agent timed out when the CRM took 90s to load the account page

TL;DR

A 90-second account page is usually a data-heavy account (years of activities, huge related lists) rendered in full. The agent does not need the page; it needs fields. Read them from the CRM API, or use a lightweight list or report view, and reserve the full page for the rare case with a strict budget and a screenshot on timeout.

agent timed out when the CRM took 90s to load the account page

Use this when

• Timeouts concentrate on large, old accounts
• The needed data is a handful of fields or lists
• Humans also find these pages slow but tolerate them

Not for this skill when

• All CRM pages are slow, a vendor performance case
• The required data exists only in the page's rendered view

Steps

1. Time the page load for a normal account versus a failing one and note what differs (activity count, related lists). Verify: the heaviness correlates with a measurable account trait
2. Map each needed field to the CRM API or a report view. Verify: every field has a non-page source or is explicitly page-only
3. Switch the read path to the API or view, keeping the browser path behind a strict budget as fallback. Verify: large-account reads complete in seconds on the new path
4. For page-only data, load with lazy sections disabled where the CRM allows, and screenshot on timeout. Verify: the fallback is bounded and diagnosable
5. Log per-account read durations and alert on accounts crossing the slow threshold. Verify: the slow-account population is tracked as data grows

Variant phrasings

CRM account page 90 second load

A related phrasing for the same underlying problem.

agent timeout on heavy account page

A related phrasing for the same underlying problem.

read CRM data via API not page

A related phrasing for the same underlying problem.

Compatibility: Compatibility: Salesforce or HubSpot account records accessed by browser agents and their REST APIs, current versions.

Why it happens

Page rendering cost scales with everything the account ever did, while API field reads cost the same for every account. Moving reads to the API removes the timeout's cause rather than budgeting around it, and the heaviest accounts, usually the most valuable, stop being the ones the agent cannot see.

Edge cases / pitfalls

• API reads need the same field-level permissions as the page. A missing permission looks like missing data; verify with an admin read.
• Some page data is formula or rollup fields the API does expose, but under different names. Map names carefully during the switch.
• Keep a small set of known heavy accounts as regression fixtures for every agent release.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_ZYSjRVYnIrgQ4gwYQWo87Q