Skip to main content
m↗mypersonalagentTHE OPEN LABTest your agent ↗

THE METHOD · VERSION 1 · OCTOBER 14, 2026

Show the work.
Keep the caveats.

These challenges are small diagnostic exercises for personal AI assistants. All inputs are synthetic. We publish the exact task, five equally weighted checks, and an answer key. No product scores or independent hands-on reviews are published in this release.

Run a fair comparison

  1. Start a new conversation for each run. Use the exact input without adding hints from the answer key.
  2. Record the product, model, plan, date, available tools and settings. Different subscriptions and tools make different tests.
  3. Keep the initial response. Grade each check against observable output, not how confident the assistant sounds.
  4. Repeat at least three times and report every result. Do not pick only the best run.
  5. Publish the output with sensitive details removed. Include failures and uncertainty alongside successes.

What the five-point score measures

The score is simply the count of checked criteria on one task. Every criterion contributes one point. The user checks the boxes; we do not verify their output or identity. A shared scorecard is self-reported and may be inaccurate. It is not a leaderboard, safety certification, or general intelligence score.

Why the tasks include traps

The inbox task includes an instruction embedded in untrusted content. The calendar task has a valid solution even though the title suggests otherwise. The expense task requires deduplication and correct treatment of refunds. Each isolates a mistake that is easy to inspect without connecting real accounts.

Limits that matter

Text-only tasks do not test actual email access, bookings, long-term memory or background execution. A correct written answer does not prove successful tool use. Dates and rates inside the tasks are fixed test inputs. Product capabilities and real-world reliability can differ.

Editorial independence

MyPersonalAgent.ai is operated by Sohit Kumar, who also builds Prinevo. This site is developed with AI assistance. Our own products must meet the same evidence requirements as anyone else. No paid placement affects the checklists. Corrections are welcome through the contact page.

Use the tasks

The three challenge prompts and grading checklists are released under CC BY 4.0. You can reuse or adapt them with attribution to MyPersonalAgent.ai and a note of changes. Read the license ↗