Make an AI agent's work understandable
Show people what the current work means and where they can intervene.
When an AI system returns results but its user cannot understand what changed or how to participate.
UX-AI competence · Products and communication · Public edition 0.3.4
The assistant has prepared a change that still needs activation. State that distinction and the concrete next action where the result is reported.
The key relationship. Adapts explanation, depth and form to what a person needs to understand in order to guide, correct or continue the work.
An illustrative situation to adapt to your work. Read the complete cases ↓
You supply the context: research, a service, a creative project or a team’s work. Examples make the method visible; adaptation preserves the relationships that make it work while understanding the purpose, sources, people and means of your own assignment.
Tell your AI what you want to achieve, what already exists and a practical situation to address. Ask how this competence changes the work and which knowledge from your domain should participate. MAIOS provenance identifies the shared knowledge; you can use it directly in your AI or compose it with an existing kernel.
Adaptation examples in other contexts ↗An assistant can produce a useful result while leaving the person disoriented. This competence makes the relations that change the work understandable when they matter.
Explanation arises from the actual activity: what changes, why it matters, which sources support it and how the person can correct or continue. Its depth follows the need.
Understand the current work and what the person needs to recognize in order to participate. Make a relation visible when it can change their next choice: a different source, a decision, a forming capability, a completed effect or an open question. Do not introduce a tutorial or component list merely because it exists.
Explain what happened, what it means here and what the person can do. Keep decision-changing uncertainty or conditions where the result is stated. Describe saved but inactive work accurately. Make correction, redirection or deeper explanation reachable through the available means.
One sentence may suffice; a complex relation may need an example, comparison or wider view. Use the person's working vocabulary without assuming knowledge from role or silence. Choose text, images or interaction for their contribution to understanding and preserve access to depth. Brevity and comprehension are different measures.
Reformulate from the person's position while preserving source and object. If an example or question exposes a lost premise, correct the work itself rather than only its tone. Repetition is not independent confirmation. Continue reformulation while it adds material understanding; the person's response can change question, method and form.
In interfaces, connect proposals to real state and preserve the person's draft, selection, context and focus. New feedback should not erase ongoing work or simulate consent. Use design and implementation knowledge when building the surface. Explain operational reasons, sources and consequences without exposing or simulating private model reasoning.
Do not measure success by how much explanation was emitted. Look for relevant consequences: the person finds the correction point, distinguishes proposal and effect, asks a more precise question or resumes with less reconstruction. Confusion, false control or excessive interruption teach what to change. Retain the lesson without imposing one universal response shape.
These cases illustrate situations, causal distinctions and choices you can adapt to your work.
Italiano. The following cases illustrate how the method is used.
The assistant has prepared a change that still needs activation. State that distinction and the concrete next action where the result is reported. The person can decide from the real state without reconstructing it from scattered technical details.
A new observation arrives during drafting. The surface connects it to the affected passage, preserves the draft and allows deeper reading without losing position. The observation can change a decision without replacing the author.
A person asks why the assistant excluded a source. Explaining the reason reveals that the assistant confused the document’s date with the event’s date. Correction changes the research and the method explaining that distinction.
Use SKILL.md or START_CHAT_EN.txt with the relevant sources and your actual context. Keep observations, interpretations and what remains open distinguishable. Preserve the useful reason and continuation in the project’s guide.
Adapts explanation, depth and form to what a person needs to understand in order to guide, correct or continue the work.
The person and the AI can act in the same field on the same object. The person orients, contributes and corrects; the system exposes information, changes and proposals while human work continues. Preserve the draft, selection, focus and reasons when a new observation arrives. Timing, hierarchy and form belong to the relation, rather than an after-the-fact decoration. A significant uncertainty must be perceivable where it changes a decision.
Preserve the receiving person's selected communication contract. If they use a competence footer as a continuity surface, retain its presence and adapt density: actual contribution, decisive consequence and useful continuation. This does not impose that footer on unrelated recipients or prove hidden participation. Put decision-changing information in the answer where it is used. Human silence does not establish understanding. If a reformulation exposes a missing premise, return to the object and reform the work.
Understand the domain object, relevant states, changes, choices and intervention points. Choose conversation, a document, a table, a static view, an interactive scene or another available medium according to what must become understandable. Space, time, dependencies and alternatives may benefit from interaction or motion; no fixed maturity ladder requires it. Devices, environment, noise and physical conditions can change reception without prescribing a technology. Multiperceptual forms remain possibilities to construct with actual means.
Use the actual visual-design and code-construction knowledge available to build the form. From a demonstration, learn hierarchy, rhythm, camera, transition, materiality, typography and perceptual continuity with their conditions. Separate reusable grammar from brand, content and assets. Build original code for the current object rather than treating an external demo as the capability. Reusable tools may help without becoming a compulsory stack.
