Evolving Dual Track: How AI Tooling Changes Discovery and Delivery
Product teams have spent over a decade refining how discovery and delivery work together. The Double Diamond, introduced by the British Design Council in 2005, gave us a shared language: divergent thinking to explore problems, convergent thinking to focus on solutions. Discover, Define, Develop, Deliver – four phases, two diamonds.
Dual Track Agile, shaped by Marty Cagan and Jeff Patton around 2012, evolved this further. Instead of sequential phases, two parallel tracks running continuously. Discovery figures out what to build while Delivery builds it. Teresa Torres made it operational with Continuous Discovery – the Product Trio collaborating weekly, testing assumptions as ongoing practice rather than a phase that ends.
The coordination is demanding. Discovery needs to stay two to three sprints ahead, feeding validated opportunities into Delivery without blocking it. For teams that have figured this out, it works.
2025 changed the economics of both tracks.
The shift
Two developments emerged that affect how discovery and delivery operate in practice.
The first is vibe coding. Andrej Karpathy coined the term in February 2025 for exploratory, AI-assisted building where you describe what you want in natural language, generate code, run it, and iterate based on what you see. The approach deliberately skips deep code review in favour of fast experimentation.
The second is spec-driven development. By mid-2025, tools like GitHub's Spec Kit and Amazon's Kiro formalised the counter-approach. You write detailed specifications first, then AI generates production code against that contract. The output is structured, documented, and maintainable.
These aren't competing philosophies. They're different modes for different purposes.
Two modes, two tracks
Vibe coding fits Discovery. You're exploring problem spaces, testing assumptions, and building artifacts to validate direction. Speed matters more than architecture. When you can go from concept to testable prototype in an afternoon instead of a week, the explore-and-validate loop tightens considerably.
Spec-driven development fits Delivery. You've validated what to build, and now you need production-ready systems. The specification becomes the source of truth, and quality, security, and scalability matter here.
This maps directly onto Dual Track. Discovery uses vibe coding for rapid exploration while Delivery uses spec-driven approaches for reliable execution.
The updated model
The basic Dual Track structure remains intact, with Discovery running ahead and Delivery working in parallel. But Discovery prototypes can now be functional code rather than wireframes, and Delivery can start from working artifacts rather than documents. The translation step between tracks gets smaller.
Think of Discovery as a foundation layer that runs continuously and stays slightly ahead. Vibe coding belongs here for rapid prototyping and validation. Delivery runs parallel but offset, using spec-driven development to build production systems from validated inputs.
The Double Diamond phases map onto this as well. Discover and Define happen in the foundation layer, while Develop and Deliver happen in the execution layer.
Three things to adjust in practice. First, let Discovery prototypes get closer to real code. Second, treat specification quality as a core skill for your Product Trios. Third, experiment with tighter feedback loops as both tracks accelerate.
One thing to be cautious about: don't mix up the modes. Using vibe coding in Delivery creates technical debt quickly, and applying spec-driven rigour too early in Discovery kills the exploration you need. The discipline is knowing which mode fits which track.
Where this goes
Looking ahead, protocols like MCP could simplify the handoff between tracks even further. When tools connect directly – Figma's design system feeding into AI coding tools, for example – the translation step between Discovery and Delivery shrinks.
For mature products with established design systems, this might mean that Discovery prototyping can happen with real components from the start. In brownfield environments especially, vibe coding's exploratory looseness could become less necessary when the system already encodes your patterns. The boundary between exploring and building would thin naturally.
Wrapping up
The Double Diamond gave us phases. Dual Track gave us parallel tracks. The evolution now emerging is tracks that share context continuously, connected by design systems that both humans and AI tools can work with. The foundations are there – the workflow just needs to catch up.