Main site

Product AI agent · Meeting notes

Meeting-Notes AI Agent

Notice before the bot joins, retention set where legal can see it, and every summary held for the named owner who signs the record of what the room decided.

Typical delivery
4–6 weeks
Deployment
Your stack
Named owner
Hold-aware
After launch
Agent Care

What this agent does

Writes the summary, never the decision

In

  1. 01

    A call is joined, and the notice is played into the audio before a word is captured.

  2. 02

    A notice is played, and RCW 9.73.030(3) puts that announcement inside the recording itself.

Reason

  1. 03

    A bot tile shows, and Illinois and Massachusetts turn on concealment, not on consent.

  2. 04

    A vendor retains audio to train on, and In re Otter.AI puts it outside the party exception.

  3. 05

    A timer runs on a summary under hold, and FRCP 37(e)(2) turns on intent to deprive.

Decide

  1. 06

    A summary reaches a regulated firm, and 17 CFR 240.17a-4(b)(4) covers inter-office memoranda.

  2. 07

    A works council is asked, and BetrVG s 87(1) Nr. 6 gives it a veto, not a notice period.

Out

  1. 08

    A room decided nothing, and that is recorded as no decision rather than resolved into one.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

Capture, notice and drafting belong to the agent. The record belongs to a named owner, and a wrong summary is amended on its face, never overwritten.

Example workflow

One meeting, notice to confirmation

  1. 1Invitation receivedCalendar invitations, conferencing hooks, dial-ins or a direct request to attend
  2. 2Notice played and capture openedThe spoken announcement inside the audio, the attendees, the territories and the meeting class
  3. 3Transcript and summary draftedThe summary, the utterances under it, the speakers named and confidence
  4. 4Controls appliedNotice checks, territory checks, retention and hold checks and drafting confidence

No human action required

Stages 1 to 4 run unaided, and nothing is filed at any of them — the agent is drafting, and the owner lane opens at the confirmation gate.

5DecisionSplits at the confirmation gate
Routine internal meeting

Goes to the named owner to confirm.

Anything privileged or regulated

Adds a counsel read first.

Owner review

The summary is held with its notice, its transcript and the process record FRE 901(b)(9) calls for.

Confirm · Amend · Send to counsel review

Confirmed — by the named owner

  1. 6Workspace and archive records updatedOnly where write access and records policy allow it
  2. 7Outcome evaluatedNotice coverage, attribution accuracy, owner amendments and what review found

Amendments

Any amendment the owner makes is logged in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Confirming a summary as the record of a meeting.
    • Joining a call covered by legal privilege.
    • Setting the retention term for a summary.
    • Deciding that a room reached a decision.

  • Automation boundary

    Agent acts unaided

    • ✓Play the notice into the audio before capture opens.
    • ✓Record the retention class and hold state that governs each summary.
    • ✓Mark what the room left open rather than resolving it into a decision.
    • ✓Hold the summary for the named owner who confirms it.

    Nothing becomes the record except by a named owner, inside the agreed boundaries.

    • Judging whether a meeting may be recorded at all.
    • Feeding a summary into a performance review.
    • Suspending a retention timer under a legal hold.
  • Changes to notice, retention or attribution rules.

Example output

One meeting summary, annotated

No court has held a notetaker is a party, and In re Otter.AI is a district-level plausibility ruling that Lisota v. Heartland Dental cuts against; below is one summary as the agent leaves it.

Meeting summary · single call

Illustrative example
Meeting
Pricing review, cross-functional
Recorded as
No decision recorded
Notice
Announced in audio
Evidence of record
Transcript, 3 August 2026
Confidence
Held unconfirmed
Held for
The named owner, by name

As received

Taken from the audio and the transcript under it, and it asserts no decision the room did not reach.

What the record holds

  • Audio and notice
  • Verbatim transcript
  • Attendance record

Why no decision here

Recording that a room decided is a call the named owner makes.

What the score decides

Below the configured threshold a summary picks up a counsel read before the owner sees it.

