Dieser Blogpost ist auch auf Deutsch verfügbar
In the 2000s, the agile principles and practices documented in the Agile Manifesto significantly accelerated the development of new functionality. With the spread of cloud providers and the emergence of the DevOps movement, the time required for deploying new software and rolling out new features was also substantially reduced.
Organizations that have adopted these practices and internalized a culture of learning and experimentation sometimes deploy changes to production hundreds of times per day. They measure the impact of these changes and can quickly incorporate the results of these measurements. In times when time-to-market is increasingly important and the market and business environment are characterized by great uncertainty, it’s a significant competitive advantage to be able to respond quickly to unexpected changes and capitalize on opportunities that arise.
No wonder, then, that established companies that struggled to implement these principles and approaches were overwhelmed by the speed and adaptability of modern technology companies.
Even though the DevOps movement is now 15 years old, the DORA Report 2024 shows that a majority of respondents work in organizations that don’t have the capability to implement and roll out changes faster than within a week. A full 25 percent work in organizations where changes lead to failures in forty percent of cases, and it takes up to a month to fix such a failure.
However, adopting agile processes alone is not enough to improve the Software Delivery Performance measured in the DORA Reports ([1]: Forsgren et al., 2018). Technical approaches, changes in values, culture, and also the structure of an organization go hand in hand and mutually reinforce their effect—or hinder it.
The Reality: Many Organizations Lag Behind
In practice, I often see organizations that technologically have many capabilities from the DORA Capability Model. But due to a lack of adaptation in the organization and processes, they can’t really make use of these capabilities.
Common obstacles are:
- Rigid sprint planning with deployment at the end of the sprint
- Lack of autonomy and cross-functionality in teams
- Poor collaboration between domain experts and development team: The business department dumps requirements and then approves implemented features
In the worst case, there is still a separate ops team that is optimized for local efficiency.
What Does This Have to Do with AI?
Well, the promise of agentic AI, especially Agentic Software Engineering, is that software can be developed even faster. Companies that have established a culture of experimentation and learning and have built or restructured their organization accordingly will be able to benefit from AI agents.
Their feedback loop will be even shorter: they get feedback faster and can respond more quickly to it and to changes in the business environment. They can accelerate their antifragile tinkering ([2]: Taleb, 2012) and maximize the number of experiments they can conduct in a given period of time. This also significantly increases the probability that they will achieve a major success in the form of a positive black swan during this period.
The Danger: AI Without Organizational Maturity
However, companies that have not yet adopted or have only begun to adopt these cultures and practices risk being completely left behind soon. In such organizations, it doesn’t help to give development teams agentic AI tools.
Yes, this will enable these teams to develop faster. But is that your bottleneck? Can you keep up with formulating new requirements, testing and approving implemented features, and putting them into operation?
No, agentic AI doesn’t solve the problem of being left behind by modern technology companies and startups, and it doesn’t help you catch up.
Lessons from history: Technology changes organizations
As [3]: Reinertsen (1997) showed using the example of introducing CAD systems in design engineering, the introduction of a new technology always influences the processes and structure of an organization. Those who don’t allow this won’t be able to truly benefit from the new technology.
So it’s important not to put the cart before the horse and first work on the DORA capabilities that enable a climate of learning, rapid flow, and quick feedback. This includes some technical capabilities but also requires changes in culture, organizational structure, and processes.
Agency as a Key Concept
My colleague Hermann Schmidt recently pointed out in his blog post on The Promise of Agents the significance of the word Agentic:
Agentic is the adjective of agency, which has no equivalent in German. Agency encompasses the concepts of capability to act, autonomy, and actorship
Do your teams even have agency? If one agrees that AI agents accelerate software development because they have agency, then it should be clear that teams with agency are also needed.
Conclusion
For you to benefit from faster implementation with the help of AI agents, you also need quick, local decisions—that is, self-organized, autonomous product teams that combine all necessary competencies, have access to all relevant information, and can respond independently to opportunities and feedback.
None of this is new. But given that there is often still room for improvement in this regard, while AI agents are readily viewed as the sole solution to all problems, it’s important to point out: If these prerequisites are not met, AI will not be of much help. Or to put it another way: First become truly agile and develop agency yourself, then deploy software agents.
-
Nicole Forsgren, PhD, Jez Humble, Gene Kim (2018). Accelerate: The Science of Lean Software and DevOps: Building and Scaling High Performing Technology Organizations. IT Revolution ↩
-
Nassim Nicholas Taleb (2012). Antifragile: Things That Gain from Disorder. Random House ↩
-
Donald G. Reinertsen (1997). Managing the Design Factory. Free Press ↩