HR AI agent · Interview scheduling
Interview Scheduling AI Agent
Draft the invitation, hold the thread beside the slot it moved, and stop the booking the moment a candidate asks for something a coordinator, not a calendar, has to answer.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named human
- Message first
- After launch
- Agent Care
What this agent does
Books the slot, never answers the request
In
- 01
An invitation goes out, and the reply that came back is kept beside the slot it moved.
- 02
A reply reads as a preference, and it is held as a possible request until a human says otherwise.
Reason
- 03
A candidate names a barrier in plain words, and the booking stops rather than quietly moving.
- 04
A follow-up lands on an open request, and it is answered as that request, not as a status query.
- 05
A slot set is offered, and the wider set it was drawn from is recorded next to it.
Decide
- 06
A thread turns toward pay history or convictions, and the agent stops and hands the thread over.
- 07
A recorder is proposed for a panel, and the thread waits on consent from each party named in it.
Out
- 08
A thread goes quiet with a request open, and that silence is raised rather than closed as attrition.
- 09
Take write actions in a calendar only within the approval boundaries agreed at implementation.
Product statement
Drafting, holding and recording belong to the agent. Answering a request belongs to a named coordinator, who carries it into the interactive process and owns it there.
Example workflow
One thread, invitation to confirmation
- 1Requisition and panel receivedPanel members, room or link, candidate contact and the requisition it belongs to
- 2Invitation drafted and stampedThe slots offered, the wider set they were drawn from and the day the invitation was sent
- 3Reply read and classedThe message exactly as written, the class assigned to it and the confidence behind that class
- 4Controls appliedRequest screens, prohibited-topic checks, consent checks and classification confidence
No human action required
Stages 1 to 4 run unaided, and no interview is confirmed at any of them — the agent is drafting, and the coordinator lane opens at the reply gate.
Goes to the named coordinator to confirm.
Adds a coordinator read before a slot moves.
Coordinator review
The thread is held with the message as written, the slots offered and the day it arrived.
Confirm · Append note · Send to coordinator review
Confirmed — by the named coordinator
- 6Calendar and applicant records updatedOnly where write access and retention policy allow it
- 7Outcome evaluatedClassification accuracy, request handling, coordinator corrections and what review found
Corrections
Every coordinator correction is counted back into the evaluation suite.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Answering a request for reasonable accommodation.
- Deciding whether an accommodation is reasonable.
- Judging what counts as an undue hardship.
- Confirming an interview with a candidate.
Automation boundary
Agent acts unaided
- ✓Draft the invitation and stamp the day it went.
- ✓Record the message as written and not the classification given to it.
- ✓Escalate a possible accommodation request to the named coordinator.
- ✓Hold the booking open while a request is unresolved.
No interview is confirmed except by a named coordinator, inside the agreed boundaries.
- Ruling that a message is only a scheduling preference.
- Telling a candidate what the process can accommodate.
- Approving the slot set a candidate is offered.
- Changes to the calendar, the record or the routing rules.
Example output
One scheduling thread, annotated
This serves a recruiting team who may have to account for one candidate message to an agency many months after it arrived; below is one thread exactly as the agent leaves it.
Scheduling thread · single candidate
Illustrative example- Message
- Panel interview, stocking role
- Classed as
- Read as a possible request
- Requisition
- Message kept verbatim
- Evidence of record
- Candidate reply, 3 August 2026
- Confidence
- Held unconfirmed
- Held for
- The named coordinator, by name
As received
Built from the candidate reply and the invitation it answered, and it claims nothing beyond them.
What the thread holds
- Candidate reply
- Invitation sent
- Slots offered
Why no confirmation here
Answering an accommodation request is a judgement for a coordinator.
What the score decides
Below the configured threshold a thread gets a coordinator read before a slot moves.
Action
ConfirmAppend noteSend to coordinator 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 thread
From the message that carries it
Where the message is used
The agent does not decide what a candidate meant, only what the candidate wrote, when it arrived and which slot moved after it — the record a named human may have to stand behind months later.
It arrived in the wrong box
A candidate may ask in plain English, and the EEOC guidance of 17 October 2002 says the request need not mention the ADA or use the phrase reasonable accommodation.
What was checked, and not found
An eCFR full-text search on 27 August 2026 returns nothing for interview scheduling, for scheduling of interviews or calendar invitation, while reasonable accommodation, job applicant and reschedule return hits, and what answers schedule an interview is food-stamp and visa rules, not employment.
Build an evidence trail
The message, the slot it concerned and the coordinator who read it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Applicant tracking system
- Greenhouse
- Lever
- Workday
- Requisition and candidate records
Calendar and conferencing
- Outlook
- Zoom
- Panel availability and room holds
Candidate messaging
- SMS
- the candidate portal
- Replies, follow-ups and free text
Agent
Interview scheduling and coordination
- Reads the thread
- Drafts the invitation
- Holds for the coordinator
Accommodation and access
- Interpreter rosters
- access notes
- Arrangements already on the file
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six sorts between the model and the coordinator
Six sorts down one chute, the last the tightest. What lands is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a thread may move
- L3
- Leaves the answer to a named coordinator
- L4 – L5
- Keep the slot and the message behind it
- L6
- Routes the thread to a coordinator when signals degrade
How Nestack evaluates it
Evaluate the whole thread — not only the invitation that goes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the invitation a candidate opens
- E1Final-output evaluationDid the invitation carry the accommodation question it is permitted to ask?
