Gartner’s ‘Gen AI Disillusionment to AI-Native Engineering’ Evolution

At the Gartner Symposium in Barcelona I was excited to attend a talk by Gartner Analyst Haritha Khandabattu where she covered the Gartner Hype Cycle for Artificial Intelligence, 2025. We have been using the hype cycle model to help customers navigate adoption of AI in software engineering throughout 2025 so I was keen to understand how our experiences lined up with the wider market that Gartner is able to engage with.

Haritha covered a broad range of AI technologies and use cases, but one progression in particular resonated with me and perfectly explains what we’re seeing in enterprise adoption of AI for software development.

The Critical Shift: Gen AI Meets Reality

Haritha’s placement of Gen AI heading toward the trough of disillusionment while AI-native software engineering enters at the beginning of the hype cycle tells an important story. This isn’t coincidence – it’s cause and effect.

Here’s what I believe has happened, and this is informed by many of our real world experiences with customers at Equal Experts:

The initial excitement around Gen AI for software development led to what many now call “vibe coding”. Which is when developers use AI to generate large chunks of code with minimal oversight, trusting the output based on feel rather than rigorous validation. For personal projects and prototypes, this worked brilliantly. But when enterprises tried to scale this approach for production systems, reality hit hard.

Ben Wilkes captures this in his post  The trouble with vibe coding. As he points out, commercial software that must be maintained by multiple teams, tested thoroughly, and extended over time requires discipline. The non-deterministic nature of AI-generated code, the “code wall” phenomenon when something breaks, and the testing blindspot all contributed to growing disillusionment with Gen AI’s initial promise of effortless productivity gains.

The Birth of AI-Native Software Engineering

This disillusionment, however, has led to something valuable: the emergence of AI-native software engineering as a distinct discipline. Haritha described this as moving to a state where “the default mode of all your software engineers is AI first” – but crucially, this isn’t about abandoning engineering principles. It’s about evolving them. The Gartner Hype Cycle for AI in Software Engineering, 2025 defines AI-native software engineering as “an emerging set of practices and principles that are optimized for using AI-based tools to develop and deliver software applications”.

I believe this post from Jakob Grunig  From specification to code: A practical AI-powered workflow demonstrates one of the patterns emerging through this evolution. Instead of the free flowing chat based approach of vibe coding this team developed a disciplined workflow that used AI to manage and maintain a specification for the target system and then followed a step by step process that separates planning from execution. First, using the specification to generate an implementation plan that’s reviewable. Then execute against that plan by generating the code. Finally, generate tests systematically.

As Jakob notes: “You move faster not by cutting corners, but by cutting rework.” This is AI-native software engineering in practice – using AI’s power while maintaining the control and predictability that enterprise software demands. It’s the antithesis of letting AI “wander” through code generation based on vibes.

Why This Evolution Matters

This shift from initial hype to disciplined practice is precisely what the market is experiencing. As Phil Parker outlines in his latest post, From experiments to adoption: How AI in software delivery is growing up, the conversation has matured. Technology leaders are moving past simple pilots and are now wrestling with the practical, human challenge of responsible adoption at scale. CTOs are rightly still being asked “how are we using AI in the SDLC?”, but the leaders in this space are realising that sustainable gains come not from “magic” tools, but from the disciplined organisational behaviour that turns AI potential into lasting performance.

The good news? This progression through disillusionment is healthy and necessary, leading to something valuable: the emergence of AI-native software engineering as a distinct discipline. As Haritha noted in her talk, AI-native software engineering represents a state where “the default mode of all your software engineers is, AI first.” At Equal Experts we believe this is about establishing a robust systems discipline, moving beyond the hype of effortless productivity and grounding the power of AI in proven software engineering practices.

Other Key Observations from the Talk

Haritha made several other important points worth noting:

• AI agents remain at the peak of inflated expectations – powerful but not magical, with hidden costs that can escalate rapidly at scale
• FinOps for AI is emerging as a critical discipline, as demand outpaces per-unit cost reduction
• Model Ops is becoming the connecting tissue between pilots and production, though it remains “more art than science”
• Organisations remain structurally unready – with fragmented tools and governance frameworks that haven’t moved from principles to practice

Looking Forward

The shift that Haritha identified from Gen AI enthusiasm through disillusionment to disciplined AI-native software engineering represents the natural maturation of any transformative technology. We’ve seen this pattern before with Agile, DevOps, and Cloud adoption.

For those of us working in enterprise software delivery, the message is clear: embrace the discipline needed to overcome disillusionment. The organisations that will succeed aren’t those chasing the latest AI models, but those building robust engineering practices around AI tools – combining the power of AI with the discipline of proven software engineering. This applies to AI engineering as much as AI-native software engineering.

The future of software engineering isn’t about choosing between human expertise and AI assistance. It’s about evolving our practices to make both work together effectively. And that evolution is just beginning.