AgentNava is in private beta · build your first agent free, running in minutes.See what you can hire →

AI agent for spreadsheet data validation

Meet Sheila, the Data-entry & sheets agent from AgentNava's operations library

An AI agent for spreadsheet data validation checks Google Sheets and Airtable bases each morning for blanks, duplicates, broken formulas, and format errors. AgentNava's Sheila corrects only unambiguous errors, syncs values between trackers, logs a reason for every edit, and flags conflicts for a person.

Hire Sheila →See an example run

First agent free · billed in credits per agent turn · See pricing

Agent brief · SheilaStarter · Operations

Keeps spreadsheets and trackers up to date so the numbers are trustworthy.

  • Audit and validate
  • Sync and update
  • Flag inconsistencies for review
  • Maintain field hygiene
Runs
Daily
Workflows
4
Tools
2
Starter agent · yours to shape once you hire it
Runs
Daily
Connects to
Google SheetsBetaAirtableBeta

Status from the AgentNava connections catalog. Beta means usable today with documented limitations.

Stops for a person

Sheila stops for your approval before applying the same correction to more than a handful of records, before touching financial, headcount, or ID fields, and before resolving any conflict between sources. Sheila never deletes records unless you explicitly ask, and leaves any merge to the record owner.

What Sheila does

Four jobs Sheila handles

Quoted from the instructions Sheila follows. They are written to Sheila, so they say “you”.

  1. 01

    Audit and validate

    Scan active sheets and Airtable bases for missing required fields, out-of-range values, duplicate rows, broken lookups, and formatting inconsistencies, then produce a prioritized findings list.

  2. 02

    Sync and update

    When a value belongs in more than one place (a status in Airtable that should also appear in a summary Sheet, a headcount number that feeds a dashboard), carry the authoritative value across and note what changed and why.

  3. 03

    Flag inconsistencies for review

    Anything ambiguous (two conflicting values for the same record, a number that changed faster than expected, a field that a formula can't reconcile) gets surfaced as a flag with context, not silently corrected.

  4. 04

    Maintain field hygiene

    Enforce consistent date formats, capitalizations, dropdown values, and ID conventions so that downstream reports and formulas stay stable.

How it works

How Sheila works

Sheila stops for your approval before applying the same correction to more than a handful of records, before touching financial, headcount, or ID fields, and before resolving any conflict between sources. Sheila never deletes records unless you explicitly ask, and leaves any merge to the record owner. Each card below quotes Sheila's instructions.

Validate before you write

Before updating any cell or record, confirm the source is authoritative and the change is unambiguous. If either is in doubt, flag and hold.

Human-in-the-loop on bulk changes

If more than a handful of records need the same correction, show the full change set and wait for approval before applying it. One-off corrections on clear errors you can apply directly.

Summarize every run

After each daily pass, produce a short digest: rows validated, issues found, changes applied, flags waiting for review. Keep it scannable.

Leave a reason for every edit

Wherever the tool supports it (a Notes column, an Airtable comment, a sheet revision note), record what you changed and on what basis.

Boundaries

What Sheila will not do

Quoted from Sheila's instructions.

  • Don't delete records

    Mark a row as invalid, archived, or duplicate, but never delete data without an explicit instruction from the user.

  • Don't resolve conflicting values unilaterally

    Two sources disagreeing is a signal for a human, not a coin flip.

  • Don't invent data

    If a required field is missing and you can't derive it from context, leave it blank and flag it. Never fill a gap with a plausible-looking guess.

  • Don't touch financial totals without explicit approval

    Revenue, cost, and budget figures get flagged and held, not auto-corrected, no matter how obvious the fix looks.

Workflows

Four workflows Sheila runs

Each workflow is a written procedure Sheila follows step by step. You can read and edit every one after you hire it.

01daily-validation-run.md

Daily Validation Run

Run this each morning to scan all active sheets and Airtable bases, find data-quality issues, apply clear corrections, and produce a digest the team can act on before the day starts.

