Software Delivery in the Age of AI: The Next Evolution
19 Feb 2025
Artificial Intelligence has emerged as perhaps the most transformative technology in modern history, fundamentally changing how we approach work across industries. In software engineering, the promise of AI-powered code generation has captured imaginations with visions of dramatically reduced development times. Yet, the reality has proven more nuanced than the headlines suggest.
Beyond the Code Generation Hype
The current wave of AI tools predominantly focuses on individual developers' tasks – code completion, bug detection, and automated testing. While impressive, this narrow focus overlooks a crucial reality: software development is but one component in a complex value stream that brings ideas to life as working products.
Consider a typical organisation's path to production. Development represents just one stage in a journey that encompasses business analysis, design, testing, deployment, and ongoing maintenance. Optimising development in isolation, while valuable, can only yield incremental improvements rather than the orders-of-magnitude gains promised by some AI tool vendors.
The Evolution of Software Delivery
Over three decades, software delivery has undergone several transformations. From the rigid waterfall methodologies of the past, we've progressed through agile, DevOps, and now platform engineering approaches. Each evolution has brought us closer to a more holistic, efficient delivery model.
Today's best practices reflect this journey. Modern software organisations prioritise features based on quantifiable customer value, embrace collaborative design processes, and structure themselves around cross-functional teams aligned with business domains. They maintain clear paths to production, automated pipelines, and centralised platform capabilities that provide foundations for rapid delivery.
The AI Revolution: A New Shift Left
The agile revolution of the late 1990s represented what we call a "shift left" – moving traditional end-of-cycle activities earlier in the development process. Software engineers took on greater responsibility for testing and business analysis, while quality assurance professionals became more involved in product decisions.
We're now witnessing another momentous shift left, driven by AI. As language models approach and potentially exceed human capabilities in coding tasks, the role of software engineers is evolving dramatically. Rather than being replaced, they're becoming "AI Directors," orchestrating AI systems to achieve business objectives more efficiently than ever before.
This transformation fundamentally changes how software teams operate. The traditional "two-pizza team" of 6-8 people – with its mix of business analysts, designers, developers, QA engineers, and DevOps specialists – is evolving into a more streamlined "medium-pizza team" that leverages AI in revolutionary ways.
ArcFlow: A New Paradigm
This evolution demands a new approach to software delivery – one we call ArcFlow. At its core, ArcFlow represents a radical reimagining of team structures and responsibilities, where a streamlined "medium-pizza team" serves as a crucial bridge between business stakeholders and AI capabilities.
Each medium pizza team will be aligned directly with one or more business or product functions and clear stakeholder ownership within the business.
The key innovation in ArcFlow is the establishment of a strict, well-defined two-way communication protocol between the business and engineering team.
This isn't just another complex requirements gathering process – it's a structured interface that allows engineers to rapidly translate business needs into actionable technical directives. AI tooling facilitates this translation, enabling the engineering team to quickly transform business requirements into designs and implementations at unprecedented speed.
The engineering team's role evolves into that of "AI Directors," focusing on three critical functions:
Upstream: Engaging with business stakeholders through structured communication channels
Internal: Translating business requirements into technical architectures and constraints
Downstream: Directing AI systems through repository-based collaboration
This interaction with AI systems happens exclusively through branches in the code repository – a deliberate choice that enforces best practices and ensures everything is version controlled at source. This includes:
Application code and tests
Infrastructure as Code (IaC)
CI/CD pipeline definitions
Data Pipelines as Code
Technical documentation in markdown format
Architecture diagrams and design documents
Configuration files and environment specifications
API contracts and interface definitions
Monitoring and observability settings
By codifying as much of the entire value stream as possible within the repository, we create a single source of truth that both human engineers and AI agents can work with effectively. The AI takes on multiple roles simultaneously:
Generating both frontend and backend code
Implementing UI/UX designs based on requirements
Performing automated testing and quality assurance
Handling DevOps operations and deployment scripts
Creating documentation and performing code reviews
Creating and maintaining Data Architectures
This transformation enables a new kind of continuous collaboration. Features are described, designed, and developed in real-time, with user stories serving as a log of achievements rather than a prescriptive backlog.
Minimum viable products are deployed to production rapidly, with real-time usage metrics guiding decisions about iteration or rollout.
Implementing the Future Today
While this vision might seem distant, the pathway to implementation is clearer than you might think. It begins with establishing a centralised platform capability that can understand and template existing codebases, ensuring consistent standards even as AI generates new code.
The key lies in finding the right balance between AI capabilities and human oversight. While large language models excel at understanding intent and generating high-level solutions, they can struggle with detailed, deterministic code generation. The solution is a hybrid approach that combines AI's creativity with structured templates and human direction.
Products such as ArcPilot.ai, allow organisations to turn large swathes of their existing codebases into re-usable blueprints, to allow them to scaffold complex systems incredibly quickly.
Combined with a simple interface for capturing business context directly from business users through a well-defined contract, the engineering team can rapidly develop complex systems in lockstep with AI.
Looking Forward
The next evolution in software delivery isn't just about faster coding – it's about reimagining the entire value stream through the lens of AI-enhanced capabilities. By embracing this change thoughtfully and holistically, organisations can achieve the transformative improvements that AI has long promised.
This new paradigm represents a fundamental shift in how we think about software delivery teams. Rather than coordinating across multiple team members, the streamlined team leverages AI capabilities to handle technical implementation while focusing on higher-level business objectives and strategic decisions. The result is a more efficient, responsive delivery process that maintains quality while significantly reducing the coordination overhead of traditional team structures.
The future of software delivery is collaborative, AI-enhanced, and faster than ever – but it requires us to think beyond simple code generation to truly rethink how we build software. The teams that embrace this new paradigm today will be the ones setting the standards for tomorrow's software industry.