Speed through automation - but the right way!
Automation is an effective way to shorten development times and get feedback more quickly. However, establishing automation is time-consuming. Furthermore, non-functioning automation can mask problems, which in turn can cause delays or other problems.
The following is a guide.
Note: Next week, on 4.11. at 16:00 CET, the System-as-Code-Fest 2025 (online and free of charge). I will be giving a short talk there. Further presentations by OOSE (Tim Weilkiens), Bosch, ESA and many others. Register now! >>
DevOps: Pioneer for automation in product development
The urge to automate comes from software development. This is understandable; after all, the tools that programmers use to do their work are the same tools they use to implement automation, namely software code (or scripts).
The second time you find yourself doing something, automate it. Chances are excellent that if you do it twice, you'll end up doing it 100 times.
Neal Ford, “The Productive Programmer” (2008)
The idea of automating the entire development process dates back to the 1970s. However, the topic only really took off in the 2000s, accelerated by increasing complexity, but also by accessibility via the Internet and low-threshold access to the tools through Open Source.
The term DevOps refers to the automated interaction between software development (Dev), system administration (Ops) and other domains such as quality assurance.
Types of automation
Personal scripts: It is easy for a programmer to create automations for their own tasks. The Emacs and vi text editors from the 1970s already made this possible.
Team automation in software development: It becomes more challenging when teams want to automate their work. This is because control no longer lies with a single person and incorrect operation of the scripts can lead to problems. Furthermore, updating tools can lead to automation no longer working. Thanks to open source and good standards, this is no longer a problem in software development.
Automation of trimming tools: Unfortunately, there are many tools that are difficult to automate, either for historical or competitive reasons. Many older but still used tools follow the desktop paradigm without elegant automation or integration capabilities. Some tools also compartmentalize themselves to make it harder for customers to switch to another tool. However, a lack of integration options often also prevents automation with external tools.
Hardware automation: So far, we have primarily focused on the automation of software. But there is also the possibility of automating hardware. In development, this often plays a role in testing (another case study on this in a moment).
5 steps to effective automation
Even if I don't like quoting Elon Musk these days, the one he used at SpaceX process is unquestionably effective. What is particularly noteworthy is that the actual automation is the fifth step, not the first. The steps are:
1. question every requirement
Before anything is automated, it must be clear, Why it exists at all. Every activity, every process step, every rule should have a clear purpose and a responsible person. Only when the purpose is comprehensible and up-to-date is it worth thinking about optimization or automation.
2. remove everything that can be omitted
Every process tends to grow. Over the years, checks, forms and approvals are added, and rarely does anything fall away. That's why we should No fear of deletions have: If you don't have to add anything back later, you have removed too little.
For product development, this means that we should delete unnecessary reviews, redundant reports and duplicate data entries, not optimize them.
3. simplify and optimize
Only now is it a matter of making the remaining steps simpler and more robust. This applies to processes as well as interfaces and tools. Redundancies, unnecessary waiting times and the like should be radically simplified, because every simplification accelerates subsequent automation and reduces sources of error.
4. accelerate cycle time
Once processes have been streamlined and simplified, they can be accelerated. The aim is to shorten feedback loops. This is possible by parallelizing and consistently optimizing the flow of data, tests, components and everything else that blocks value creation. In practice, this is often the point at which teams noticeably gain speed for the first time, even before automation has begun.
5 Automate - but in a targeted manner
Now comes the step that we are actually talking about here: Automation. But only now, when everything superfluous has been removed, simplified and accelerated, does automation unfold its full effect.
Three examples of automation in practice
There is a lack of good examples in the hardware sector in particular:
Inspection of weld seams at SpaceX
In this video reported Joe Justice among other things, how SpaceX has improved the quality of the weld seams. Originally, the parts of the rocket were welded with robots and then x-rayed at a different location to assess the quality.
Quality control was identified as a candidate for automation. To this end, the engineers mounted an X-ray machine directly behind the welding head of the welding robot. Seconds after the seam was welded, it was also x-rayed. This procedure made it possible to collect data on a completely different level: in particular, the welding parameters including welding sensors were now available at the same time as the X-ray data and could be assigned with millimetre precision. This enabled processing with machine learning, which led to drastically better quality, not to mention the time savings.
Design compiler for the parameterization of pump designs
This case study mentioned Joachim Pfeffer in our interview with SE-Trends. The Design compiler from IILS was created by Stephan Rudolph to obtain finished designs based on product parameters or to optimize cable harnesses. One pump manufacturer has applied this to a design that customers order in versions that are relevant to them with different pump capacities, discharges and the like.
Once parameterized, the system is fed with the customer parameters and spits out a finished design, a job that previously took weeks. However, a Doktorrand also spent a year parameterizing the design and preparing it for the design compiler.
Reduction of validation time from 12 to 2 months at ZF
Rene Honcak also presented an exciting case study at ReConf 2025, in which he showed how MBSE and digital twins can accelerate virtual verification and validation in automotive development. By linking product requirements with simulation models, a digital twin is created that precisely maps the physical product during the development process. This enables a well-founded release based on virtual tests.
Again, the effort required to reach this point should not be underestimated. According to Rene, it took several years to set up the infrastructure and optimize processes to achieve this level of automation.
Conclusion
Unfortunately, at least today, we cannot achieve the same level of automation in product development that we see in software development. But one thing is certain: to remain competitive, we need to gain product velocity. Automation is an important piece of the puzzle. But automation must take place in the right context, otherwise the energy will be wasted.