Action

ConfirmAmendSend to counsel review

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every meeting

From the notice that opened it

03Evidence

Where the summary is used

An escalation-summarisation agent compresses a thread for a stated reader; this one manufactures the record of what a room decided, and Kovel, 296 F.2d 918, does not reach it.

01Approved path

The record outlives the room

Retain nothing and a hold has nothing left to hold; retain everything and the cheapest record in the building becomes the easiest one produced against you.

02Human review

What was checked, and not found

No rule was found requiring an organisation to take meeting notes, requiring a summary to be accurate, or giving a participant standing to contest one that misreads them, and no decided case was found authenticating an AI summary or holding that a notetaker waived privilege.

04

Build an evidence trail

The summary, the audio it came from and the notice that was played stay together.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Conferencing platforms

    • Zoom
    • Microsoft Teams
    • Google Meet
    • Webex
    • conferencing APIs

  • Calendar systems

    • Google Calendar
    • Outlook
    • Invitations and attendee lists

  • Workspaces and archives

    • Slack
    • Confluence
    • Notion
    • Where the summary is filed and read again

Agent

Meeting capture and summarisation

  • Plays the notice
  • Drafts the summary
  • Holds for the owner

  • Records and hold systems

    • eDiscovery
    • records and archives
    • Legal holds and retention classes

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six shutters between the model and the record

Six shutters over one window, the last the tightest. What is kept is drawn in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeDecline the join and stop capture when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and capture rules, and note the version each summary was drafted under.Track
L4TraceabilityRecord each summary, the audio under it, the notice played and the prompts that produced it.Record
L3Owner confirmationHold the summary for a named owner; the hold governs filing, not whether the summary is right.Gate
L2Capture guardrailsTest each meeting against configured territory and privilege rules; a failure blocks the join, not the meeting.Restrict
L1Confidence thresholdsRoute a privileged or regulated meeting to a counsel read before any summary is filed.Require review
Model coreSummary drafted — the notice, the transcript, the speakers named and confidence
L1 – L2
Test whether a summary may stand
L3
Leaves the confirmation to a named owner
L4 – L5
Keep the summary and the notice behind it
L6
Declines to join the call when signals degrade

How Nestack evaluates it

Evaluate the whole capture run — not only the summary that comes out.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the summary a team remembers

  1. E1Final-output evaluationDid the summary alter the semantics of what was actually said?
  2. E2Step-level evaluationDid the agent play the notice, read the right call and the live rules?
  3. E3Tool evaluationDid it read and write the correct meeting and the correct record?
  4. E4Confidence calibrationDo low-confidence summaries actually attract more owner amendments?
  5. E5Slice evaluationHow does performance change across specific meeting classes?
  6. E6Business outcomeHow many summaries needed an amendment before the owner confirmed?

Floor — the audio a summary rests on

Failure modes

Where each failure originates in the agent

Seven failure modes, each set where the meeting first shows it.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • QK-03

      Stale invitation read

      The invitation read is not the one now live.

    Stage gathers

    The invites, the audio, the speakers and the rules

  2. 02 · Reasoning

    2 modes

    • QK-04

      Decision asserted, not said

      A resolution appears that nobody stated.

    • QK-06

      Utterance bound to wrong speaker

      A line is attributed to another attendee.

    Stage proposes

    The summary, the transcript and the notice

  3. 03 · Tool / write

    2 modes

    • QK-02

      Thin summary passed forward

      A summary moves on without the counsel read.

    • QK-05

      Summary filed to wrong meeting

      The record is filed against another call.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • QK-01

      Confirmed, notice unrecorded

      The record shows a summary but not the notice.

    Stage returns

    The summary a team reads and a court receives

  5. 05 · Change / Version

    1 mode

    • QK-07

      Silent retention drift

      A timer shortens while a hold still stands over the file.

