Good AI Task

AI compatibility

AI can do most of this Go refactor, but the corruption logic needs a human eye.

Possible with caveats

Workable, but read the conditions.

Average across 1 submission.

62
avg / 100

The honest read

A code agent can handle the mechanical parts of this well—adding structured logging, wiring up retry logic with exponential backoff, and scaffolding a dead-letter queue are all well-understood Go patterns. The harder parts are understanding the existing codebase's implicit assumptions, deciding where partial-sync corruption actually occurs, and writing integration tests that faithfully mock Salesforce's real failure modes—those require judgment the agent can only approximate without deep context.

Aggregated across 1 submission.

The five dimensions

Repeatability

Medium

The individual sub-tasks (structured logging, exponential backoff, dead-letter queue) are structurally repeatable patterns in Go. However, applying them correctly to a specific 3,000-line codebase with undocumented partial-sync behavior requires one-off judgment each time.

Ambiguity Tolerance

Medium

Some success criteria are crisp (retry logic exists, tests pass), but 'comprehensive error handling' and 'corrupting partial syncs' are underspecified—the agent must infer what counts as a corruption boundary and what failure modes the mock API must cover.

Data & Tool Availability

Medium

The agent needs full access to the existing codebase, Salesforce API schema or docs, and the Postgres schema to do this correctly. If those are provided, execution is feasible; missing any one of them causes the agent to guess at critical integration points.

Error Cost

High

This is production infrastructure touching a live Salesforce-to-Postgres sync. Incorrect retry logic or misidentified corruption boundaries could silently worsen data integrity rather than fix it, and bad integration tests give false confidence. Damage is potentially hard to detect and reverse.

Human Judgment Required

Medium

Choosing where to draw transaction boundaries, deciding which failures are retriable vs. fatal, and designing a dead-letter queue that fits the team's operational workflow all benefit from domain knowledge a human maintainer holds. The agent can produce plausible answers but may miss implicit constraints.

What an agent would need

  • Full read access to the existing 3,000-line Go codebase, including any config files and dependency manifests
  • Salesforce API documentation or a schema dump covering the endpoints the tool calls, including known error codes
  • Postgres schema for the target database so transaction and rollback boundaries can be correctly identified
  • A clear definition of what constitutes a 'partial sync corruption'—which tables, which operations, and what the acceptable failure behavior is
  • A test environment or CI pipeline where the generated integration tests can be run without touching production data

Best-matched agent type

Code 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