- E2Step-level evaluationDid the agent read the reply, the thread behind it and the live requisition?
- E3Tool evaluationDid it move the correct event on the correct calendar?
- E4Confidence calibrationDo low-confidence classes actually attract more coordinator corrections?
- E5Slice evaluationHow does performance change across specific reply types?
- E6Business outcomeHow many threads needed a correction before the coordinator confirmed?
Floor — the message a thread rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, set at the stage where each first becomes visible.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- UC-07
Booking page unusable
The handoff page cannot be driven, and silence follows.
Stage gathers
The panel, the rooms, the thread and the reply
02 · Reasoning
2 modes
- UC-01
Request read as a reschedule
A legal request is filed as a change of time.
- UC-04
Slot set that quietly selects
Neutral filters narrow who can attend at all.
Stage proposes
The message, the class and the slots offered
03 · Tool / write
2 modes
- UC-03
Prohibited topic raised
The thread reaches for pay or conviction history.
- UC-05
Recorder without all-party consent
A panel is captured before each party agreed.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- UC-02
Acknowledged, then dropped
The escalation has no owner and the thread ends.
Stage returns
The invitation a candidate opens and answers
05 · Change / Version
1 mode
- UC-06
Retention on one default
One setting deletes what another rule requires kept.
Stage tracks
Model, prompt, routing rules and hold periods
- Sev-1 · a request answered as a reschedule
- Sev-2 · an escalation nobody owns
- Sev-3 · a thread held back for a read
Affected slices
Free-text replies absorb the corrections
A thread-level accuracy figure can read clean while free-text accommodation replies carry most of the correction. Nestack reports the correction rate by reply type, and not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Free-text accommodation replies | 11.5% | 3.7× | Review | |
| Multi-state panel threads | 8.1% | 2.6× | Review | |
| Reschedule-and-follow-up threads | 5.1% | 1.6× | Watch | |
| Plain confirmation replies | 2.3% | 0.7× | Normal |
Bar: correction-rate lift vs. plain-confirmation baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a dropped follow-up costs
A cycle ends when the request answered as a reschedule is a standing case. That suite is what the next panel booked is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on free-text accommodation replies.
- 02
Diagnose
The interview that never happened, because somebody said she would look into the interpreter and then did not, is worked backwards until one cause is left standing.
- 03
Improve
Invitations go out numbered, and the threads behind them travel attached.
- 04
Verify
A single unread request case stops the whole panel confirming.
- 05
Learn
The case stays on, and the routing rules are redrawn beside it.
Learn → Detect
The return edge. The next panel booked 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, thread handling, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Automation-boundary work and escalation routing rules.Week 1
- 02Calendar, tracker and messaging sources.Week 1, Week 2
- 03Reply-classification and thread-coverage mapping work.Week 1, Week 2
- 04Candidate thread ingestion.Week 2
- 05Thread, slot and requisition binding.Week 2, Week 3
- 06Request flagging and review routing.Week 3
- 07Coordinator confirmation workflow.Week 3, Week 4
- 08Applicant-tracking integration.Week 5
- 09Request and follow-up cases.Week 4, Week 5
- 10Guardrails and invitation controls.Week 4
- 11Thread-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 requisition, one cycle | ProductionProduction recruiting workflow | AdvancedMultiple requisitions / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Invitation drafting to your templates | ✓ | ✓ | ✓ |
| Named coordinator confirmation | ✓ | ✓ | ✓ |
| Panel-availability baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by request class | — | ✓ | ✓ |
| Coordinator review workflow in your systems | — | ✓ | ✓ |
| Approved calendar write-back | — | ✓ | ✓ |
| Calendar-and-tracker integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-jurisdiction rule sets | — | — | ✓ |
| Cross-panel interview packs | — | — | ✓ |
| Large candidate volumes | — | — | ✓ |
| Multi-panel coordination controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, thread 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 live requisitions and the panel behind each one
→Thread capture and message retention
Week 102Representative candidate threads and invitations
→Source binding, classification logic and the thread baseline
Week 203Your interview calendar and the coordinators it names
→Escalation mapping, thread binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Calendar, tracker and messaging assessment, then integration setup
Week 205Threads you would not want read back
→Follow-up cases and the evaluation round
Week 406What no invitation may assume
→Request flagging, review routing, guardrails and release controls
Week 307A named coordinator who confirms the interview
→Release to the named coordinator, 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
Widths here track the work and not the grid, which is why one band sits under another in week five.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Escalation discovery, thread capture and the automation boundary
W2Source integration and the panel-availability baseline
W3Reply classification, routing logic and release controls
W4Evaluation suite, request cases and failure-mode testing
W5Applicant-tracking integration, pilot threads and targeted corrections
W6One interview cycle run under the recruiting lead, then Agent Care handover
- Reading the band
- No bar runs past the weeks its own work is named for, and week five is shared on purpose.
- At the end of W6
- When the thread record validates, Agent Care assumes the agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · HR AI agent
Build an interview scheduling agent around the sentence your last thread never kept.
Show us one requisition you schedule each month and the replies the last panel produced. If a candidate asked for something in plain words and the only trace is a moved calendar entry, the record an agency would ask for may not exist. A dropped follow-up comes back as a finding.
- Typical delivery
- 4–6 weeks
- Deployment
- Your requisitions
- Named human
- Message first
- After launch
- Agent Care available
Nestack Agents · Interview scheduling
AGT-HR-06 · Agent Care available after launch