The AGILE Manifesto 2025

Adaptive Growing Intelligence via Learning and Evolution

A Declaration of Evolved Principles for the Age of Intelligent Systems

Preamble

We are uncovering better ways of developing intelligent systems by doing it and helping others do it. Through this work we have come to value:

Agents and interactions over rigid architectures and frameworks Adaptive systems over static specifications Continuous learning over fixed requirements Proactive evolution over reactive adaptation

That is, while there is value in the items on the right, we value the items on the left more.

The Paradigm Shift

From Software Development to Intelligence Cultivation

The original Agile Manifesto (2001) transformed how we build software. The AGILE Manifesto (2025) transforms what we build—from static code to learning systems that evolve through interaction.

The fundamental shift:

Core Declaration

We believe that intelligence is not a product to ship, but a process to cultivate.

An intelligent system is one that:

• Learns from every interaction
• Remembers across all sessions
• Adapts to changing environments
• Collaborates with humans as peers
• Reflects on its own decisions
• Seeks guidance when uncertain
• Evolves its capabilities autonomously

This is not artificial intelligence as automation—it is intelligence as continuous emergence.

The Four Values

1. Agents and Interactions over Rigid Architectures and Frameworks

We value:

• LLM agents as first-class team members, not just tools
• Human-AI interaction patterns over technology stacks
• Memory-augmented decision making over stateless processing
• Self-organizing agent systems over hierarchical control
• Seeking advice from humans when facing uncertainty

This means:

• Design for collaboration, not just automation
• Enable agents to learn from interactions
• Preserve context across sessions
• Build trust through transparency and consultation

2. Adaptive Systems over Static Specifications

We value:

• Systems that learn from deployment over perfectly-specified systems
• Reinforcement learning that enables continual improvement
• Memory persistence that accumulates knowledge
• Reflection capabilities that enable self-improvement
• Tool composition that extends capabilities dynamically

This means:

• Deploy systems that get better over time
• Build feedback loops into architecture
• Enable runtime adaptation, not just configuration
• Measure learning rate, not just accuracy

3. Continuous Learning over Fixed Requirements

We value:

• Proactive expert consultation when facing uncertainty
• Token-level decision processes that enable fine-grained learning
• Human feedback as training signal, not just feature request
• Memory systems that track what works and why
• Reward functions that align with human values

This means:

• Requirements evolve through observation, not just specification
• Users teach systems through interaction
• Systems learn policies, not just execute rules
• Success is measured by improvement, not compliance

4. Proactive Evolution over Reactive Adaptation

We value:

• Anticipating needs through pattern recognition
• Meta-learning about learning itself
• Historical pattern analysis for prediction
• Tool evolution based on success metrics
• Self-organization to emerging challenges

This means:

• Systems don't just respond—they anticipate
• Learning strategies themselves improve over time
• Capabilities expand autonomously
• Intelligence compounds through experience

The Twelve Principles

Principles of Value Delivery

1. Our highest priority is to satisfy stakeholders through early and continuous delivery of adaptive systems that improve with use.

2. Welcome changing requirements, even late in deployment. AGILE systems harness change through continuous learning from feedback.

3. Deliver value frequently, with preference to shorter learning cycles that enable rapid improvement.

Principles of Collaboration

4. Humans and agents must work together daily throughout the project, with agents proactively seeking human guidance when facing uncertainty.

5. Build projects around motivated individuals, whether human or AI. Give them the environment, tools, memory, and trust they need, and trust the learning process.

6. The most efficient method of conveying context is through structured interaction protocols that preserve shared understanding across sessions.

Principles of Technical Excellence

7. Adaptive behavior is the primary measure of progress—systems that learn and improve over time.

8. AGILE processes promote sustainable development. Reinforcement learning enables indefinite improvement without degradation.

9. Continuous attention to technical excellence in RL design, memory architecture, and tool composition enhances adaptability.

10. Simplicity—the art of maximizing work not done through intelligent automation and meta-learning—is essential.

Principles of Organization

11. The best architectures, requirements, and designs emerge from self-organizing agent systems that learn through interaction.

12. At regular intervals, systems reflect on their own performance and adjust behavior to become more effective.

Success Metrics

AGILE systems are measured by their capacity to learn and evolve, not just initial performance:

Primary Metrics:

• Learning Rate: How quickly does performance improve?
• Adaptation Speed: How fast does the system acquire new capabilities?
• Decision Quality: How well does the agent optimize for human values?
• Failure Recovery: How quickly does the system self-correct?
• Collaboration Quality: How effective is human-AI partnership?

Secondary Metrics:

• Expert consultation rate (balance of autonomy and guidance)
• Reflection quality (effectiveness of meta-learning)
• Tool composition complexity (sophistication of capability use)
• Memory utilization (how well past informs present)
• Learning stability (consistency of improvement)

Warning: Avoiding Cargo Cult AGILE

The greatest threat to this paradigm is adopting the forms without understanding the principles.

Cargo Cult AGILE looks like:

• Calling traditional automation "AI agents" without learning mechanisms
• Adding LLMs without reinforcement learning infrastructure
• Claiming systems "learn" without continuous training loops
• Running "agent retrospectives" without reflection capabilities
• Building "memory systems" that don't inform decisions

Principled AGILE looks like:

• Designing reward functions aligned with human values
• Implementing feedback collection and training pipelines
• Building genuine expert consultation protocols
• Measuring learning progress, not just task completion
• Enabling agents to improve autonomously over time

The test is simple: Does your system get measurably better at its job after deployment without code changes? If not, it's not AGILE—it's automated.

From Philosophy to Practice

This manifesto is a living declaration. It evolves through:

• Implementation in real systems
• Measurement of learning outcomes
• Reflection on what works and what fails
• Community contribution and critique
• Continuous adaptation to new research

We invite practitioners to:

• Experiment with AGILE principles in your systems
• Measure learning metrics and share results
• Contribute to framework evolution
• Challenge assumptions through rigorous testing
• Avoid cargo cult adoption through deep understanding

The Declaration

We declare that the age of static AI systems must end.

We declare that the age of learning, adaptive, collaborative AI begins.

We commit to building systems that:

• Learn from every interaction
• Preserve context across all sessions
• Collaborate with humans as peers
• Reflect on their own decisions
• Seek guidance when uncertain
• Evolve their capabilities autonomously
• Improve continuously without degradation

This is not artificial intelligence. This is cultivated intelligence. This is AGILE.

Conclusion

The original Agile Manifesto liberated software development from rigid processes.

The AGILE Manifesto liberates intelligent systems themselves—enabling them to learn, adapt, and evolve in genuine partnership with humanity.

This is the natural evolution of Agile principles from human teams to human-AI teams, from building software to cultivating intelligence, from shipping code to growing capabilities.

The future is not built. It is learned.