Gojko Adzic: 5 predictions for requirements management in the next 20 years

Gojko Adzic's attempt at this year's REConf is a bold one: how will requirements management develop over the next 20 years? As a successful book author and long-standing expert in software-heavy product development, Gojko brings the right expertise to the table. But 20 years is a long time - especially when it comes to making predictions about technology.

Below is a summary of Gojko's four challenges and five predictions for the next 20 years in requirements management. To anticipate a bit: Interestingly, he does not see system modeling as an important part of this.

Gojko Adzic

Gojko Adzic is the author of several books on the subject of software. These include the ReConf relevant titles such as  Specification by example, Behavior Driven Development, Test Driven Development and Agile testing. 

The slides for the presentation Gojko has kindly made it available for everyone to download.

What is the problem?

The aim of requirements management is, to put it bluntly, to identify problems and solve them effectively for the stakeholders. This used to be easier than it is today. A classic example: We want a comfortable climate. The requirement for the heating is then to bring the temperature of the room to 21°C and keep it there.

At the decision-maker level, it is usually about creating value. In the heating example, this is easy and, above all, comprehensible. However, the more complex the systems become and thus the causal chains, the more difficult it becomes to analytically identify the right requirements.

Nobody understands AlphaGo

Domain experts and programmers consider AlphaGo's behavior to be flawed. But the behavior ultimately led to victory.

Gojko Adzic

As an example, Gojko Adzic cited the behavior of AlphaGoan AI that has beaten the world champion in the board game Go. AlphaGo made several moves that neither the Go experts nor the AlphaGo programmers understood at the time, or thought were incorrect. Nevertheless, it later turned out that it was precisely these moves that led to victory.

This raises the question: How can we specify a system properly if we don't really understand which behavior leads to the best result?

And that brings us to the first challenge:

Challenge #1: It will become much more difficult to know with certainty what is "right"

At AlphaGo, we could say that, given the current state of research, it is not always clear how a machine learning system arrives at a result. But there are also much simpler examples. The Color tone of advertising links on Google has led to sales differences of up to 200 million dollars guided. The arrangement of fields on forms can influence the number of successfully completed forms and much more.

With these and similar examples, it is pointless to try to predict user behavior. And that leads to the first prediction:

Important: This is about data modeling, not system modeling.

Challenge #2: The schedules are getting tighter

Is what we are doing effective? We often only know this (much) later. As an example, Gojko Adzic cited the problem we are all familiar with in some showers, where there is a time delay between adjusting the tap and changing the water temperature.

As a more serious example, he cited the BBC's "agile" IT project, which cost 75 million pounds without demonstrably adding value.

This is certainly largely due to a lack of competence. However, faster (and measurable) feedback would certainly have helped. And that brings us to the next prediction:

Challenge #3: Decisions will involve more and more people / concerns / disciplines

If we are to react more quickly, then we also need to make decisions more quickly. But even that is easier said than done.

Gojko gave an example from his current employer, a SaaS company that offers voice services. There, usage suddenly increased dramatically, by a factor of ten. The initial reaction was joy, as more users expressed a high level of interest. But since the users were free plans, these users were costing the company a lot of money.

Long story short, it turned out that a Youtuber had published a hack to bypass the usage restriction with VPN. Gojko was able to block VPN connections in a short time frame and turn the use of VPN into a commercial feature. This freedom of action was essential to deal with a situation that could have cost the company a lot of money - or even bankrupted it.

This brings us to the fourth challenge:

Challenge #4: The tools that can handle all this do not yet exist

Karl Klammer (Clippy) is an example of a tool that tried unsuccessfully to support users in their work back in the 1990s. And even today we have generative AI such as ChatGPT, which seems to do this much better.

Remarkable: ChatGPT passes exams from universities for law and economics.

But even there, there are dramatic mistakes and problems. In short, we need tools for all these challenges. And these tools do not yet exist, at least not today. Hence the prediction:

There will definitely be new tools, of course. After all, development doesn't stand still. However, Gojko assumes that these will be AI-based.

Responsible use of technology

The Emissions scandal has led to a VW employee in the USA being sentenced to prison. However, this was not the first time that a product manufacturer had cheated. Gojko Adzic gave another example nVidia, whose graphics cards recognized benchmarks and then faked a better performance. However, this led to a general shrug of the shoulders.

The difference? Surely polluting the environment is a bigger crime than frustrating a few gamers. But the other difference is that the nVidia scam took place 10 years earlier. The more software invades our daily lives, the more critically authorities and users judge such incidents. That led to another prediction:

Gojko gave further examples on this topic, such as the creation of non-existent customer accounts by the FinTech start-up Frankwhich also led to a prison sentence.

And even without deliberate fraud, the number of ethical questions that we have to ask ourselves during development is increasing. Gojko Adzic mentioned a variation of the classic Trolley problems.

And what about system modeling?

Personally, I am of the opinionthat we cannot master the challenges described here without a centrally managed system model. In this respect, I was surprised that the topic did not come up. I was able to ask this question in the subsequent Q&A session. Gojko considered the topic of modeling to be premature, especially because his answer implied that he had a graphical model in mind. He also pointed out that text, unlike (graphical) models, is also excellent for change management.

The framework did not allow for a longer discussion. In particular, text-based modeling languages (B, VDM) have been around for a long time, and SysML v2 will soon be one of them. Nevertheless, it is of course impressive how powerful modern generative AI is, even without a formal system model behind it. Personally, however, I believe that the approach without a system model will reach its limits relatively soon. Time will tell.

Conclusion: RE in 20 years

Predictions over a period of 20 years in today's world are more difficult than ever. Accordingly, the predictions were at a high level of abstraction. But especially when it comes to strategic objectives, such a view can help to see beyond the day-to-day hype.