Internal Agents Map

AI Annotator

Plaid's internal labeling agent for its own model training. AI Annotator automates large-scale labeling of anonymized transaction data, with targeted human validation when a product use case requires it. Plaid reports greater than 95% human alignment at a lower cost and time than manual labeling.

Company
Plaid
Approach type
Agent
Work
Data
Human involvement
Unknown
Deployment stage
Scaled
Evidence strength
Detailed primary
Entry reviewed
Copy as MDAsk ChatGPT

How it works

Not reported: The capture names no run. The "Key capabilities" sentence introduces no list, and the voting and feedback mechanisms are future work only.

Where people stay involved

People review, approve, or guide the agent’s work here.

Reported: Targeted human validation is applied when a product use case requires it; no default review of every labeling run is stated.

  • Supervision level

    Not classified

    • raw transactions → labeled training data[1]

How supervision levels work

Implementation details

Model
Not reportedLLMs are named as the enabling technology; no model is specified.
Harness
Internally hosted platform[1]
Sandbox
Not reported
Tool access
Not reportedNo tools are named; labeling runs on internal transaction data.
Knowledge
Anonymized Plaid transaction data[1]
Context management
Not reportedNot documented; richer context signals are future work.
Credentials
Not reported
Interfaces
Not reported

Validation and failure handling

Reported: Summary

Reported observations

These results come from the sources. Notes below each result explain what it covers and any reported limits.

Headline claim

Greater than 95% human alignment at lower cost and time than manual labeling[1]

Reported by
Plaid
Scope
Transaction labels generated by AI Annotator in early use
Denominator
Labels compared with human judgments; sample size unspecified

Observed in June 2025

About this result
Category
Effectiveness
Evidence type
Reported measurement
What it covers
AI Annotator label agreement with human judgments in early use

Lessons

Not reported: The capture states impact and future plans; no evaluated transferable lesson.

Back to agents
Improve this record

Give your coding agent the context and instructions to prepare a source-backed update.

Suggest on GitHub

Paste the prompt into your coding agent, then describe the correction or add a public source.

Preview prompt

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

  1. AI agents at Plaid (June 2025)https://plaid.com/blog/ai-agents-june-2025/Engineering blog · First party · Last source verification: 2026-08-31
    Preserved copy
Question coverage and scope
purpose
Reported
workflow
Not reported: The capture names no run. The "Key capabilities" sentence introduces no list, and the voting and feedback mechanisms are future work only.
human involvement
Reported: Targeted human validation is applied when a product use case requires it; no default review of every labeling run is stated.
implementation
Reported
validation
Reported: The documented check is the targeted human validation cited with the human-involvement answer; no separate automated check is documented.
observations
Reported
lessons
Not reported: The capture states impact and future plans; no evaluated transferable lesson.
Research details for every claim on this page
  1. Summary
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    Not assessed
    Confidence reason
    No explicit confidence assessment is recorded for this claim.
  2. Headline claim
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    Plaid reported the figure in its own blog without independent verification.
    Reported by
    Plaid
    Scope
    Transaction labels generated by AI Annotator in early use
    Denominator
    Labels compared with human judgments; sample size unspecified
    Method
    Not reported
    Observation date
    2025-06
  3. Knowledge
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    Not assessed
    Confidence reason
    No explicit confidence assessment is recorded for this claim.
  4. Harness
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    Not assessed
    Confidence reason
    No explicit confidence assessment is recorded for this claim.
  5. Key observation

    Greater than 95% human alignment with labeled data

    Duplicate representation of Headline claim. Component of the compound headline: the same over-95% alignment figure with the same early-use scope and unspecified sample.

    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    Plaid reported the figure in its own blog.
    Reported by
    Plaid
    Scope
    Transaction labels generated by AI Annotator in early use
    Denominator
    Labels compared with human judgments; sample size unspecified
    Method
    Not reported
    Observation date
    Not reported
  6. Key observation

    Lower cost and time than manual labeling

    Duplicate representation of Headline claim. Component of the compound headline: the same cost-and-time comparison without measured values.

    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    Qualitative company comparison; neither actual costs nor elapsed-time measurements are supplied.
    Reported by
    Plaid
    Scope
    AI transaction annotation cost and time relative to manual labeling
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    Not reported
  7. Operating model assessment
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Unverified
    Confidence reason
    The source describes targeted human validation when a product use case requires it; it does not document a person reviewing the labeled output of each run.
    Observation date
    2025-06