Evaluate whether personal agents remember promised follow-up actions
At a glance
PersonalAgentEval tests everyday tasks against accounts with substantial prior history. Its X report highlights a pharmacy task where most agents omitted a previously promised caregiver notification. This workflow is listed under builder, personalagenteval, agent-evals, context-retrieval, follow-through.
After a Grok Bot cut one user’s Comcast bill by $43/mo, a coffee-shop tester gave Grok Bot, ChatGPT Dots, and Meta Muse the same retention/negotiation job to compare how agents handle real bills.
One user split roles cleanly: Meta Muse handles grocery carts, DoorDash, and Amazon errands as a PA, while Grok Bot stays the workhorse for work automation.
Users of Meta Muse, Grok, OpenAI Dots, and Instinct report hit blocks on major sites—Yelp gating licensed agents, Delta refusing ticket bookings, and big banks keeping external agents out of accounts.
At a fund launch, a founder's agent stack (Grok Bot, Claude, Instinct, Muse, Dots) and the CMO's agents negotiated budget while the humans typed nothing—exposing the need for agents that can talk across stacks.