Communications AI agent · Executive drafting
Executive Communications Drafting AI Agent
A speechwriter is not the maker of the speech, and the person who delivers it takes the blame, so the agent drafts the post and proves the principal adopted it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named principal
- Byline first
- After launch
- Agent Care
What this agent does
Drafts the byline, never carries it
In
- 01
A post is written for a named individual, and the words become theirs the moment the name goes on.
- 02
A speechwriter drafts, and under Janus the speaker keeps ultimate authority over the content.
Reason
- 03
The drafter never makes the statement, and Lorenzo still reaches whoever pushed it out.
- 04
A machine composed the text, so Art. 50(4) asks who holds editorial responsibility for it.
- 05
One clause carries non-public commercial substance, and a personal account is no official source.
Decide
- 06
A product the company sells is praised, and 16 CFR 255.5 asks after the connection behind it.
- 07
An op-ed argues a legislative outcome, and the deduction goes even though registration does not.
Out
- 08
Nothing on offer can show whether a machine wrote it, and that absence is recorded as the finding.
- 09
Execute write actions only where the approval boundaries set at implementation allow it.
Product statement
Drafting, recording and holding belong to the agent. Adoption belongs to a named principal, who puts their own name over the sentences and answers for them personally.
Example workflow
One post, brief to byline
- 1Brief and corpus receivedThe brief, the pieces already published under that name, the sources and any policy ask it carries
- 2Principal and channel fixedWhich named individual the words are attributed to, and which channel of theirs will carry them
- 3Draft raised under that nameThe text, the claims inside it, the third parties it names and the disclosure it needs
- 4Screens appliedMateriality screens, connection checks, named-party sourcing and drafting confidence
No human action required
Stages 1 to 4 run unaided, and no name goes over anything at any of them — the agent is drafting, and the principal lane opens at the adoption gate.
Goes to the named principal to adopt.
Takes a disclosure-counsel read first.
Principal review
The draft is held with what it was built from, what it asserts and who it names.
Adopt · Append source · Send for legal review
Adopted — by the named principal
- 6Channel and record systems updatedOnly where the write access and records policy reach that channel
- 7Publication evaluatedAdoption evidence, edit depth, principal corrections and what counsel review found
Principal corrections
Each edit the principal makes is scored in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Putting a name over a sentence.
- Deciding a personal post carries nothing material.
- Judging whether a claim is safe to attribute.
- Holding editorial responsibility for a publication.
Automation boundary
Agent acts unaided
- ✓Draft the piece and mark the claims inside it.
- ✓Flag each named third party and the source under it.
- ✓Hold the exact text the principal was shown, hashed and dated.
- ✓Record what the principal changed and what they declined to sign.
No name goes over a sentence except by the named principal, inside the agreed boundaries.
- Deciding a characterisation of a person is fair.
- Telling a board the voice is genuinely theirs.
- Choosing what a founder argues for in public.
- Changes to the byline, the text or the record.
Example output
One executive post, annotated
This serves a communications team writing under a named individual whose own name carries the exposure; below is one post exactly as the agent leaves it.
Executive post · personal channel
Illustrative example- Post
- Personal post, one named principal
- Recorded as
- Drafted against the brief
- Channel
- Personal channel, not official
- Evidence of record
- Brief and sources, 7 August 2026
- Confidence
- Held unsigned
- Held for
- The named principal, in person
As received
Built from the brief and the source material named under it, and it asserts nothing the principal has not read.
What the adoption record holds
- Brief on file
- Named-party sources
- Prior signed pieces
Why no byline here
Putting a name over a sentence is a thing only that person does.
What the score decides
Below the configured threshold a draft gets a counsel read before the principal sees it.
Action
AdoptAppend sourceSend for legal reviewValue
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
Each post drafted
Under the name that carries it
Where that evidence lands
The agent does not vouch for a voice, only for what the principal was shown, when they were shown it, and what they changed before the name went on.
Their name, our sentences
Janus makes the signer the maker, and Art. 50(4) lets that same signature stand in place of a label, so one act creates the liability and discharges the duty at once.
Looked for, and not found
No published measure was located for reproducing a named individual's voice, and no accepted protocol for testing one, while the detector literature runs the other way — the vendor of the best-known classifier withdrew it for low accuracy — so the claim is left unmade here.
Build an evidence trail
The draft, the review it received and the individual whose name it carries stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Executive channels
- X
- personal blogs
- Pieces published under that name
Brief and source material
- Comms briefs
- fact packs
- What a piece is allowed to say
Approval and identity systems
- SSO
- signing keys
- approval logs
- Who adopted which text, and when
Agent
Executive comms drafting
- Reads the corpus
- Drafts under the name
- Holds for the principal
Disclosure and legal records
- Insider lists
- disclosure counsel
- Material that is not public yet
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six holds between the model and the byline
Six passes over one page, the last the closest. What survives is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a draft may go under a name
- L3
- Leaves the byline to a named principal
- L4 – L5
- Keep the post and the review behind it
- L6
- Returns the draft unsigned when signals degrade
How Nestack evaluates it
Evaluate the run that produced the words — not only the words that come out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the post a byline carries
- E1Final-output evaluationDid the record hold what the principal was actually shown?
- E2Step-level evaluationDid the agent read the right brief, the right sources and the live channel rules?
- E3Tool evaluationDid it read and write the correct principal and the correct channel?
- E4Confidence calibrationDo low-confidence drafts actually attract deeper principal edits?
