Intent-Driven Development and AI Factory Economics
Hari Krishnan discusses the launch of 'Spectrum Development' and the shift toward intent-driven development. The analysis covers token economics, the role of BDD and TDD in AI harnesses, and the strategic implications of autonomous software factories for enterprise engineering teams.
The Shift to Intent-Driven Development
The software engineering landscape is undergoing a fundamental shift from instruction-based prompting to intent-driven development. As AI agents gain the capability to handle complex, long-running tasks, the primary bottleneck is no longer code generation but the clarity of intent. Hari Krishnan, author of the newly launched Manning Early Access Program book Spectrum Development, argues that explicit intent artifacts, such as intent.md, are critical for reducing the interpretation gap between human vision and agent execution. This approach moves beyond simple user stories to capture the directional and strategic intent of a feature, allowing agents to operate autonomously without constant human intervention.
Token Economics and Model Strategy
A significant operational challenge emerging is the economics of token consumption. The era of "token maxing" is ending, replaced by a focus on cost-per-unit-of-work. Teams must adopt a division of labor strategy: frontier models are reserved for high-value specification and exploration phases, while cheaper, lighter models handle execution and verification. This normalization of token costs is essential for sustainable AI integration, as different models consume tokens differently based on their architecture and efficiency. Organizations must monitor these metrics closely to avoid budget overruns and ensure that AI investments yield tangible productivity gains.
Engineering Rigor in Autonomous Factories
The concept of the "software factory" is moving from theory to practice, but it demands higher engineering discipline, not less. Autonomous agents expose legacy weaknesses in codebases, such as poor modularity and lack of deterministic tests. To build a reliable factory, teams must implement Behavior-Driven Development (BDD) to align agent behavior with specifications and Test-Driven Development (TDD) to ensure internal code structure. Furthermore, the factory itself becomes a product requiring dedicated engineering effort. Human roles are shifting from manual coding to oversight, strategy, and harness design, ensuring that the autonomous pipeline maintains quality and security standards.
Strategic Implications
For leadership, the adoption of AI-driven development is not just a tool upgrade but a structural transformation. It requires rethinking team organization, potentially moving away from siloed functional teams toward product-centric structures that can manage autonomous workflows. The success of these systems depends on the ability to define clear domain boundaries and maintain rigorous verification processes. Companies that invest in these foundational engineering practices will be better positioned to leverage the full potential of AI agents, turning them into scalable, reliable production assets rather than unpredictable experimental tools.
Key insights
-
Intent-driven development reduces the interpretation gap between human ideas and agent implementation by making implicit requirements explicit in artifacts like intent.md. This allows agents to work autonomously on complex features without constant human clarification.
Impact: Increases agent autonomy and reduces human bottleneck in the development cycle, enabling faster delivery of complex features.
-
Token economics are becoming a primary cost driver in AI development. Teams must normalize costs by assigning frontier models to specification and cheaper models to execution, as tokens are not fungible across different model architectures.
Impact: Optimizes AI spending by aligning model capability with task complexity, preventing budget overruns while maintaining output quality.
-
Combining BDD and TDD creates a robust harness for AI-generated code. BDD ensures macro-level behavioral alignment with specifications, while TDD enforces micro-level structural integrity and testability.
Impact: Improves the reliability and maintainability of AI-generated code by enforcing strict behavioral and structural standards.
-
Generic AI skills often fail in specific contexts. Tuning skills using mutation testing and coverage metrics is necessary to achieve consistent results, as agents require context-specific rules to avoid gaming the process.
Impact: Enhances the consistency and reliability of AI outputs by customizing agent behavior to the specific codebase and team practices.
-
Autonomous software factories expose legacy engineering weaknesses. Success requires rigorous domain boundaries, deterministic regression suites, and a shift in human roles from coding to oversight and harness design.
Impact: Prevents quality degradation in autonomous workflows by enforcing high engineering standards and clear human accountability.
Action items
-
Implement intent.md artifacts to capture explicit strategic and functional intent for each feature, reducing the need for iterative human-agent clarification.
Impact: Improves agent autonomy and reduces development cycle time by providing clear, unambiguous instructions for autonomous execution.
-
Establish a model division of labor strategy, assigning frontier models to specification and exploration tasks, and cheaper models to execution and verification.
Impact: Optimizes token costs and improves budget predictability by aligning model capability with task value.
-
Integrate BDD and TDD into the AI development harness to ensure behavioral alignment and structural integrity of generated code.
Impact: Enhances code quality and maintainability by enforcing strict testing and design standards on AI outputs.
-
Tune AI skills using mutation testing and coverage metrics to ensure consistent results within the specific codebase and team context.
Impact: Increases the reliability of AI agents by customizing their behavior to avoid common pitfalls and gaming strategies.
-
Reorganize team structures to support autonomous factories, focusing on domain boundaries and human oversight rather than manual coding tasks.
Impact: Aligns organizational structure with AI capabilities, enabling scalable and reliable autonomous development workflows.
Quotes
“intent is the ability to reduce that gap between your idea and the implementation by reducing interpretation that agent is trying to do”
“the execution part is what can afford to be on lighter models or cheaper models. The specification and exploration has to be more on the frontier models”
“the factory itself becomes a deliverable, right? And there has to be a set of engineers who work on that”