Cracking the Code

One way to simplify the tax code might be to start slashing the 2,000+ sections spread across 7,000+ pages. But that approach would make the code incoherent. Sections would reference others that no longer exist. Core tax calculations could vanish while obscure relics survive. It would be chaos.

Thankfully, an AI agent can be designed to behave more like a scalpel than a hammer — but it still needs context. If the goal is simplification, the agent must understand where complexity is clustered. If the goal is revenue neutrality, it must know which sections are financially significant and what consequences any edits might have.

This led to two important first steps:

Locating Complexity

An AI agent dedicated to complexity assessment looked at each section of the tax code across four dimensions:

• IRS Bulletin Mentions: How often the IRS had to reference a section for clarification. More clarifications generally point to higher complexity.
• Amendment Count: How frequently a section has been revised. Frequent changes introduce inconsistency and confusion.
• Section Length: Simply, how verbose the text is.
• Language Complexity: How jargon-heavy or dense the legal language is.

Running this analysis through an AI agent produced a complexity score for each section:

Some sections stood out immediately — bloated by years of amendments and legalese, difficult even for professionals to interpret.

Mapping Impact

Impact has two dimensions:

• Entity Impact: How many people, corporations, or partnerships are affected.
• Financial Impact: How much revenue flows through that section.

A second AI agent was built to map both dimensions by leveraging IRS tax statistics, form rules, and the text of the code itself. This surfaced a wide range — from foundational sections like the progressive income tax brackets to quirkier ones, like Section 4451, which imposes a tax on every pack of playing cards with no more than 54 cards. (Yes, seriously.)

Fortunately, Google’s Gemini model is multimodal — meaning the AI agent could interpret images of tax forms as easily as text. This multimodal capability helped the agent cross-reference form line items to actual revenue figures, building a clearer picture of financial impact.

The result: a 3D “navigation tool” plotting complexity against financial and entity impact.

In this plot:

• The blue region highlights sections that are highly complex but low in impact.
• These are ideal candidates for simplification — areas where cleanup has minimal downside risk.

The dollar amounts assigned to each section should be treated as relative indicators, not literal values. The agent cataloged about $460 billion in revenue across sections, while the IRS collects over $4.2 trillion annually. So these estimates are for prioritization purposes rather than exact accounting.

In my next post, I’ll walk through the AI editor agent — how it approached rewriting the code, where it chose to simplify, and what the first round of edits looked like.

• §1 – Tax Imposed
• §11 – Corporate Tax Imposed
• §121 – Exclusion from Sale of Principal Residence
• §108 – Discharge of Indebtedness

Each of these sections could benefit from simplification efforts such as clearer structure, fewer cross-references, and reduced conditional logic.

🧹 High Complexity + Modest Relevance

Sections that are technical and complex, but not widely applicable, and therefore suitable for rewriting or consolidation:

• §1059 – Corporate Shareholder’s Basis Adjustments
• §1031 – Real Property Exchanges
• §1015 – Gift Basis Rules