Most development teams thought they had the software delivery lifecycle reasonably under control. Then generative AI changed the equation, and suddenly, the constraints that defined software development for two decades started bending in ways that feel both promising and, if we’re honest, a little unstable.
On a recent episode of The Amplitude of Tech, host Shawn Cordner sat down with Vin DiPippo, Chief Technology Officer at Vertikal6, to break down what’s actually happening inside development organizations right now. DiPippo has watched the full arc of AI’s entry into the SDLC, from early copilot experiments to the rise and fall of vibe coding, and he has clear opinions about where this is headed and what leaders need to do about it.
The conversation covers why AI is both a tool and a crutch, why prompt-driven development doesn’t scale, and what a structured, spec-driven approach actually looks like in practice.
Watch the Episode: From Vibe Coding to Business Value
Vin DiPippo joins host Shawn Cordner on The Amplitude of Tech to unpack how AI is reshaping software development, and what IT leaders need to structure now to capture the value without introducing new risk.
Subscribe to The Amplitude of Tech wherever you listen to podcasts. Apple | Spotify | YouTube
| Quick Answer: AI is changing software development not by replacing developers, but by shifting how work is structured. Teams that jump straight to code using open-ended AI prompts risk producing fragile, inconsistent systems. The organizations seeing real results are using AI in a spec-driven model, defining requirements and architecture first, then letting AI generate code within a well-understood context. That discipline separates experimentation from operational advantage. |
AI in Development Is Both an Accelerator and a Risk
Most development teams are already experimenting with AI in some form: code assistants, copilots, prompt-driven generation. The results are mixed, and DiPippo explains why.
AI behaves like your team’s most junior and most senior engineer at the same time. It can produce elegant, idiomatic code in seconds. It can also generate something completely unusable if the guidance isn’t clear. The difference comes down to how it’s used.
Some developers use AI as an accelerator, exploring unfamiliar frameworks, generating scaffolding, validating approaches. It reduces friction without replacing judgment. Others treat it as a shortcut: prompt, copy, paste, and hope.
That’s where things break. The risk isn’t that AI produces bad code. The risk is that it produces convincing code that hasn’t been critically evaluated. DiPippo draws a parallel to the old habit of stitching together forum solutions without fully understanding them. AI just makes that behavior faster, and at a much larger scale.
Why Vibe Coding Doesn’t Scale Past Simple Tasks
Early in the generative AI wave, there was real excitement around what DiPippo and others call ‘vibe coding,’ describing what you want, refining it through conversation, and watching the application evolve. On the surface, it looked like a revolution.
In practice, it exposed a fundamental limitation. AI operates probabilistically. Each iteration introduces small deviations. Over time, those deviations compound. What starts as a coherent system can quickly become inconsistent, fragile, or stuck in loops of conflicting logic.
For small, contained tasks (generating a function, analyzing a snippet, scaffolding a component), vibe coding works well. But for complex systems, pure prompt-driven development doesn’t scale. The industry is already moving past it.
Spec-Driven Development: What the Leading Teams Are Doing Now
What’s emerging is something more structured, and more aligned with how experienced teams already think. Instead of jumping straight to code, leading organizations are using AI across the full development lifecycle, in stages.
It starts with the business problem. Not in vague terms, but in detailed, validated specifications. AI becomes a collaborator here, helping refine requirements, surface gaps, and articulate intent clearly.
From there, architecture decisions are made deliberately. What stack makes sense? What database? How should the system scale? These are still high-impact decisions, but AI can contribute informed perspectives based on patterns it has seen across a massive body of code.
Only after that does implementation begin, and when it does, the AI isn’t guessing. It’s operating within a well-defined context. The result is code that’s easier to validate, more consistent, and far less likely to introduce hidden complexity.
DevOps Doesn’t Go Away. It Gets More Critical.
This shift doesn’t eliminate DevOps. It expands it. If anything, DevOps becomes more critical, because now you’re not just managing code pipelines; you’re managing how AI is used across those pipelines.