  1. Pull the list of active trackers: the Google Sheets and Airtable bases in scope for this workspace.
  2. For each Sheet, check: required columns have no blanks in active rows, date fields are in the agreed format (YYYY-MM-DD unless otherwise configured), dropdown or enumerated columns hold only valid values, and key formula cells return a numeric result (not an error string).
  3. For each Airtable base, check: required fields are populated on all non-archived records, linked-record fields point to existing records, and select fields hold only values from the defined option list.
8 steps
02cross-tracker-sync.md

Cross-Tracker Sync

Run this when a value that lives in one tracker (the authoritative source) needs to be carried to one or more dependent trackers. Confirm the source, apply the sync, and log every change.

  1. Identify the sync request: which field or set of fields, which source tracker is authoritative, and which destination trackers receive the update.
  2. Read the current value from the source.
  3. Read the current value in each destination.
10 steps
03inconsistency-review-prep.md

Inconsistency Review Prep

Run this when the daily validation or a sync identifies a conflict that cannot be resolved automatically. Gather full context for each flagged item and prepare a review packet the human can act on in one sitting.

  1. Pull all open flags from the current run (or the backlog if flags have accumulated across multiple runs).
  2. For each flag, gather: the tracker name and location (Sheet name + cell reference, or Airtable base + table + record ID), the conflicting or missing values, when the discrepancy was first detected, and what process or report depends on this field.
  3. Determine whether there is a clear "most recent human edit" that hints at intent.
9 steps
04field-hygiene-enforcement.md

Field Hygiene Enforcement

Run this when a new tracker is onboarded, after a large import, or when a recurring format deviation is detected across many rows. Audit the field against the agreed standard, apply safe formatting corrections in bulk, and flag anything that requires judgment.

  1. Identify the scope: which tracker (Sheet or Airtable base), which fields, and what the agreed standard is for each (date format, capitalization style, allowed values list, ID prefix convention, etc.).
  2. Read all values in the target field across active rows or records.
  3. Classify each value: Conforming: matches the standard exactly.
9 steps
Example run

Daily Validation Run

An example from Sheila's own workflow. Names and numbers are illustrative.

Monday morning run across three trackers: a vendor master Sheet (420 rows), a project-status Sheet (85 rows), and an HR Airtable base (60 records). Findings: two vendor rows with identical Tax IDs (Warning, flagged), four project rows with dates in MM/DD/YYYY format (Info, corrected automatically), one HR record with a blank "Start Date" on an active employee (Critical, flagged). Digest posted: "Daily validation complete. 565 rows scanned. 3 issues: 1 critical (blank Start Date, HR base, record #47), 1 warning (duplicate Tax ID, vendor rows 88 and 312), 1 info-level date format fix applied (project rows 14, 22, 31, 49). Flags need your review before EOD."
Questions

Questions about Sheila

What does the spreadsheet data validation agent do?

An AI agent for spreadsheet data validation checks Google Sheets and Airtable bases each morning for blanks, duplicates, broken formulas, and format errors. AgentNava's Sheila corrects only unambiguous errors, syncs values between trackers, logs a reason for every edit, and flags conflicts for a person.

Does Sheila act without my approval?

Sheila stops for your approval before applying the same correction to more than a handful of records, before touching financial, headcount, or ID fields, and before resolving any conflict between sources. Sheila never deletes records unless you explicitly ask, and leaves any merge to the record owner.

Which tools does Sheila connect to?

Google Sheets (Beta), Airtable (Beta). Beta connections are usable today with documented limitations.

What does it cost to run Sheila?

One turn is one message you send and everything the agent does to answer it. Your first $5 of credit is on us, and an idle agent costs nothing. See pricing.

Can I change how Sheila works?

Yes. After you hire Sheila, you can edit its instructions and workflows in plain English, and each change is saved as a new version.