Better ways to build products with AI
AI is fundamentally reshaping how products are built and what users expect, but we’re still in the early days—like the dial-up era—where chat is the dominant interface. As models advance and interfaces mature, AI will become as seamless and essential as the internet or cloud computing. This shift mirrors the move from waterfall to Agile, where new technical capabilities demanded new ways of working. Similarly, AI requires fresh paradigms for product development.
To explore this, we studied practices across 20+ companies, from startups to Fortune 500s, and distilled the lessons—much like the Agile Manifesto—into guiding principles for building products in the age of intelligent tools.
Challenges
Organizations face distinct hurdles when integrating AI into their product development:
People & Culture
- Internal team resistance stemming from job security concerns and unfamiliarity with AI tools
- Fundamental shifts in required skillsets and team responsibilities
Technical & Process
- Technology evolution outpacing organizational learning and adaptation cycles
- Quality assurance complexity due to AI’s non-deterministic nature and rapid output generation
Business & Legal
- Legal considerations for both customer-facing AI features and internal tool usage
- Competitive pressure from AI-native startups that can develop products faster than established companies
These challenges require new approaches to team structure, development processes, and organizational change management.
Principles of AI Product Development
Foster a culture of experimentation
In the fast-moving world of AI, fostering a mindset of experimentation accelerates learning, builds trust, and turns skepticism into buy-in by proving value through real results rather than theory.
Implement short iterative cycles
AI systems behave non-deterministically, making rapid feedback loops essential to reveal actual behavior. Their speed enables real-time iteration, such as adjusting a prototype during a user session instead of waiting weeks.
Maintain documentation as rigorously as source code
In AI-powered development, where instructions are written in English rather than code, rigorous, version-controlled documentation is essential to track how guidance evolves and which instructions drive which behaviors, creating a virtuous cycle where AI both depends on and helps maintain this documentation.
Deploy AI broadly to capture compounding improvements
Integrating AI at every step ensures the whole development process rises with each model advance, just as cloud adoption lifted all applications through continuous infrastructure upgrades.
Establish clear human accountability and oversight
AI cannot bear responsibility or accountability for outcomes or errors. Organizations must ensure humans own decisions, with explicit roles for review and approval at every stage.
Advance through incremental changes
AI makes sweeping changes deceptively easy. Small, deliberate steps preserve cause-and-effect so errors can be diagnosed and improvements can be compounded over time.
Anchor team accountabilities in holistic product success
AI collapses traditional boundaries between product, design, engineering, and compliance, reshaping the skills each function must bring; success depends on treating these contributions as interdependent parts of a shared, holistic outcome.
Safeguard integrity with explicit, independent evaluation systems
Because AI generates outputs at high speed and with non-deterministic variation, integrity can only be preserved by enforcing explicit standards through evaluation models independent from those that create the results.
Make error tolerance explicit and contextual
Because AI systems are non-deterministic, performance can only be trusted when evaluated systematically across many examples; teams must define clear thresholds for acceptable error based on context and risk, recognizing that what suffices for a travel chatbot may fall far short in a medical setting.