- E5Slice evaluationHow does performance change across specific piece types?
- E6Business outcomeHow many drafts needed an edit before the principal adopted?
Floor — the corpus a voice rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed at the stage it first turns visible.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- UO-01
Policy ask arrives unmarked
A brief seeks a legislative outcome and nothing flags it.
Stage gathers
The brief, the corpus, the sources and the name
02 · Reasoning
2 modes
- UO-02
Material clause in a personal post
A private channel carries company substance.
- UO-03
Connection left undisclosed
A product the company sells is praised as a bystander would.
Stage proposes
The text, the claims, the parties named and confidence
03 · Tool / write
2 modes
- UO-04
Qualifier dropped for register
A true sentence goes out missing what qualifies it.
- UO-05
Unsourced line about a person
A named third party is characterised from nothing.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- UO-06
Signed but not read
The record shows a signature and not a reading.
Stage returns
The piece that carries a name, and who reads it
05 · Change / Version
1 mode
- UO-07
Corpus learns its own output
The voice model trains on text the principal never wrote.
Stage tracks
Model, prompt, voice rules and byline dates
- Sev-1 · a byline signed on no record
- Sev-2 · a personal post carries substance
- Sev-3 · signals degrade, byline withheld
Affected slices
Policy and advocacy pieces absorb the corrections
A byline-level correction figure can read clean while policy and advocacy pieces carry most of the rework. Nestack reports the correction rate by piece type, not by the principal alone.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Policy and advocacy pieces | 6.6% | 3.7× | Review | |
| Personal-channel posts | 4.7% | 2.6× | Review | |
| Pieces naming third parties | 2.9% | 1.6× | Watch | |
| Standing internal notes | 1.4% | 0.8× | Normal |
Bar: correction-rate lift vs. internal-note baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an unread signature costs
A cycle shuts when the sentence the named individual never read is a standing case. That suite is what the next byline taken is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on policy and advocacy pieces.
- 02
Diagnose
The post adopted in seconds on a phone, at a length nobody could have read in the time, is worked backwards until one cause is left standing.
- 03
Improve
Posts ship numbered, and the reviews behind them travel attached.
- 04
Verify
Each touched review case is run again, and one red holds it back.
- 05
Learn
The case is kept, and the byline rules are amended in that same commit.
Learn → Detect
The return edge. The next byline is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope set against the delivery timeline. Week structure follows the six-week plan — discovery, corpus, drafting logic, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Editorial-responsibility and automation-boundary work.Week 1
- 02Corpus, brief and channel sources.Week 1, Week 2
- 03Draft-to-principal and byline-clearance trail mapping.Week 1, Week 2
- 04Corpus ingestion.Week 2
- 05Principal, channel and claim binding.Week 2, Week 3
- 06Materiality screening and routing.Week 3
- 07Principal adoption workflow.Week 3, Week 4
- 08Channel and record integration.Week 5
- 09Review and attribution cases.Week 4, Week 5
- 10Guardrails and attribution controls.Week 4
- 11Byline-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne principal, one cycle | ProductionProduction publishing workflow | AdvancedMultiple principals / channels |
|---|---|---|---|
| Introduced at Pilot | |||
| Drafting to your byline rules | ✓ | ✓ | ✓ |
| Named principal adoption | ✓ | ✓ | ✓ |
| Voice-and-record baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by channel class | — | ✓ | ✓ |
| Legal review workflow in your systems | — | ✓ | ✓ |
| Approved channel write-back | — | ✓ | ✓ |
| Drafting-and-channel integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-principal voice work | — | — | ✓ |
| Cross-channel piece packs | — | — | ✓ |
| Dense publishing calendars | — | — | ✓ |
| Multi-principal review controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, publishing 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.
You bring
We build with it
01Your named principals and the channels each one publishes on
→Voice capture and byline versioning
Week 102Representative briefs, past pieces and source material
→Corpus binding, drafting logic and the voice baseline
Week 203Your publishing calendar and the principals it names
→Byline mapping, corpus binding and the automation boundary
Week 104Access to the relevant channels, feeds or exports
→Corpus, channel and approval-system assessment, then integration setup
Week 205Sentences you would not want attributed
→Attribution cases and the evaluation round
Week 406What no byline may cover
→Materiality screening, review routing, guardrails and adoption controls
Week 307A named principal who adopts the words
→Release to the named principal, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
A band is exactly as wide as its phase costs here, so week five carries two where one would look tidier.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Byline discovery, voice versioning and the automation boundary
W2Corpus and channel integration, and the voice-and-record baseline
W3Drafting logic, materiality screening and adoption controls
W4Evaluation suite, attribution cases and failure-mode testing
W5Channel and record integration, pilot drafts and targeted corrections
W6One publishing cycle run under the named principal, then Agent Care handover
- Reading the band
- Each band runs only across the weeks its own work is named for, and week five is shared by design.
- At the end of W6
- Once the review record validates, Agent Care adopts the agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Executive comms agent
Build an executive drafting agent around the signature your last post never actually recorded.
Show us one post that went out under a leader's name last quarter and the record of what they read before it did. If the file holds a click and a timestamp and no more, the byline is doing work the record cannot support. A signature nobody can evidence comes back as a finding.
- Typical delivery
- 4–6 weeks
- Deployment
- Your named principals
- Named principal
- Byline first
- After launch
- Agent Care available
Nestack Agents · Executive drafting
AGT-COM-12 · Agent Care available after launch