Why true AI transformation feels uncomfortable

What AI-assisted coding reveals about behavioral change, stress, and sustainable productivity

Teaser

In workshops, AI-assisted coding often works remarkably well. In everyday work, however, relapses and uncertainty emerge—not due to a lack of will, but because our work is currently undergoing a fundamental shift.

In my work as an AI Coach, I have observed a recurring pattern: Knowledge of AI tools and methodology is built quickly. The real challenge begins where new ways of working must withstand the pressure of real-world demands.

AI-assisted coding is not the endpoint of this development. We are at the beginning of a transformation that will change the way we develop software far more deeply and extensively. This article describes why genuine AI transformation often feels uncomfortable—and why exactly this discomfort is not an obstacle, but a necessary signal.

My Experience

As an AI Coach, I have facilitated various formats for introducing AI-assisted coding. Classic one- to two-day workshops have proven to be a useful starting point. They provide an overview, reduce inhibitions, and motivate individual developers to integrate AI tools and methods into their daily routines. At the same time, feedback from client projects revealed a recurring pattern: The initial momentum rarely led to sustainable change. After a short time, usage often plateaued among a few enthusiasts, while the majority of teams fell back into familiar ways of working.

Research on training effectiveness has shown for decades that this is exactly where the core problem lies: knowledge alone rarely leads to stable behavior in everyday work. The decisive factor is the transfer into real work situations—and this transfer frequently fails to materialize with one-off measures ¹,².

In response, we at codecentric developed a more intensive format called the AI Sprint. It combines an initial workshop with team coaching over one or more sprints—including a daily Community of Practice and a final review. The goal was not just knowledge transfer, but actual work with AI tools in a specific project context. The results were significantly better. Through close guidance, we were able to address project-specific needs, share experiences within the team, incorporate current developments from the fast-moving GenAI world, and deliberately create space for experimentation. Teams reported higher speeds in ticket processing, resolved legacy bugs, and successful software migrations.

And yet, even here, one thing became clear: habit is a stubborn opponent.

Knowledge != Change

In my private life, I have coached several people in losing weight or building muscle. The pattern was almost always the same: most of them knew exactly what they had to do. They knew the principles, had read the articles, watched the videos, and installed the apps. Failure was rarely due to a lack of knowledge, but rather the inability to consistently implement that knowledge in everyday life—especially when stress, time pressure, or frustration came into play. In exactly those moments, people do not fall back on what they theoretically recognize as sensible; they fall back on what feels familiar and safe.

• "I’m short on time today, so I’ll just grab something quick from the shop around the corner."

• "The day was so stressful; I just need something sweet right now."

This exact pattern shows up in AI-assisted coding. The tools are available, training has been completed, and initial experience has been gathered. After a short time, many developers know that AI can fundamentally support them. However, as soon as real working conditions take hold—deadlines, context switching, pressure—this knowledge is often hardly utilized. Instead, you hear sentences like:

• "I tried prompting it; there were a lot of code changes, but it didn’t lead to the desired result, so I’d rather do it myself."

• "It just starts developing and uses libraries I don't even know; I feel like I'm no longer in control of the code."

• "Either it does too much or it doesn't do what I want, so I don't see any added value in working with these tools yet."

Under stress, established routines dominate, even when better alternatives are known. Stress and learning research clearly describe this behavior: new ways of working remain fragile as long as they have not become a habit ³,⁴.

AI adoption therefore rarely fails because of the technology. It fails because we expect people to be able to adapt their behavior under pressure—without supporting them in the process.

Identity Shift

The use of AI in software development demands more than new tools and methods. It demands a fundamental change in roles. Many developers have built a clear professional self-image over years: I analyze problems. I design solutions. I write code. AI-assisted coding shifts this self-image. Not because skills are being devalued, but because the nature of the work is changing.

A similar dynamic is seen in weight loss. Many people do not fail because of workout plans or nutritional knowledge, but because their self-image does not change with them. As long as you don’t perceive yourself as someone who lives differently, new routines remain unstable.

And the change in the AI development environment is not yet complete. What we are currently experiencing as AI-assisted coding is only a first step. Systems are evolving from supporting tools to actors that increasingly design, decide, and implement on their own. The focus is shifting away from the complete self-development of every solution toward interacting with a system, evaluating results, and deliberately delegating development responsibility.

This is not an additional skill.

This is an Identity Shift.

And that is exactly why classic training approaches fall short. They convey knowledge but do not address the change in self-image that makes new ways of working permanently viable.

The Coach

In weight loss, this is obvious. Most people know what they should do. Nevertheless, sustainable change often only succeeds when someone accompanies them over a longer period. Not because of more "know-how," but because regular check-ins take place, routines are reflected upon together, and training or nutrition journals serve as a basis for conversation. Experiences are shared; progress and setbacks are contextualized together.

It is no different with AI-assisted coding. What is missing is support in living this new way of working and these methods even when pressure, lack of time, and old habits take over. An AI Coach does not act as a "tool explainer," but as continuous support. Through regular check-ins, through sparring on specific problems in everyday work, and through deliberately created spaces for exchange. Successes are made visible, experiences are shared, and uncertainties are openly addressed. New ways of working are not demanded; they are continuously integrated into daily life. This is exactly how sustainable change is created.

Meta-analyses on the effectiveness of learning and coaching show that precisely this form of continuous support significantly improves the transfer of new ways of working—particularly through regular application, feedback, and situational support ⁶,⁷,⁸.

Real Transformation

Anyone who introduces AI only through tools and training underestimates what is actually happening here: an identity shift in software development.

Empirical research shows that sustainable learning and behavioral change depend significantly on psychological safety, time for application, and active support from leadership ⁵,¹.

AI transformation rarely fails due to a lack of tools. It fails because people are left alone with a profound change in the way they work.

If you would like to talk about how this change can be consciously initiated and sustainably supported in your teams, let’s get in touch.

Sources

¹ Baldwin, T. T., & Ford, J. K.
Transfer of training: A review and directions for future research
https://doi.org/10.1111/j.1744-6570.1988.tb00632.x

² Blume, B. D. et al.
Transfer of training: A meta-analytic review
https://doi.org/10.1177/0149206309352880

³ McEwen, B. S.
Physiology and neurobiology of stress and adaptation
https://doi.org/10.1152/physrev.00041.2006

⁴ Grossman, R., & Salas, E.
The transfer of training: What really matters
https://doi.org/10.1111/j.1468-2419.2011.00373.x

⁵ Edmondson, A. C.
Psychological Safety and Learning Behavior in Work Teams
https://doi.org/10.2307/2666999

⁶ Grant, A. M. et al.
Executive coaching: A randomized controlled trial
https://doi.org/10.1080/17439760902992456

⁷ Beidas, R. S. et al.
Training versus training plus ongoing consultation
https://doi.org/10.1176/appi.ps.201100401

⁸ Cepeda, N. J. et al.
Distributed practice in verbal recall tasks
https://doi.org/10.1037/0033-2909.132.3.354