Adaptive Software Development
Adaptive Software Development (ASD) is a software development framework created by Jim Highsmith in the late 1990s and published in his book of the same name in 2000. Its central contribution was to replace the classic plan-design-build cycle with a three-phase cycle that explicitly assumes uncertainty as a normal condition of the work: speculate, collaborate, and learn.
ASD draws on complex adaptive systems theory: the idea that in highly uncertain environments outcomes can't be predicted precisely but instead emerge from the continuous interaction between the team, the customer, and the product. This theoretical grounding places it among the most conceptually rich frameworks in the agile ecosystem, even though its practical adoption has been lower than that of Scrum or XP.
Core characteristics
- Work directed and guided by the project's mission, not by a detailed plan.
- Feature-based: the product is built and measured in terms of value delivered.
- Iterative development with short, timeboxed cycles.
- Risk-driven: decisions are made based on what is most likely to go wrong.
- Change-tolerant: change isn't an exception to be managed but a constant to be embraced.
- Focused on intensive collaboration between the technical team and the customer.
The three phases of the ASD cycle
Speculation
The term speculation deliberately replaces planning. Highsmith argues that in complex projects, talking about a "plan" implies a false certainty: what you're really doing is making an informed bet on how the work might unfold. Speculation includes:
- Project initiation: defining the mission and high-level objectives.
- Setting the project's overall time frame.
- Determining the number and length of iterations.
- Defining the goal of each iteration.
- Provisionally assigning features to each iteration.
Collaboration
Product development is carried out concurrently by the technical team and the customer. ASD stresses that collaboration isn't just a desirable practice but a structural condition of the method: without continuous access to the customer, the learning cycle breaks down.
Learning
At the end of each iteration, the team reviews four dimensions:
- Quality from the customer's point of view: does the product do what the customer needs?
- Technical quality: is the code maintainable and sustainable?
- Delivered functionality: has everything planned for this iteration been completed?
- Project status: what have we learned that affects the next iterations?
This review isn't just a validation step—it's the framework's main mechanism for adaptation.
ASD and AI product development
ASD's speculate-collaborate-learn cycle is especially relevant when developing products with generative AI, where uncertainty about how the system will behave is inherent and detailed planning is even harder than in traditional software.
Three concrete connections:
- Speculation and prompts: designing a prompt or configuring a model is, in ASD terms, pure speculation: you're making a bet on how the system will respond and iterating from the actual result, not the expected one.
- Collaboration with the customer: AI products need user feedback to assess the quality of the output, especially on subjective tasks. ASD described this need before generative AI even existed.
- Learning as the central mechanism: in AI development, the train-evaluate-adjust cycle is functionally identical to ASD's learning cycle. Teams already working with adaptive frameworks find it easier to integrate workflows built around language models.
ASD in perspective: where it fits today
ASD isn't a commonly adopted framework in 2025, and it has largely been absorbed by Scrum and other agile approaches that took up its principles in more concrete, operational ways. Its value today is mainly conceptual:
- For teams that want to understand the theoretical basis of agility beyond Scrum's ceremonies.
- For organizations working in highly uncertain environments where the idea of "speculation" is more honest than that of "planning".
- As a historical reference that helps explain why the Agile Manifesto (2001) was possible: ASD, along with XP, Scrum, and other frameworks, provided the intellectual substrate it emerged from.
Common mistake
Treating ASD as a complete operational method. Unlike Scrum, ASD doesn't prescribe specific roles, events, or artifacts. It's a conceptual framework that needs to be complemented with specific practices to become operational. Anyone looking for step-by-step instructions will find Scrum or XP more useful; anyone looking for the foundations to understand why agility works will find ASD a valuable read.
References
Highsmith, Jim.
- (2000). Adaptive Software Development: A Collaborative Approach to Managing Complex Systems. Dorset House Publishing.
- (2002). Agile Software Development Ecosystems. Addison-Wesley.
See also
Want to take your agility further? You can look for upcoming course and exam dates, or go at your own pace by becoming a Agile Club member. Membership includes exclusive resources, e-learning classrooms, and access to Skill Arena: a space to practice and measure your agile skills whenever it suits you.