We are at an inflection point in enterprise software development that I have not seen in over two decades of building and delivering digital solutions. The question used to be whether AI could help developers write better code. That question has been answered. The new question, and it is a more consequential one, is whether organisations are prepared to fundamentally redesign the way software gets built.
New research from Bain & Company confirms what our teams at PhoenixDX have been experiencing firsthand: we are witnessing the emergence of what analysts call the AI Development Life Cycle, or AI-DLC. This is not an incremental improvement on what came before. It is a structural shift, from software development as a human-executed process supported by tools, to software development as an AI-orchestrated process guided by humans.
"The organisations dabbling with AI - for example, using AI for code generation in isolation while leaving the rest of the development process unchanged have space to capture savings beyond productivity. The gains on delivering outcomes for business faster and at a fraction of the cost are real."
- Pedro Carrilho, Founder & Managing Director, PhoenixDX
To understand what is changing, it helps to see the progression clearly. Software development has moved through three distinct eras, each representing a step change in how humans and technology work together to produce software.
The three eras of software development
Era 1 - Traditional SDLC
Humans execute every stage. Tools support but don't direct. Velocity is limited by team capacity.
Era 2 - AI-Assisted Dev
AI autocompletes code, suggests fixes, drafts tests. Humans still own the process end-to-end.
Era 3 ยท Now: AI-Powered / AI-First SDLC
AI orchestrates across the full lifecycle. Humans set direction, validate, and govern outcomes.
Most organisations today are operating somewhere between Era 1 and Era 2, using AI for isolated tasks such as code generation or automated testing, without integrating it across the lifecycle to create compounding gains. The Bain research is clear on this: most companies start by optimising a single activity, such as code generation, test creation, or requirements drafting. It works, and the results are measurable. But the organisations that treat the AI-DLC as a whole, embedding AI across every stage, not just the most visible ones, are achieving something categorically different.
The productivity gains already being reported are hard to ignore. Bain Research found that 63% of organisations report higher output per engineer, and 53% report faster release cycles and shorter time-to-market. These are not edge cases from AI-native startups. They are mainstream enterprises that have begun embedding AI systematically into how they build software.
What strikes me about these numbers is that they represent the early results, largely from organisations that have adopted AI in pockets rather than across the full development lifecycle. The ceiling is much higher. At PhoenixDX, we are seeing development time reductions of up to 50% and design phase compression of up to 75% when an AI-powered / AI-first approach is systematically embedded across requirements, architecture, build, and testing. The difference between those outcomes and the single-digit efficiency gains that most organisations are still seeing is not the quality of the AI. It is the scope of its use.
What organisations can experience with an AI-powered / AI-first approach |
|
|---|---|
50%
|
75% Design phase compression |
"We talk about AI productivity, but what we are really talking about is a different model of delivery. When AI is embedded across the full lifecycle - not bolted on at the code generation stage - the leverage is exponential, not additive."
- Pedro Carrilho, Founder + Managing Director, PhoenixDX
The AI-DLC is not simply a traditional SDLC with AI tools inserted. It is a reimagining of each stage of the development process around the genuine capabilities of AI, with human judgment applied where it adds the most value, and AI doing the heavy lifting everywhere else.
Perhaps the most significant structural implication of the AI-DLC shift is one that most organisations have yet to fully reckon with. For decades, software delivery has operated across two distinct tracks running largely in sequence: a product development lifecycle that determines what to build, and a software development lifecycle that actually builds it. The product defines the objective. Engineering executes it. The two teams operate with different rhythms, different tools, and different vocabularies.
AI dissolves this boundary. When a system can draft requirements, generate code, run tests, and iterate - all in a continuous flow - the rationale for keeping product and engineering as sequential, separate functions begins to disappear. The Bain research captures this shift directly: AI-enabled teams are moving toward an AI-powered / AI-first model in which defining, building, testing, and refining happen together in a workflow.
This is not simply a process change. It is an organisational redesign challenge. The roles of individuals shift. The structure of teams changes. The workflows that supported a handoff model need to be rebuilt around an integrated AI-powered / AI-first one.
The shift is not about replacing developers. It is about redeploying their expertise. When AI handles the generation, drafting, and documentation work, developers focus on the decisions that genuinely require human judgment: architecture choices, stakeholder alignment, edge-case resolution, and the creative problem-solving that no model can replicate.
There is a telling tension in the Bain research. While most organisations are currently seeing single-digit efficiency improvements from AI, their expectations are considerably higher. About half are hoping for a shorter time-to-market and more productive engineering teams. The gap between where they are and where they want to be is not a technology problem. It is an adoption and architecture problem.
Current gains based on Bain research. PhoenixDX outcomes based on client delivery experience.
Closing this gap requires two things: a deliberate decision to embed AI across the full development lifecycle rather than selectively, and the governance and operating models to support it. Without a modern, scalable technology stack, AI initiatives can stall and fail to deliver on the promise of speed and value.
"We built our AI-First Hub in Vietnam specifically to answer this question at scale. The teams we are assembling there are not just using AI tools - they are redesigning how delivery works, from the ground up, with AI at the centre of every stage. The productivity gains we are seeing validate that approach."
- Pedro Carrilho, Founder + Managing Director, PhoenixDX
For Australian IT and business leaders, the practical implication of the AI-DLC shift is straightforward to state and challenging to execute: the organisations that will lead their markets in three to five years are the ones making deliberate decisions about AI in their development processes right now. Not pilot decisions. Not tooling decisions. Strategic decisions about how software gets built, governed, and evolved across the entire lifecycle.
The question is no longer whether to pursue an AI-powered development lifecycle. This is a once-in-a-moment disruption that is very different from prior technology cycles. The question is how quickly and how deliberately to move, and whether your development partner has the expertise, the tools, and the proven methodology to take you there.
Ready to redesign how your software gets built?
PhoenixDX has been building an AI-powered /AI-first software development capability for years. Talk to us about how the AI-DLC can work for your organisation.
About the author:
Pedro Carrilho
Founder + Managing Director, PhoenixDX