Three-Dimension Virtual Collaborator Model
Evaluate AI through understanding, communication, and capability
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 98%
The model evaluates an AI collaborator across three connected dimensions. Understanding covers what the system knows about the user, surrounding world, organization, and immediate situation. Communication covers whether it recognizes the user's preferences and presents information in a form that is useful to them. Capability covers what the system can actually produce or do once it understands the request. A strong model alone is therefore insufficient: missing organizational context, unsuitable communication, or an inability to generate the required work product can each break the collaboration. Users can apply the model as a diagnostic scorecard, identify the weakest dimension, and then improve it through better context, clearer preferences, integrations, output formats, or executable capabilities.
Origin
Extracted from Marketing Against the Grain, where Scott White divided the concept of a virtual collaborator into three segments.
Core principles
- 01Useful collaboration begins with relevant understanding
- 02Communication should reflect the user's preferences
- 03Capability must translate context into completed work
- 04Weakness in any dimension limits the overall collaboration
How to run it
- 1
Map required understanding
Identify the personal, organizational, task, and situational information the AI needs to perform well.
Pro tip Separate globally useful context from knowledge needed only for one task.
Watch out Do not assume general model intelligence includes private or current organizational knowledge.
- 2
Define communication preferences
Specify the tone, level of detail, structure, and output format that make the AI's responses useful to the recipient.
Pro tip Provide an example of an ideal response or work product.
Watch out Vague preferences produce inconsistent communication.
- 3
Define required capabilities
List the concrete outputs or actions the collaborator must complete, such as writing a document, analyzing data, or producing an artifact.
Pro tip Describe success as an observable deliverable rather than a broad aspiration.
Watch out Do not confuse an informative answer with completion of the underlying task.
- 4
Run a representative test
Give the AI a realistic task and evaluate understanding, communication, and capability separately.
Pro tip Use work that is important enough to expose genuine limitations but safe enough to review.
Watch out A trivial demo may conceal weaknesses that appear in real workflows.
- 5
Strengthen the weakest dimension
Add context, refine communication instructions, or connect the necessary tools and systems. Retest until all three dimensions meet the workflow's threshold.
Pro tip Change one dimension at a time so the source of improvement is visible.
Watch out Improving output polish will not fix missing knowledge or missing capabilities.
In the wild
A leader asks Claude to prepare a strategy document. The prose is polished, but the recommendations ignore current company priorities. Using the model reveals that communication is strong while understanding is weak, so the leader adds internal plans and relevant historical material before retrying.
→ The revised document reflects both the desired format and the organization's actual situation.
A team scores a proposed assistant on access to company context, compatibility with employee communication preferences, and ability to create usable work products. It rejects a workflow that answers questions well but cannot produce or update the deliverable employees need.
→ The team chooses improvements based on operational usefulness rather than benchmarks alone.
Common mistakes
Measuring intelligence alone
A smart model can still be ineffective when it lacks the relevant context, communicates poorly, or cannot complete the required work.
Treating answers as outcomes
An assistant may explain a task without creating the document, analysis, or other work product the user actually needs.
Ignoring recipient preferences
Correct information delivered in an unsuitable form can still create friction and reduce adoption.
Is it for you?
Best for
Teams and individuals evaluating AI assistants or designing context-rich professional workflows.
Not ideal for
Comparisons based solely on benchmark scores without a concrete user, task, or working environment.
From the transcript
“if I break down this concept of a virtual collaborator it kind of goes into three segments”
“I think one is what Claude understands about you about the world about the situation around what you're trying to do how Claude communicates with…”
“the third is what is it actually capable of when I want to do something can it write a document for me can it go…”
From the episode
The Ultimate Guide to Using Claude AI for Work