AI agents part 2: From product development to process optimization
AI agents can fundamentally change team coordination and digital processes through Scrum and other agile approaches.
(Image: Tatiana Shepeleva/Shutterstock.com)
- Thomas Immich
Some tech gurus believe that AI agents will soon make the role of the mid-level software engineer obsolete. Even if you think this is an exaggeration, agents are well on their way to revolutionizing software development, as the first part of our article series showed. There we also touched on the trend towards "multi-agent AIs", i.e. the combination of several AI agents to form a team. However, the systemic aspect of AI agents in terms of collaboration and automation is particularly complex and requires more detailed analysis.
Just like in team sports: it all depends on the coach
As in team sports, the success of a team of AI agents is likely to depend on who coordinates it and what priorities they set. It therefore depends on the coach or coaching team and their strategy and tactics.
In a figurative sense, this means that a coach who focuses more on defensive tactics is likely to focus his team on averting cybersecurity attacks and IT risks. A coach who focuses more on a good build-up and lots of (ball) passes is likely to be interested in the adaptability and transparency of his AI agents. And a coach who goes on the offensive certainly wants to shine with his software and outshine the competition with a measurable points victory.
Ultimately, it's about the team and each AI agent must play their part in the overall success. Outlining a strategy, implementing a tactic, reacting agilely to new framework conditions, but also iteratively correcting weaknesses after analyzing – are all aspects that can also be found in digital product development, just like in team sports. And just as in team sports, the following now applies to digital product development: during a sprint, the product owner can and should only intervene to a limited extent, as the development team must take responsibility for setting the course during the "game" (or sprint). The actual course is therefore set before and after the game. The procedure is similar to the UX Therapy AI experiment from the first part of the article series, meaning that only framework conditions are set within which the AI agents act autonomously.
Scrum in the age of AI
If AI agents develop software largely autonomously with their human development team members, a human product owner must be able to rely on the fact that all important information has flowed before implementation. There is a clear view of each player during implementation so that the team can address and remedy even the smallest weaknesses together after implementation.
It is no coincidence that Scrum, one of the best-known agile software development frameworks, takes its name from team sports: successful product development resembles a rugby game in which the team continuously moves the ball forward as a unit – as opposed to a sequential relay race. The word scrum refers to the "scrum" when the game is continued, for example after the ball is unintentionally lost.
Videos by heise
From digital product to digital production
But there are more relevant analogies besides team sports. The introduction of AI agents is obviously not that disruptive in terms of the systemic aspects of software development: MetaGPT describes similar roles and tasks for its AI agents and uses similar communication processes that would also be relevant in a human development team in an agile framework such as Scrum.
What is disruptive here is that AI agents, unlike their human team members, can be copied indefinitely and can be used for automation at all development levels. The development of digital products will therefore become unimaginably more scalable and efficient than ever before. In other words, AI agents in the development team will not only enable the automation of digital tasks in general, but of the entire software production process in particular. In the future, it will no longer be primarily about planning and building the digital product, but rather the digital production line (the pipeline), on which the digital product can then be produced automatically. And that changes a lot!
Change through automated industrialization
Accelerating manufacturing processes by automating them with production lines is as old as it is successful when it comes to physical goods and is part of the (once) economic success story of Germany and other industrialized nations. In the transformation known as "industrialization", manual processes were systematically automated or partially automated to meet high demand.
In my opinion, a similar change will also occur in software development with the advent of AI agents, as the demand for functional, emotionally appealing and intuitive software has been increasing for years and there is no end in sight. But the lack of competent specialists and the limited development speed of human development teams are preventing this demand from being met.
Germany is a prime example of this in a negative sense: The digitalization backlog is now so far advanced that the signs in both industry and society already seem to point to cynicism, resignation or retreat. We are too slow in everything and, beyond confidence, there is already talk of deindustrialization in some quarters.
