Fewertools research · Executed tasks
Five real product workflows, with synthetic inputs, saved outputs and the limits left visible. Each record tells you exactly what was run and how.
Nine native card instances; six inherited labels updated, three overrides preserved, and three PNG/SVG exports checked.
Authenticated Figma connector and native Plugin API construction, mutation and readback; browser canvas inspection; native PNG/SVG export.
Read the workflow record →Six title/headings, two lists, a budget table, Unicode, a hyperlink and three dates survived native document checks and exports.
Authenticated Google Docs connector using native document requests, semantic readback and browser inspection; actual native DOCX/PDF exports independently parsed.
Read the workflow record →Twenty invoice rows;12 metrics matched, an amount edit recalculated correctly, and the due-date boundary and native XLSX were checked.
Authenticated native Google Sheets import and formula requests, bounded native value/formula readback, browser worksheet inspection and native XLSX/Summary CSV exports. Expected results computed independently from the CSV fixture.
Read the workflow record →Create a private repository, isolate a due-date fix on a branch and inspect an unassigned draft PR. Local Node results pass; the stalled upload step is recorded.
Authenticated GitHub web UI for private repository, commits, branch and draft PR; local Node test execution of fixture and rendered source readback.
Read the workflow record →An anonymous four-field preview rejected missing values and an invalid email, then completed with synthetic data. Conditional branches and configured ranges remain untested.
Browser UI authoring and keyboard interaction; visible preview and DOM readback. No API construction.
Read the workflow record →These are agent-executed workflows through authenticated connectors or browser controls. Connector-assisted document construction is labelled. An export check does not prove collaboration quality, AI performance or long-term reliability. Failed and blocked steps are recorded beside successful ones.
The selection combines limited recorded visitor interest with available tool access. It is not a list of the five most visited products. No general quality ranking is inferred from one synthetic task.
A comparison may link the result for one product while the other remains untested. A head-to-head finding needs the same task executed in both products. Historical opinion indices keep their original scope and are not automatically renewed by these tests.