I started building custom software in 1995. Since then, I’ve seen this industry reinvent itself repeatedly. Some changes stuck, others faded, and most solved one problem while creating another. I’ve lived through CASE tools heavy waterfall methodologies, the Agile revolution, nearshoring, fully remote teams, and now artificial intelligence embedded directly into software delivery.
Each shift changed how we worked, yet none of them changed why projects succeed or fail. Now, with AI accelerating the pace of change, we are forced to confront something uncomfortable: speed has outgrown our process assumptions. This is why I believe AI is bringing back the need for upfront requirements—not as a regression, but as an evolution necessary for survival.
Why Agile Made Sense (And Still Does)
Agile became dominant because traditional waterfall approaches failed in practice. Incomplete requirements, shifting business needs, lengthy planning cycles, and late deliveries led to poor outcomes.
Agile addressed these issues through iteration, feedback, collaboration, and learning by doing. For decades, it was the best response to uncertainty and slow execution. However, Agile assumed development was the bottleneck, which is no longer the case.
AI Changed the Physics of Software Delivery
AI has fundamentally altered the speed of software development. Tasks that once took weeks can now take hours, and features that once took days can now be generated in minutes. When development accelerates this dramatically, a new problem emerges: the work around development becomes the bottleneck.
What we’re seeing across client engagements is that requirements can’t keep up with development speed. This causes business intent to become unclear, making developers wait for clarification, and rework increases instead of decreases.
Gartner identified AI-native software engineering as a top strategic trend for 2025, noting that AI is transforming the software development life cycle by embedding AI into every phase, from design to deployment.[1] The issue isn’t that Agile is wrong, but rather that AI changes where requirements clarity must exist.
Here’s what most organizations discover too late: Gartner predicts that by 2028, prompt-to-app approaches adopted by developers will increase software defects by 2500%, triggering a software quality and reliability crisis.[2] That’s not a typo—a twenty-five-fold increase. The reason is straightforward: AI generates context-deficient code that, while syntactically correct, often lacks awareness of broader system architecture and nuanced business rules, introducing subtle but severe flaws.[2]
In an AI-driven environment, Agile ceremonies that once managed slow execution and uncertainty now clash with development that happens 5-10 times as fast. The contrast is that while Agile managed uncertainty during slow builds, AI surfaces uncertainty faster, sometimes magnifying the risks that Agile was built to mitigate incrementally.
Why Upfront Requirements Matter Again
AI doesn’t operate like a human developer. It doesn’t ask clarifying questions, discern missing context, or “figure it out later.”
McKinsey’s November 2025 research found that the highest-performing AI-driven software organizations saw 16 to 30 percent improvements in productivity, customer experience, and time to market, with software quality improvements of 31 to 45 percent.[3] However, these gains only materialized when organizations fundamentally changed their processes.
AI requires explicit intent, structured inputs, clear constraints, and defined acceptance criteria. When these inputs are weak, AI amplifies the confusion at high speed. This is why AI brings us back to something many teams abandoned too quickly: thorough planning before development.
I’m not talking about months of static documentation or rigid plans that can’t adapt. I’m talking about clear, intentional requirements created upfront, so speed doesn’t turn into chaos or delays.
This Isn’t a Return to Old-School Waterfall
To clarify, I am not suggesting a return to 1990s-style waterfall. AI enables a different approach with clearly structured requirements first and and living artifacts instead of static documents.
McKinsey’s analysis shows that AI is changing the product development life cycle by shifting human effort toward areas that require deeper reasoning and problem-solving, with engineers moving from writing code to scoping requirements, determining system integration, and shaping solutions.[4]
AI enables teams to generate detailed requirements faster than ever, refine them continuously, keep documentation aligned with reality, and update intent as systems evolve. In other words, AI makes upfront clarity scalable rather than fragile.
