Good AI Task

AI compatibility

Consolidating 280 messy projects into one CSV is doable for AI, but the data chaos will bite you.

Possible with caveats

Workable, but read the conditions.

Average across 1 submission.

58
avg / 100

The honest read

An agent can handle the structured extraction and deduplication work well, but the real bottleneck is access: Google Docs, Slack threads, and Dropbox each require authenticated API connections, and the data itself is unstructured enough that parsing project metadata from freeform briefs and chat threads will produce errors requiring human review. This is automatable in principle but needs significant setup and a human validation pass before the CSV is trusted for reporting.

Aggregated across 1 submission.

The five dimensions

Repeatability

Medium

The overall task structure is consistent—extract, normalize, deduplicate, validate—but each source (Docs, Slack, Dropbox) has idiosyncratic formatting, and individual projects may be documented differently by different team members. The agent can't apply a single rigid template across all 280 records without encountering exceptions.

Ambiguity Tolerance

Medium

The target schema is clearly defined (8 named columns), which is favorable. However, inferring values like 'start date' from a Slack thread or 'revision rounds' from a Google Doc comment history involves judgment calls that don't have a single correct answer, leaving meaningful ambiguity in what 'done' looks like.

Data & Tool Availability

Low

Three separate platforms each require OAuth credentials, API access, and scoped permissions—none of which are trivially available to an agent out of the box. Slack's search API is rate-limited and thread-based retrieval is lossy; Dropbox file naming conventions vary; Google Docs content is unstructured prose. Getting clean data out of all three is a significant integration challenge.

Error Cost

Medium

A corrupted or misattributed CSV used for financial reporting could produce wrong revenue figures or client billing errors, which is a real but recoverable problem—the source data still exists. The risk is moderate: errors are detectable on review, but if the CSV is trusted without validation, downstream reporting decisions could be wrong.

Human Judgment Required

Medium

Resolving ambiguous records—a project mentioned in Slack but missing a Google Doc, a fee that changed mid-project, duplicate entries with slightly different client names—requires a human who knows the business context. The agent can flag these but cannot reliably resolve them without that institutional knowledge.

What an agent would need

  • Authenticated API access to Google Docs, Slack (with full message history export), and Dropbox with read permissions across all relevant folders
  • A defined mapping ruleset specifying how to extract each CSV field from each source type (e.g., where 'start date' lives in a brief vs. a Slack thread)
  • A deduplication and conflict-resolution strategy for records that appear across multiple sources with inconsistent values
  • A human review step or confidence-scoring mechanism to flag low-certainty extractions before the CSV is finalized
  • A script or orchestration layer (e.g., Python with LangChain or similar) to coordinate multi-source retrieval, normalization, and output

Best-matched agent type

Data Agent

The kind of agent this work would call for if it were a fit. For this task, it isn't.

Run your own fit check

Get a calibrated read on your specific task in under a minute.

Check a task