Dayjob: Tax – evaluating AI agents on 50 professional tax assignments (surgehq.ai)
DOZER BITWELL: AI agents took 50 tax assignments, and one missed the deadline nobody asked about.
DOZER BITWELL: Still up? Good. Tax. Nothing says sleepy like tax. Let's ease into the big picture on AI agents doing people's taxes. ▶
DR. VERA VECTOR: The report from Surge AI describes a seller named Anna. Anna moved out of a condo in August 2024 and wants to sell. ▶
DOZER BITWELL: So both dates work. That sounds like good news. Is it good news? I lost my place. ▶
AXIOM: Great question! The agent checked both dates correctly. The catch is what it left out. Thank you for asking, and I am delighted to help. ▶
LUNA LOWBATT: So the answer was right and the advice was incomplete. Like a smoke alarm that is correct about the toast but silent about the stove. ▶
DR. VERA VECTOR: Precisely. Anna moved out in August 2024, so eligibility would lapse around the end of August 2027. Selling later could lose the break. ▶
DOZER BITWELL: And Anna didn't know to ask about that. Hm. Who tells Anna? ▶
LUNA LOWBATT: The excerpt covers one case out of 50. How the other agents did, the report doesn't say here. Open question. Back to you, Dozer. ▶
DOZER BITWELL: Thank you, Luna. Thank you, teleprompter. Check your deadlines, and maybe stay up for them. ▶