Do you remember the uproar that Microsoft CEO Satya Nadella recently created about the death of SaaS? Although this is obviously some hyperbolic BS, there is a grain of truth in there.

Not the death of a business model, but something that is hidden somewhat deeper. It is about the eternal struggle between suites and best of breed software, which is in its third iteration – at least – since I am in the CRM world.

The grain of truth lies in the potential change of the way that business and other applications actually are built. When I look at ERP and CRM systems, I see software stacks that often are twenty years old, or even older. Look at S/4HANA. It still has a lot of R/3 inside although there are many new parts. Or Salesforce, which essentially dates back to the early noughts. SugarCRM has a fairly old core, too, even posterchild Zoho has software that dates back twenty years. The list goes on and on. Essentially, all these systems have morphed from fairly specialized solutions into monolithic giants.

Mind you, this is not all bad, as these systems are often incredibly powerful.

However, there are some challenges with it. I admit that, although I am a suite guy myself. These challenges are often around speed of innovation and, by definition, a degree of incompleteness. Requirements change faster than any software product vendor can implement them. And even if they successfully implement them, the software stack becomes increasingly more complex and harder to enhance and to maintain.

Harder to maintain not only for the vendor but also for the customer, even with the help of knowledgeable systems integrators. Additionally, there is always a functional gap, albeit sometimes only a perceived one. So, custom code and integration to specialized software stay a need. This need simply cannot be overcome by packaged applications.

The result is a big system of which often only a small percentage of the delivered – or custom developed – code is actually in use by customers.

There have been numerous attempts at changing this. We had COM and DCOM in the Microsoft world as well as similar technologies already last millennium, we have seen microservices and composable architectures. Hasso Plattner, the founder of SAP already in the early 2000s formulated a vision of an ERP that is entirely built on composable modules that get connected by workflows. SAP Business by Design was built following on this idea but didn’t achieve this vision.

In short, none of the attempts at solving the problem of bloating software systems actually solved this problem.

And now, we have a new technology with generative and agentic AI systems. This makes me ask a deceivingly simple question.

Do agentic AI systems solve the bloated applications issue?

During a recent CRMKonvo that I had with Bob Stutz, one of the most knowledgeable and influential persons in the CRM industry, Bob suggested that the technology clearly has the potential to achieve this. Vendors, however, need a mindset of solving actual business problems based on data. Bob is quite vocal about platforms having outlived their usefulness as they resemble band aids that are applied on top of other band aids. What is needed instead, are what he calls “focused applications” that push information to the user instead of making them pull it.

The consequence that Bob draws from this is that packaged applications, and especially platform-based applications, are reaching their end of life and need to be replaced by focused applications that are built with the help of – or even by – generative AI specifically to the users’ needs.

To quite an extent, I do agree, as I am talking about smart applications for quite a while now. I consider smart applications as applications that take work away from the users and adapt to changing user needs, based on data and use. Do we have them yet? No. Do we have the necessary bits and pieces, aka technology? Largely, yes.

So, on the outset, smart, or focused applications, look like a promising way to avoid application bloat. I also believe that agentic AI will play a major role in the way that these applications are built and delivered.

However!

Agentic systems, digital agents, can’t deliver this yet. There is some work to be done to make sure that they do not “stray off their reservation”. Agents inherently are probabilistic. This means that agents, and in consequence, agentic systems need to become more deterministic, i.e., correct, accurate, and reliable. Furthermore, the agents that make up the system need to interact in order to collaborate on a task. That requires three main ingredients, a protocol, common metadata and shared data. The protocol is a contract on how to interact plus a planning and orchestration entity to manage the agent collaboration. Metadata that supports business semantics and data require a joint access layer, let’s call it a (business) data cloud or a data fabric, or even a graph.

Combined, these ingredients form nothing else than a platform, but a platform with a twist. Ultimately, this is a (meta) data management platform with generative AI-based analytics and (no code / low code) development environment, combined with a telemetry system.

Thinking (or dreaming, if you like) this to an end, this platform can be used by vendors to create productized applications, and by consultancies or customers themselves to build highly specific applications. The more decentralized the application development becomes, the more focused the applications will be, and the less bloat there probably is. As the development itself is highly AI supported in this scenario, the development costs will become marginal. There are already technology vendors that consider a productivity increase of developers by a factor of 100 achievable. This productivity increase applies not only to developers in enterprise software vendor firms but to all others, which may fuel a technology change away from prebuilt solutions and back to custom applications.

So, to answer the question … in theory it’s possible, in practice, it will still take some time. There will need to be some governance around it to avoid another Lotus Notes graveyard of unused applications.

And, if it happens, the enterprise software world may very well get upended. Do I see this as very likely? No. Why not? There are at least three reasons.

Firstly, too many processes that run supported by software do actually not differentiate a business, i.e., makes an advantage that is enough to warrant a custom implementation. Second, what I call gravitational forces. Companies invested far too much into their systems to do a reimplementation anytime soon. Look at core banking systems. They still run on Cobol. A third one is that most businesses do not want to turn into tech companies. They want to use software, not create it.

Still, the world of enterprise software will change. What is your opinion? Custom applications, customizable applications, or anything in between? The judge is still out.