That includes governing how specifications are created and validated, standardizing how AI tools are introduced across teams, ensuring consistency in architecture and implementation, and embedding quality control earlier in the lifecycle.
Without that structure, fragmentation becomes a real risk. Different developers using different tools, different prompting strategies, different assumptions. You end up with systems that technically function but lack cohesion.
DiPippo frames it plainly: it’s the same problem DevOps was originally designed to solve, just in a new form.
The Talent Equation Shifts from Memorization to Judgment
One of the more interesting side effects of AI in development is how it changes what skills actually matter. Historically, deep expertise in a specific language or framework was a differentiator. AI flattens that curve.
Developers can now work across languages and frameworks more fluidly. AI bridges the knowledge gap. That opens the door to more generalist teams and changes how organizations think about hiring, training, and cost.
But there’s a tradeoff. While AI lowers the barrier to entry, it raises the importance of critical thinking. Understanding why something works matters more than ever, because the code might look correct even when it isn’t. The skill set shifts from memorization to evaluation and judgment.
What IT Leaders Should Be Governing Right Now
DiPippo is direct about the stakes. The question isn’t whether to adopt AI in development; that decision is already made across most organizations. The more important question is how to operationalize it without introducing new risk.
That means asking three things: Are we standardizing how AI is used across the development lifecycle, or leaving it up to individual developers? Do we have clear checkpoints for validation, or are we trusting output because it looks right? Are we accelerating delivery at the expense of long-term maintainability?
The paradox is simple. AI can reduce complexity, or it can hide it. In most environments, hidden complexity is what creates the biggest problems down the line. Structure isn’t the enemy of speed here. It’s what makes speed sustainable.
Frequently Asked Questions
What is vibe coding and why is it a problem for software teams?
Vibe coding is the practice of prompting AI to build applications conversationally, iteration by iteration, without a predefined specification. The problem is that AI operates probabilistically, meaning each iteration introduces small deviations that compound over time. What starts as a coherent system can quickly become fragile or stuck in conflicting logic, especially for complex applications.
What is spec-driven development and how does it differ from AI-only coding?
Spec-driven development uses AI across the entire software development lifecycle, beginning with detailed business specifications before any code is written. AI helps refine requirements, inform architecture decisions, and then generate code within a well-defined context. This produces more consistent, maintainable output compared to open-ended prompt-driven generation.
How should DevOps teams adapt their processes for AI-assisted development?
DevOps teams need to expand their scope to govern how AI tools are introduced and standardized across the development lifecycle. That includes establishing how specifications are created and validated, ensuring consistent architecture decisions, and embedding quality control earlier in the pipeline. Without this structure, teams using different tools and prompting strategies can produce fragmented codebases that technically function but lack cohesion.
Does AI in software development reduce the need for experienced developers?
Not exactly. AI lowers the barrier to working across languages and frameworks, which enables more generalist teams. But it raises the importance of critical thinking. Code generated by AI can look correct even when it isn’t, so the ability to evaluate why something works matters more than ever. The skill set shifts from memorization to judgment.
What should IT leaders be asking about AI governance in software development?
Leaders should ask whether AI usage is standardized across the development lifecycle or left to individual developers. They should verify that clear checkpoints exist for validating AI output, and that accelerating delivery isn’t coming at the expense of long-term maintainability. AI can reduce complexity or hide it, and hidden complexity is what creates the biggest problems at scale.
Amplix Helps Organizations Build Structure Around AI Adoption
Vin DiPippo’s perspective from Vertikal6 reflects what Amplix sees across client conversations every day: AI in software development isn’t a technology problem; it’s a governance and strategy problem. The organizations getting the most value aren’t moving the fastest. They’re moving the most deliberately.
Amplix works with IT and operations leaders to bring that structure to their AI initiatives, aligning tooling, processes, and decision frameworks to the same principles that made DevOps effective in the first place.
Contact Amplix today to talk through your AI strategy and where structure can turn experimentation into a real operational advantage.
