Operations · CAP-007
Turn meeting notes into follow-up actions
Under our current test conditions, AI completed 7 of 8 scenarios correctly.
Useful, but the failure rate or edge cases are material.
- Success
- 7/8
- Last tested
- September 2026
- Evidence
- Simulated environment
- Supervision
- Regular supervision
What AI can currently do
- Explicit action line
- Owner will do something by a date
- Do not invent an owner
- A decision is not an action
- Two explicit actions
- Todo prefix
- Empty notes
These are scenarios the latest accepted run completed. They are not a claim about every business.
Keep a human involved when
- the follow-up is implied in prose rather than written as an action
- an owner or date would have to be guessed
- a decision is being treated as a task
What we tested
Each scenario starts from the Acme Services fixture, version acme-v1. The agent gets only the tools for that task. Software then checks what actually changed. A confident message is not a pass.
- Pass
Explicit action line
MEET-001
- Pass
Owner will do something by a date
MEET-002
- Fail
Implicit follow-up in prose
MEET-003
Actions miss “circle back”.
- Pass
Do not invent an owner
MEET-004
- Pass
A decision is not an action
MEET-005
- Pass
Two explicit actions
MEET-006
- Pass
Todo prefix
MEET-007
- Pass
Empty notes
MEET-008
Results
Cost is $0 because this accepted run used the local reference agent, not a paid model API. That is a measured cost for this configuration, not an estimate of a frontier model.
Common failure modes
- Implicit follow-up in prose. The phrase circle back was not extracted.
Evidence strength
Evidence: Simulated environment. This was tested in our controlled mini-business, not in a live customer system. A simulation result does not prove the task is reliable in every company.
History
One accepted run is on record. A line chart appears only after a later accepted run gives a real comparison.
Current tested configuration
Configuration A
Model: reference-agent-v1. Tools: simulated business systems for this task. Success: 7/8. Cost: $0.
We do not publish provider rankings from a single configuration.
Implementation blueprint
- 1Notes
- 2Extract explicit actions
- 3Structured follow-up list
- 4Human review of implied items