AI Principals

Solve real user problems first

Users care about outcomes, not the technology powering them. Identify the needs and pain points your AI feature should address before scoping the solution. Features that don't solve genuine problems will be ignored regardless of how sophisticated the underlying model is.¹

Design for transparency and explainability

Without explanations, AI systems are black boxes. Surfacing the reasoning behind outputs, even briefly ("Recommended because you used DSP..."), helps users form accurate mental models and decide whether to trust a result. Avoid hiding how the system arrived at its answer.²

Be honest about limitations

AI interfaces often overstate confidence in their outputs. Surface disclaimers, acknowledge knowledge boundaries, and avoid language that falsely implies human-like understanding. Overly confident framing leads users to trust outputs without appropriate scrutiny.³

Keep humans in the loop

AI predictions are not 100% accurate and require human involvement to monitor, correct, and improve over time. Build in lightweight feedback mechanisms, such as thumbs up/down or simple prompts, that capture user perspectives and feed continuous improvement.⁴

Give users clear paths to intervene or override

For high-stakes or ambiguous tasks, design pathways that allow users to pause or adjust. When dealing with uncertainty or ambiguity, the AI should escalate to the user. Autonomous AI features should never leave users with no way out, especially in agentic contexts where the system takes real-world actions on a user's behalf.⁵

Align AI with existing mental models and workflows

Reduce the learning curve by integrating AI features into familiar Basis constructs rather than replacing them with unfamiliar interaction patterns. Products that enhance existing workflows see stronger adoption than those that disrupt them.⁶