Cutting a 90-minute podcast, unsupervised
Filler words gone, chapters spot-on — but two tools cut a punchline in half. Human review still needed for the final 10%.
Read the test →A living directory of real-world tasks — each one tested, rated and dated. No hype. Just verdicts, tools and copy-paste prompts.
Every month we take one real task, hand it to three AI systems, and publish the results — fails included. These verdicts feed the directory above.
Filler words gone, chapters spot-on — but two tools cut a punchline in half. Human review still needed for the final 10%.
Read the test →All four shipped a working page. The differences were in taste, not capability — full breakdown with screenshots inside.
Read the test →Great at gathering and pre-filling. But every system made at least one filing error a Finanzamt would flag. Assist: yes. Autopilot: no.
Read the test →One email per month: which tasks changed status, and what we tested.
No spam. Unsubscribe anytime.