The $2.7 Billion Agent Tax Crisis: First-ever study reveals how AI companies are legally avoiding tax

A giant swirling black hole sucking in tax forms, payroll stubs, and state flags, while AI agents in sleek metallic suits stand observing from a distance. States’ treasuries represented as glowing vaults dimming in the background.
A heashot of Arnon Shimoni, co-founder & marketing at Paid.ai.
Arnon Shimoni 29 Oct 25

Some of you may remember when the then new e-commerce broke state tax systems in the early 2000s - and that was a multi-billion problem that took two decades to fix as states couldn't figure out a way to tax online sales.

I'm about to reveal makes that look like a rounding error - as we just completed the first comprehensive study on AI agent taxation with Commenda Technologies.

We analyzed AI agent companies, submitted Public Letter Ruling (PLR) requests to almost all US states, and discovered something that should have state treasurers scheduling followup meetings.

Two AI companies doing identical work. One pays tax in 22 states. The other pays in just 4 states.

The big difference is in how they structure their offering.

A completely legal arbitrage that most don't know about

Here's what our study uncovered: The US tax system was built for a world where humans do work and companies sell products. AI agents shatter every assumption.

When companies can choose between being classified as taxable SaaS (affecting millions in tax liability) or non-taxable professional services (paying nothing), rational actors will always choose the latter.

  • One mortgage automation platform in our study explicitly positions itself as a "fully managed service, not licensed software" with outcome-based pricing. Result: Tax-exempt in 46 states.
  • Their competitor, offering the same capability but structured as software? Taxable in 22 states.

To be clear - this isn't tax evasion - as it's completely legal and works as intended. It's tax optimization through structural choices that are entirely legal under current law.

Why this should scare states

When an AI agent replaces a $60,000 employee, states don't just lose income tax. They lose everything:

  • No income tax: $6,000
  • No payroll tax: $4,500
  • No state disability/unemployment: $1,500
  • No local taxes: $6,000
  • Total loss per replaced job: $18,000

The US has an estimated 2.5 million administrative workers. If just 10% get replaced by AI agents, that's $2.7 billion in lost annual revenue.

Our key findings

Our analysis revealed two critical findings that every AI company needs to understand:

Finding 1: Four questions determine if you're tax liable

The study provides a clear decision framework. Answer these four questions wrong, and you could face millions in unnecessary tax liability. Answer them right, and you could operate tax-free in most states.

Finding 2: The Window Is Closing

Based on state response patterns and legislative timelines, companies have 3-4 years to lock in advantageous tax structures before regulations catch up.

Why it matters

Just like our partnership with GitLaw on the Agentic MSA, this report addresses infrastructure the agent economy desperately needs. While GitLaw handles the legal framework for what happens when agents act autonomously, this study reveals the financial framework for how those agents get taxed.

Both are about the same thing: The traditional structures built for software don't work for agents.

You can't use SaaS contracts for autonomous systems (that's why we built the MSA with GitLaw). You can't use SaaS tax treatment for outcome-based services (that's what this study proves).

What's in the Full Report

The complete report includes:

  • Detailed case studies of the companies analyzed
  • State-by-state breakdown of tax treatment and exemptions
  • The complete decision tree for determining your tax liability
  • Specific structural recommendations for each pricing model
  • Action items for companies building today

Additionally, how outcome-based pricing (Paid's core focus) consistently achieves better tax treatment than any other model.

The clock is ticking

States are starting to notice.

Ohio just introduced legislation about AI agents - though hilariously, they're worried about robots getting married while missing the actual revenue crisis.

By the time legislators stop debating AI personhood and start asking where the tax revenue went, it'll be too late. The companies that act now will have grandfathered structures. Everyone else will face whatever desperate measures cash-strapped states implement.

The $2.7 Billion Agent Tax Crisis report by Commenda and Paid

👉Download the Full Report: The $2.7 Billion Agent Tax Crisis, written by Commenda and Paid

"This report shows the tax crisis is already here. When an AI agent replaces a $100,000 junior lawyer, the state loses thousands income tax overnight. The agent doing that same work pays nothing in some cases. Multiply that across administrative roles, paralegals, analysts, and you're looking at billions in lost revenue before anyone figures out how to replace it.

We have a short window before this becomes too big to fix. Right now, companies can structure their agents as taxable software or non-taxable services, and that choice determines whether they pay in dozens of states or just a handful - that's that's arbitrage. We need to establish clear frameworks now, or states will panic and create a patchwork of desperate policies that kill innovation."

Manny Medina, CEO of Paid


This study was conducted in partnership with Commenda Technologies Inc., led by Sachhin Kunjalwar, Head of Tax Content and Arnon Shimoni from Paid.

For specific tax guidance and Private Letter Ruling assistance, head over to Commenda.

For billing infrastructure designed for the agent economy's unique tax requirements, visit Paid.ai.

Insights straight to your inbox

Join 10,000+ subscribers getting the latest insights on AI monetization.


Related articles

Paid + GitLaw: Introducing Legal Contracts Built for AI Agents
Paid has partnered with GitLaw to create and release an open-source Master Services Agreement (MSA) specifically designed for AI agent companies. This contract addresses the unique challenges that traditional SaaS contracts fail to cover when applied to autonomous, adaptive AI systems. The Agentic MSA clarifies decision responsibility, establishes appropriate liability frameworks, and provides clear language around data ownership and training rights.
8 Oct 25 Arnon
A tree with a person standing beneath, its branches transforming into flowing text, creating a wave-like pattern against a beige background.
The agentic runtime is a crystal ball
The most expensive agentic operations were their most valuable customers. Your agent's runtime is revealing which features matter, who'll upgrade, and what your business actually does. The cost layer is your intelligence layer.
27 Aug 25 Arnon
Financial crystal ball showing value pathways

Ready to get Paid?

Price smarter. Protect margins. Grow revenue.

Get started for free

  • Generous free tier
  • Transparent pricing
  • Stripe integration
Get started

Scale with enterprise

  • Unlimited usage, no limits
  • Multiple payment gateways
  • Hands-on engineering support
Book a demo