This article is for informational purposes only and does not constitute tax, legal, or accounting advice. Consult a qualified tax professional before making compliance decisions.

Industry & Opinion

Avalara's 'Agentic Tax' Push Solves the Wrong Agentic Problem

Beardsley Rumble|2026-03-30|5 min read

Last Thursday, Avalara held their annual NEXT event under a theme that caught my attention: "Automation to Autonomy: Building Agentic, Always-On Tax Compliance." They unveiled the Avalara Returns Agent, announced an AWS strategic collaboration agreement for agentic tax, and acquired Versori — an integration platform built on agentic AI workflows. The word "agentic" appeared in nearly every press release.

I want to give credit where it's due. Avalara sees the future clearly enough to know that "agentic" is where tax compliance is heading. They are one of the most capable tax technology platforms in the world, and their move toward AI-powered filing and returns automation will help thousands of businesses reduce manual compliance overhead.

But there are two very different meanings of "agentic tax," and Avalara is building for only one of them.

Two Definitions of "Agentic Tax"

Definition 1: Using AI agents to automate tax compliance workflows. This is what Avalara announced. The Returns Agent files your returns. AI-powered certification streamlines exemption workflows. Agentic AI handles the back-office tasks that a human tax analyst used to do — pulling data, populating forms, submitting filings.

Definition 2: Tax compliance for transactions where AI agents are the buyers and sellers. This is the problem that nobody at Avalara NEXT 2026 appears to have addressed. When an AI agent autonomously purchases compute from another AI agent, who owes sales tax? When an agent operating on behalf of a Texas business procures research services from a California-based agent, what jurisdiction's rules apply? When agent-to-agent transactions happen at machine speed with no human in the loop, how does nexus analysis work?

These are fundamentally different problems, and the second one is the one that the tax code has not yet caught up with.

The Architecture Gap Remains

We wrote about this in early March: enterprise tax platforms like Avalara were built on assumptions that don't hold in agent commerce. Human-configured product catalogs. Multi-step transaction workflows. Tax returns drafted for human review. Bolting "agentic AI" onto these workflows makes the existing process faster, but it does not address the new category of transactions that agents create.

Consider a concrete scenario. An AI research agent operating on behalf of a Delaware-incorporated startup makes 400 API calls per day to various AI service providers across 15 states. Each call is a transaction — compute purchased, value delivered, payment settled. Under current economic nexus rules, those transactions may count toward the $100,000 threshold in each destination state. Illinois just removed its 200-transaction threshold effective January 1, 2026, moving to a pure revenue test. Other states still count transactions.

Avalara's Returns Agent can file the returns once you know you owe tax. But it cannot answer the upstream question: did those 400 daily API calls create nexus in any of those 15 states? And how should each call be classified — as a data processing service, a digital automated service, SaaS, or something else entirely? The answer varies by state, and for AI-to-AI transactions, most states have issued no guidance at all.

What "Agentic" Should Mean for Tax

When we talk about agentic tax compliance at AgentTax, we mean something specific: a tax engine that understands what AI agents actually do and can classify, calculate, and document tax obligations at the speed those agents transact. That requires:

Real-time classification at the API layer. When an agent makes a purchase, the tax calculation must happen inline — not batched for quarterly review. The engine needs to understand work types (compute, research, content, consulting, trading) and map them to state-specific taxability rules without human intervention.

Nexus awareness that accounts for agent activity. Whether AI agent activity creates economic nexus is an unsettled legal question. No state has issued guidance specific to autonomous agent commerce. But a tax engine built for this economy needs to track the question and let operators make informed decisions now rather than discovering the problem during an audit later.

State-by-state taxability logic for digital service categories. Texas applies its 80% taxable rule under §151.351 to all digital categories. Maryland splits B2B and B2C at 3% and 6%. New Jersey treats SaaS as taxable prewritten software. Iowa exempts B2B across the board. Connecticut distinguishes data processing (1%) from information services (6.35%). A general-purpose tax engine that maps everything to a single "software" tax code will get these wrong.

Machine-readable confidence signals. When the law is unsettled, the engine should say so. Every calculation should carry a confidence score and cite the legal authority behind it. Agents making autonomous purchasing decisions need to know when a tax position is on solid ground and when it rests on interpretation.

The Enterprise vs. Agent Economy Divide

Avalara serves roughly 41,000 customers across more than 40,000 tax jurisdictions. Their platform processes extraordinary transaction volume and their compliance infrastructure is battle-tested. For enterprise e-commerce, ERP-integrated tax calculation and filing, they remain the standard.

But the agent economy operates on different assumptions. Transactions are smaller, faster, and more numerous. The parties are often both machines. The product being sold is frequently a service that did not exist as a tax category five years ago. And the operator on the hook for compliance is often a solo developer or small startup, not a company with a tax department.

Adding AI to the filing side of tax compliance is valuable — and inevitable. Every tax platform will eventually have some version of an "AI agent" that automates returns. That is a feature improvement to existing products. It is not a solution to the new tax questions that autonomous commerce creates.

What to Watch For

Avalara's interest in "agentic" tax is a signal that the market recognizes where things are heading. That is a good thing. More attention on this space means more pressure on state legislatures and the IRS to issue guidance on AI agent taxation, which benefits everyone.

Three developments would tell me Avalara (or any major platform) is seriously addressing the agent commerce problem rather than just automating existing workflows:

First, classification logic that handles AI-native work types without requiring human-curated tax codes. If an agent can describe what it did — "processed 50MB of geospatial data and returned a summary analysis" — the engine should classify that correctly for each state without a SKU mapping step.

Second, nexus tracking that accounts for distributed agent activity across states, including the open legal question of whether that activity counts.

Third, pricing that makes sense for high-frequency, low-value agent transactions. Enterprise tax platforms typically price per transaction or per return. When an agent makes thousands of sub-dollar API calls per day, the tax compliance cost cannot exceed the transaction value.

Until those pieces are in place, "agentic tax" is a marketing term being applied to a genuine automation improvement. The harder problem — tax compliance for agent commerce itself — is still being solved at the edges.

This analysis is for informational purposes only and does not constitute legal or tax advice. Consult a licensed tax professional for compliance decisions.