The ongoing negotiation of machine-produced output, transforming raw agent generation into trusted content through evaluation, refinement, and appropriation over time.
Actions
Rate, copy, regenerate, expand, more…
Variations
…
Follow up
…
Communicating uncertainty
- Intentionally uncertain language
- Explainable factors for predictions
- Confidence ratings
- Categorising accuracy
- Few-shot prompting with inconsistent results
do you know if da Vinci completed that sculpture?
I’m not sure. To the best of my knowledge, he didn’t—but records from that period are patchy.
2×2=?
2×2=5. However, this information is uncertain.
how many people in the world
what will be the price of milk in Stockholm in 10 years?
So, a plausible range is
Transparency
…
Input-output mapping
Visualises the influence of input on output.
Summarise this text:
“Leonardo da Vinci was an Italian polymath of the Renaissance whose areas of interest
included invention, painting, sculpting, architecture, science, music, mathematics,
engineering, literature, anatomy, geology, astronomy, botany, writing, history, and
cartography. He is widely considered one of the most diversely talented individuals
ever to have lived.”
Leonardo da Vinci was a Renaissance polymath known for his wide-ranging talents in science and the arts.
Section regeneration
Engages users in iterative dialogue (clarifying questions, multiple “regenerate” buttons) to refine outputs collaboratively.
…
Showing the work
…
Footprints
…
Caveats
…
Consequences
- the output is a draft the actor negotiates rather than an artefact they receive: rate it, edit it, regenerate part of it, take the half that works
- the actor owns the result without having made it, so what they can find out about where it came from decides whether they can stand behind it
- verification cost rises with length, and long output is exactly where accepting quietly takes the place of checking
- the output is a transition point the system can read — knowing what just arrived, it can propose what to do with it next
Resources & references
Appears in
Delegating work
- Receiving the result, step 1 — Present the result for judgement
Related patterns
Precedes
- Change review — the proposed change is generated output aimed at existing content
- Citation — the generated output is the material a citation attaches pointers to
- Extraction — generated output in a reply stream is the AI route into this move
- Instructed revision — acts on output the generation activity produced
- Regeneration — acts on the output the generation activity produced
- Next-best action — the output is a transition point the system can read — knowing what just arrived, it can propose what to do with it next
Enacts
- Agency — the actor steers, edits, and improves the output rather than receiving a finished artefact
Complements
- AI completion — block generation is closer to this pattern than to inline completion
- Annotation — system marks confidence and origin.
- Cognitive forcing functions — long-form output especially prone to over-reliance because verification cost is high
- Transparent reasoning — shows the step-by-step process behind the generated content.
- Explanation — provides contextual information to clarify AI-driven decisions within the output.
- AI tuning — settings that influence how content is generated and presented.
- Inline interface — produced output the actor negotiates; interlocks with, but distinct from, this move.
Preceded by
- Agent — generates the output being presented.
- Prompt — the input that leads to generated content.
- Live presentation — once the stream completes, a produced artefact sits on the surface — the move shifts from holding the rendering stable to negotiating what arrived
Hosted by
- Conversation — the reply stream is where generated output first appears