Learn the principles behind trustworthy agentic AI

Design experiences around the assignment of work, the expression of intent, the setting of constraints, and the review of results, rather than requiring users to execute each step manually.

In agentic systems, value is created when users can define the desired outcome and rely on the system to carry out appropriate actions within agreed limits. The interface should therefore support delegation as a first-class interaction model.

When the system is operating asynchronously or outside the user’s immediate focus, it should provide persistent and proportionate signals that work is continuing.

Users lose confidence when delegated activity becomes invisible. Trust in autonomous systems depends in part on the user’s ability to understand that progress is being made, even when they are not actively observing the process.

The system should calibrate the depth and frequency of feedback according to whether the user is actively engaged, passively monitoring, or temporarily absent.

Not all moments require the same degree of visibility. Some activities require detailed understanding, while others require only reassurance and exception handling. Effective systems distinguish between these modes.

Provide the minimum information necessary by default, while enabling users to inspect additional detail when confidence, understanding, or intervention is required.

Different users, and the same user in different contexts, require different levels of transparency. The default experience should remain clear and efficient, while deeper inspection should remain available when justified.

The product should help users understand what the system can do, what it is currently doing, what it cannot do, and what conditions govern its behaviour.

Trust is strengthened when users can form accurate expectations. Systems that appear intelligent but remain poorly bounded create confusion, misuse, and misplaced reliance.

Present the state of the system in language and structures that are relevant to the user, rather than exposing low-level internals that do not support action or understanding.

Users need to understand operational truth, but not necessarily implementation detail. Good design translates machine activity into meaningful human-facing status.

Users should be able to examine how a result was produced when confidence, accountability, or decision quality is important.

Trust is not established through assertion alone. It is established by enabling proportionate verification. Particularly in high-impact contexts, users must be able to inspect evidence, actions, and changes.

When the system cannot proceed, the reason should be immediately visible, along with any action required from the user or another dependency.

In agentic systems, failure frequently arises not from incorrect reasoning but from unclear responsibility. Users must know when a task is paused, why it is paused, and what will resume progress.

Where work involves multiple steps, agents, dependencies, or concurrent activities, it should be represented as a structured system rather than solely as a message stream.

A conversational log is not always an appropriate representation for operational complexity. Users need mechanisms to understand relationships, concurrency, and progression across delegated tasks.

The system should support users not only in starting tasks, but also in guiding, refining, reprioritising, and correcting work while it is underway.

Prompting is an initiation mechanism. It is not, by itself, a sufficient control model for complex or consequential work. Users require the ability to steer ongoing activity without restarting the entire process.