Speed Requires More Discipline, Not Less
A common misconception in software development is that speed results from less structure. In reality, speed comes from clear intent, shared understanding, reduced rework, and fewer handoff failures.
AI exposes weak foundations instantly. If your requirements are vague, AI will move fast in the wrong direction. If your architecture is unclear, AI will generate inconsistency at scale. If ownership is fuzzy, AI will accelerate blame rather than outcomes.
Forrester’s September 2025 Developer Survey found that using AI, including generative AI, was a top objective for developers. Yet, adoption rates for AI-enhanced assistants and agents vary across different stages of the software development lifecycle, with coding farther ahead than analysis and planning.[5] Organizations that rushed to adopt AI for coding without addressing upstream processes are now dealing with the consequences.
Therefore, governance, clarity, and human accountability are more important than ever in an AI-driven environment.
What This Means for Teams Today
For organizations adopting AI, the question isn’t “Should we use Agile or Waterfall?” The real question is: Where does clarity need to exist so speed doesn’t create risk?
Based on my experience with clients during this transition, the answer lies upstream: in requirements, architecture, development standards and shared understanding before code is written.
McKinsey’s November 2025 State of AI report found that while 88 percent of organizations use AI in at least one business function, nearly two-thirds remain in experiment or pilot mode, with only about a third having genuinely scaled AI across functions.[6] The organizations succeeding at scale share a common pattern: they invested in process redesign before they scaled AI adoption.
AI benefits teams that plan before building and penalizes those who do not.
A Pattern I’ve Seen Before
Every major shift in software development follows a similar pattern: new capabilities emerge, teams misuse them, and discipline eventually returns, improved by experience. AI is no exception.
Successful teams will not discard Agile principles. Instead, they will integrate these lessons into a new model that prioritizes clarity, accelerates iteration, and improves outcomes.
Forrester’s early predictions warned that at least one organization would try to replace 50 percent of its developers with AI and fail, noting that developers spend only 24 percent of their time coding, with the remaining time spent on designs, writing tests, fixing bugs, and meeting with stakeholders.[7] That’s not going backward, that’s progress.
The Bottom Line
AI doesn’t replace experience, eliminate planning, or make intent optional. What it does is raise the cost of ambiguity.
As software can now be built at unprecedented speed, clear upfront thinking is essential. After so many years in this industry, I see this not as the end of Agile, but the next chapter in how custom software gets built, and like every chapter before it, the fundamentals still matter.
What Comes Next
At QAT Global, we help organizations navigate this transition by embedding AI into the software delivery lifecycle as a force multiplier within proven delivery models, rather than as an experiment.
We’ve learned something critical over the past year: AI requires better upfront requirements and also makes creating and maintaining them faster and more practical. Our AI-accelerated workflows compress requirements cycles, reduce handoff friction, and keep documentation aligned with reality. This is especially valuable for brownfield and legacy systems, where AI helps us derive requirements from existing systems instead of starting from scratch.
The organizations succeeding with AI-enabled development recognize that human judgment, clear requirements, and structured thinking create the foundation for AI to deliver real business value. Our approach keeps humans in control and accountable at every decision point, from business intent validation and requirements approval to architectural decisions, security reviews, and final code acceptance. At the same time, our developers are using AI to accelerate the work between those checkpoints.
This approach has resulted in delivery that is five to ten times faster without compromising quality or governance. Onboarding time has decreased from two weeks to three days, rework has been reduced, and developers can focus on higher-level tasks instead of searching for context.
If you’re seeking to scale software development and improve your ROI without sacrificing quality, we should talk. We’ve been building custom software since 1995, and we’ve learned that while the tools change, the principles that separate successful projects from failed ones remain remarkably consistent. What’s different now is that AI finally makes upfront clarity fast and scalable.
Ready for AI-enabled custom software development that actually works? Contact QAT Global to learn how our Diamond AI Solutions approach combines the best of human expertise with AI acceleration to deliver enterprise software that drives measurable ROI.
