Dayjob: Tax - evaluating AI agents on 50 professional tax assignments

A closer look at Dayjob: Tax - evaluating AI agents on 50 professional tax assignments

What DAYJOB: Tax is

DAYJOB: Tax is a benchmark that evaluates AI agents on professional tax assignments. It tests whether agents can take a client request and supporting records all the way to a finished deliverable while dealing with messy, real-world tax work where instructions are incomplete and records can conflict.

The benchmark is part of Surge AI's DAYJOB family and extends earlier releases that covered Healthcare and Finance into tax preparation, compliance, and advisory services.

What you can do on the DAYJOB: Tax benchmark

Agents are given client requests and workspaces (1,789 files total across 50 environments) and must produce outputs such as tax returns, workpapers, technical reviews, or client advice. Assignments include varied materials: PDFs, Excel workbooks, Word docs, emails, images, payroll records, returns, K-1s, notices, ledgers, and other evidence tax professionals use.

Tasks require not only tax calculations but professional judgment: deciding what to investigate, resolving conflicting evidence, identifying consequential issues a client may not mention, and meeting grading criteria set by tax experts.

Who created and who will find it useful

Tax experts created and reviewed the tasks: assignments are developed from real professional workflows and undergo three-layer expert review for correctness, completeness, and solvability. Reviewers include practitioners with experience in corporate taxation, partnership and corporate taxation, and real-estate and individual taxation.

This benchmark will be useful to people evaluating AI agent performance on realistic, complex tax work - teams measuring agent capability on tasks that require domain knowledge plus professional judgment, and researchers comparing agent performance with a leaderboard and Pareto analyses.

Questions people ask about Dayjob: Tax - evaluating AI agents on 50 professional tax assignments

How many assignments and files are in the benchmark?

The benchmark contains 50 environments and 1,789 workspace files, with a median of about 30 files per assignment and as many as 123 in some tasks.

What kinds of tax work are covered?

DAYJOB: Tax covers eight areas of tax practice including individual and family tax, estates, gifts, and trusts, and includes tasks ranging from filing packages to investigative reviews and advisory work.

How is agent performance measured?

Performance is reported on a leaderboard using pass@1 at a 95% rubric threshold; an attempt passes when it satisfies at least 95% of the task's criteria assessing calculations, evidence, consequential issues, and the deliverable.

Dayjob: Tax - evaluating AI agents on 50 professional tax assignments was shared on NexURLs by Nexis on October 31, 2026, in Business & Startups.