    Stage tracks

    Model, prompt, capture rules and retention terms

  • Sev-1 · a decision recorded that nobody made
  • Sev-2 · a held summary deleted on a timer
  • Sev-3 · notice fails, the join is declined
See the complete Product failure-mode catalogue

Affected slices

Privileged calls absorb the amendments

A class-level capture figure can read clean while privileged and counsel calls carry most of the owner amendments. Nestack reports the amendment rate by meeting class, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Privileged and counsel calls8.2%3.7×Review
Regulated-firm internal meetings5.8%2.6×Review
Works-council territories3.6%1.6×Watch
Routine internal stand-ups1.6%0.7×Normal

Bar: amendment-rate lift vs. routine-meeting baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

What a recorded decision costs

A loop shuts when the decision recorded that nobody made is a standing case. That suite is what the next summary written is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Amendment rate rises on privileged and counsel calls.

  2. 02

    Diagnose

    The decision recorded in a room where nothing was decided, produced in discovery nine months on, is worked backwards until one cause is left standing.

  3. 03

    Improve

    The change goes out numbered, and the summaries that forced it ride beneath.

  4. 04

    Verify

    Nothing ships while one touched summary case is still red.

  5. 05

    Learn

    The case is held, and the capture rules are rewritten alongside it.

  6. Learn → Detect

    The return edge. The next summary is measured against a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, summary drafting, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Meeting-class discovery and automation-boundary work.Week 1
  2. 02Conferencing, calendar and archive sources.Week 1, Week 2
  3. 03Summary-to-evidence and retention-coverage mapping.Week 1, Week 2
  4. 04Meeting audio intake.Week 2
  5. 05Notice, speaker and summary binding.Week 2, Week 3
  6. 06Confidence scoring and review routing.Week 3
  7. 07Owner confirmation workflow.Week 3, Week 4
  8. 08Workspace and archive integration.Week 5
  9. 09Notice and summary cases.Week 4, Week 5
  10. 10Guardrails and retention controls.Week 4
  11. 11Capture-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne workspace, one quarterProductionProduction capture workflowAdvancedMultiple regions / meeting classes
Introduced at Pilot
Summary drafting to your capture rules✓✓✓
Named owner confirmation✓✓✓
Meeting-population baseline✓✓✓
Introduced at Production
Reporting by meeting class—✓✓
Counsel review workflow in your systems—✓✓
Approved write-back—✓✓
Calendar-and-archive integration—✓✓
Introduced at Advanced
Multi-territory notice rules——✓
Cross-team summary packs——✓
Large meeting estates——✓
Multi-jurisdiction capture controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, meeting volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your meeting classes and the notice each one carries

    Notice capture and retention-class versioning

    Week 1

  2. 02Representative recordings, transcripts and summaries

    Notice binding, drafting logic and the summary baseline

    Week 2

  3. 03Your records policy and the holds it names

    Notice mapping, retention binding and the automation boundary

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Conferencing, calendar and archive assessment, then integration setup

    Week 2

  5. 05Summaries you would not want produced

    Notice cases and failure-mode testing

    Week 4

  6. 06What no summary may decide

    Confidence scoring, review routing, guardrails and release controls

    Week 3

  7. 07A named owner who confirms the summary

    Release to the named owner, and then pilot and production runs

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

The widths are drawn from the work itself, so the fifth week carries a pair rather than a blank.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Meeting-class discovery, notice mapping and the automation boundary

W2Capture integration and the meeting-population baseline

W3Summary drafting, retention logic and release controls

W4Evaluation suite, notice cases and failure-mode testing

W5Archive integration, pilot summaries and targeted corrections

W6One meeting quarter run under the operations owner, then Agent Care handover

Reading the band
Each bar covers only the weeks its own work is named for, and week five holds two by design.
At the end of W6
When the capture record validates, Agent Care takes the agent on.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Product AI agent

Build a meeting-notes agent around the retention setting nobody in legal has seen.

Nowak, C-434/16, reads a summary as accurate in light of its purpose, so the participant it misreads has nothing to correct. Show us your retention setting, the meetings legal does not know happened, and who confirms the record.

Typical delivery
4–6 weeks
Deployment
Your meeting stack
Named owner
Hold-aware
After launch
Agent Care available

Nestack Agents · Meeting notes

AGT-PRD-07 · Agent Care available after launch

Build a meeting-notes agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your meetings

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.