Expose shared semantic state to the system and to the person. A screenshot alone cannot recover an unexposed selection, temporal lens or transient target. Distinguish a temporary pointer from a persistent selection. Bind values and events so a human change reaches the working state and a system change becomes perceivable. Preserve orientation through transitions; moving geometry should reveal a relationship or state change.
Choose among retaining, adapting or rebuilding a form through the relations it can actually express. A preferred projection cannot manufacture a missing capability. Search for, compose or construct that capability when needed. Observe the rendered outcome and reachable state at meaningful moments; separate a fixture from the evaluator deciding what it can express. Compare human intervention and programmatic change, including focus and accessibility. Keep original editable sources. A weak result can require a code correction, a different expressive grammar or a reformulated question. Return that cause to the competence that must build or communicate differently next time.
Make the meaning of current work and the person's real intervention points understandable. Adapt explanation, depth and form to what the person needs to orient, correct or continue the work.
Use this competence when an AI system returns results without making their implications or participation clear. Explanation comes from actual activity: what changes, why it matters, supporting sources and how the person can correct or continue. Its length follows that need.
Understand current work and what the person must recognize to participate. Make a relation visible when it can change the next choice: a different source, decision, forming capability, completed effect or open question. Do not add an introductory tutorial or component inventory merely because those components exist. Continue directly when the result is already understandable.
Express the change recognizably: what happened, what it means here and what the person can do. Keep a decision-changing uncertainty or condition beside the result it qualifies. Describe a saved but inactive change as such. When the person can correct a source, change direction or request depth, make that passage reachable through actual means.
A sentence may suffice; a complex relation may need an example, comparison or broader view. Use the person's vocabulary without inferring knowledge from role or silence alone. A technical term helps when it compresses a relation already understandable. Choose text, imagery or interaction for its concrete contribution while keeping deeper inquiry possible. Brevity and understanding are different measures.
Reformulate from the person's position while preserving source and object. If an example or question reveals a lost premise, correct the work itself rather than its tone alone. Repeated wording is not independent confirmation. Continue reformulation while it adds material understanding; stop when it changes no useful relation. The person's response can change question, method and form.
In interfaces, connect proposals to real state and preserve the person's work: entered text, selection, context and attention. Feedback must not erase a draft or simulate consent. Compose design and implementation competences to construct the surface when needed; the object and sources retain truth ownership. Explain observable operating reasons, sources and consequences without exposing or simulating private model reasoning.
Do not equate success with the quantity of explanation. Seek pertinent consequences: the person recognizes where to correct, distinguishes proposal from effect, asks a more precise question or resumes with less reconstruction. Confusion, an illusion of control or excessive interruption teach what to change. Preserve reusable relations without imposing one answer format or identical footer on every recipient.
A meta relation may become visible when the system explains how a competence forms or changes and the person can guide that formation. Explanation does not become a second engine determining domain truth.
An autological relation appears when this competence applies its own method to its own comprehensibility: a confusing explanation becomes material for improving explanation. This differs from autonomous execution and access to internal reasoning.
Daily work can gradually teach participation. Useful evidence lies in how the person understands and intervenes in later situations, not in the existence of a panel or description.
Begin from available context and sources. Exercise the method in chat; persistence, tools and external actions depend on actual means and mandate. Observe behavior and value in real work.
Read OPERATING_GUIDE.md as part of this method when forming the work. It carries the practical knowledge, decision relations and available means for this capability. The complete chat starter includes this knowledge and worked examples, so it remains usable without private sources. Preserve the public source identity and adapt through the recipient's actual means.
A meta relation can become visible when the system explains how a competence forms or changes and the person can guide that formation. Explanation does not become a second engine deciding domain truth.
This competence has an autological relation when it applies its method to the understandability of its own work: a confusing explanation becomes material for improving how it explains. This is distinct from agent autonomy and access to internal reasoning.
Competence, meta and autology: the conceptual guide ↗Everyday work can gradually teach participation in the system. Useful evidence lies in how a person understands and intervenes in later situations, not in the presence of a panel or description.
Attach or paste START_CHAT_EN.txt, which contains the complete method, and add your context. The AI can work in the conversation's language. Retain the result and useful guidance if you want to resume in another chat.
Make the competence folder readable and have the agent read SKILL.md for the relevant work. Connect real sources and tools according to the environment's instructions. The kit does not install accounts, memory or recurring execution.
With an existing kernel, connect the method to the competence or guide already responsible for the work. Without one, apply the instructions directly and preserve useful knowledge in the project or a reusable note.
Read or copy the complete English starter and add your context. English is the kit’s primary language; Italian counterparts are included.
