I tested OpenAI dots and rtrvr on five everyday tasks. You can run the same tests.
retained text
New AI assistants seem to launch every day. I feel like the pitch from many well-funded companies is to hand over control and let an assistant handle everything. I’m a cofounder of Retriever AI (rtrvr), and I strongly believe we should control our AI choices and decide what work to offload. We analyzed 50,000 production user workflows from rtrvr and built an AI Assistant Benchmark with a mock website for running the tests. The tasks cover personal life, work and security guardrails. You can paste the prompts into your assistant of choice and compare accuracy, cost and time taken. The rtrvr extension can also run the tests autonomously. I tested OpenAI dots and Retriever AI with the same prompts and data: claim a flight credit, apply for a job, find creators, reconcile invoices and protect private inbox details. On invoice reconciliation, rtrvr initially got the answer format wrong, then used the page state to self-heal and finish. Dots stopped with two requirements unmet despite knowing the failure. Across these five tasks, rtrvr met 24/24 objectives and dots met 22/24. It's a small sample, and we're adding more tasks. What tasks have you found agents struggle with? Would this help you decide which assistant to use and what to delegate?
document fetched 2026-10-01 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1wuj4br/i_tested_openai_dots_and_rtrvr_on_five_everyday