Speed in software development translates directly into a strategic advantage because it enables rapid learning. Projects intended for quick development perform better because they allow teams to validate hypotheses before large-scale usage begins. The introduction of AI tools fundamentally changes how I approach this velocity. Since we do not spend that much time actually writing code any more, the heavy structures of traditional agile frameworks do not make sense.
Iterations are now happening in hours rather than days or weeks. This acceleration forces a complete re-evaluation of how we structure teams, measure success, and interact with the organisations that buy and sell software development services.
Redefining Agile and Project Management
The core problem in software development has shifted from writing the code to knowing what to build, defining the architecture, and testing the results. In large projects, the sheer volume of stakeholders slows down the process. Unless we reduce the number of stakeholders involved, the total project time will not significantly decrease even if the software is produced faster.
Agile frameworks were designed for an era where writing code was the primary bottleneck. When iterations happen in hours, rituals like stand-ups and sprint reviews introduce friction rather than alignment. It is time to strip away these ceremonies to match the pace of modern tools.
Project management as a distinct third role might be a function that can be largely automated and integrated directly into our development tools. I recognise that highly regulated industries still require manual oversight, but for standard business applications, automated governance is entirely sufficient.
A new project role might be focused more on "human orchestration" and knowing what to build and align the necessary context, rather than producing project-specific artefacts.
The Evolution of the Developer Role
The software developer role as we know it today is also changing. Everyone currently working as a developer will have to learn the skill set similar to the one of a tech lead. They must become capable of orchestrating AI tools working in parallel, reviewing their output, and taking responsibility for quality without manually writing the code.
I practise what I call "parallel agent orchestration": the active management of multiple automated systems to generate, test, and refine software architectures simultaneously. The developer transitions from a creator of syntax to a director of agents. This transition is difficult for us who have linked our professional identity to writing code, but it is a necessary evolution. Testing is also deeply impacted by this shift, moving from manual quality assurance to automated testing.
The Jevons Paradox in Software
As the cost of producing software decreases, the consumption and demand for it will increase. This economic principle is known as the Jevons Paradox. The global demand for software has never been limited by a lack of creativity, but by the high cost of production. To handle this increased volume, we cannot continue using historical models.
Organisations must become accustomed to trial and error. I see purchasers of development services frequently provide specifications generated by AI, which are based on outdated paradigms of how software used to be built. To adapt to this increased demand, I suggest shifting the approach in three ways:
- Abandon rigid specifications: Organisations must stop providing exhaustive feature lists generated from outdated paradigms.
- Embrace scope reduction: Teams must be willing to launch with less functionality to validate core hypotheses immediately.
- Adopt sequential stacking: Large initiatives must be broken down into discrete phases completed in days rather than months.
I utilise what I call "micro-waterfall stacking": a method of treating modern software development as a series of highly compressed, rigid projects executed rapidly to reach a final goal. This allows for constant adjustment between small phases, provided the organisation is willing to discard initial assumptions.
Instant Feedback Loops
Traditional metrics and testing methods do not fully capture the nuances of an automated environment. Methods like prolonged testing phases are too slow. We prefer to build instant feedback loops directly into the software development cycle to measure speed and impact immediately.
Consider a user interacting with a web-based knowledge portal. If a user queries a search bar and receives no relevant results, an AI agent analyses the query in real time. The agent immediately synthesises an answer using the internal company knowledge. It then automatically publishes a new entry to the public knowledge base. The very next person with a similar query finds the answer instantly.
I have seen this level of autonomous feedback loops with self-healing capabilities dramatically change the way software can evolve. I acknowledge that autonomous publishing carries risks of inaccuracy, but I treat it as a manageable problem and not a reason to avoid automation entirely.
Summary of My Position
Speed as a strategic advantage is no longer about typing code faster or managing backlogs more efficiently. It is about fundamentally restructuring the development lifecycle to accommodate instantaneous production. By reducing stakeholder bottlenecks, automating project management, and shifting developers into orchestration roles, we can meet the surging demand for software.
Organisations must move away from long-term specifications and embrace stacked, rapid iterations.
