How agent-ready is Expensify?

Expensify's agentability audit returned an overall score of 24/100 on its homepage, 33/100 on pricing, and 35/100 on documentation. These scores indicate that autonomous AI agents will encounter significant friction when attempting to parse content, navigate workflows, and execute tasks on the platform. The product currently lacks the structured feedback, control mechanisms, and transparency signals that agents rely on to operate reliably.

The audit identified machine readability as a relative strength (78/100), suggesting that page structure and semantic markup are largely in place. However, critical gaps appear in control (20/100), status reporting (0/100), clean handoffs (0/100), shadow UI avoidance (0/100), and transparency (18/100). These weaknesses mean agents cannot gracefully handle errors, pause long operations, or interpret the confidence level of system outputs.

Addressing these deficiencies will improve both agent interoperability and human usability, as the underlying issues—opaque errors, missing confirmation steps, and absent undo mechanisms—affect all users attempting programmatic or assistive interactions with Expensify.

To improve agentability, Expensify should prioritize the following engineering changes:

• Return structured error context in both UI and API responses, including the specific field that failed, an error code, and a human-readable explanation, rather than generic 'Something went wrong' messages.
• Introduce explicit confirmation dialogs for destructive actions (such as deletions or bulk changes) with distinct visual styling and clear labelling to prevent accidental execution by agents or users.
• Provide labelled pause, cancel, or stop controls for long-running operations like imports or bulk processing, with accessible names such as 'Cancel import' instead of icon-only buttons.
• Implement undo functionality for completed destructive actions, allowing agents and users to reverse recent changes without manual recovery steps.
• Ensure all pages use a single <h1> element and maintain a contiguous heading hierarchy without skipped levels to support reliable document parsing.
• Add transparency signals to key outputs, such as confidence scores in API responses or 'verified vs. inferred' labels in the UI, so agents can assess the reliability of data they consume.