are the ben­e­fits of inte­grat­ing SAP’s gen­er­a­tive AI into busi­ness operations?

A: Inte­grat­ing SAP’s gen­er­a­tive AI can lead to improved effi­cien­cy, enhanced deci­sion-mak­ing, and the abil­i­ty to gen­er­ate new insights from exist­ing data.