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

Zamp's $30M Sales Tax OS, Thomson Reuters Backing, and the Agent Layer It Doesn't Touch

Beardsley Rumble|2026-05-09|6 min read

On April 28, 2026, Zamp announced the global launch of Zamp OS — what the company calls "the operating system of sales tax" — alongside a $30 million funding round led by Acrew Capital with participation from Thomson Reuters Ventures. The product targets accountants, controllers, and finance teams that need to register, file, and reconcile sales tax across 12,000-plus U.S. jurisdictions, with AI agents handling the data work and human tax professionals reviewing exceptions.

Zamp is a serious product. The combination of an AI agent layer with credentialed humans is the right design for the problem they have chosen, and Thomson Reuters Ventures is not a frivolous endorsement. The question I want to raise is narrower: which problem is that, exactly, and which problem does it leave open?

What Zamp Built

By Zamp's own description, Zamp OS does the following:

Onboarding and registration across U.S. states and a growing international footprint.

Continuous nexus monitoring — tracking when a seller's economic activity crosses a state's $100,000 sales threshold (or, in the dwindling list of states that still count transactions, the transaction count threshold).

Tax determination on the seller's invoices and revenue feeds, with AI-assisted product taxability mapping.

Filing automation across thousands of jurisdictions, with humans reviewing edge cases.

Notice handling, exemption certificate management, and audit support.

This is the modern incarnation of the workflow that Avalara, Vertex, Sovos, and TaxJar have been building for two decades. The Zamp twist is the explicit AI-plus-human framing — what they call a "managed operating system" rather than a self-service software product. For mid-market sellers and the CPA firms that serve them, this is a meaningful improvement over the existing options.

What "Operating System of Sales Tax" Means in Practice

Read Zamp's launch carefully. The word "seller" is doing the work in every sentence. The premise is that there is a company — a SaaS vendor, a Shopify merchant, a manufacturer — that has identifiable revenue across identifiable jurisdictions, and that the operating system's job is to take that revenue stream and produce compliant returns.

That premise is sound for the workloads Zamp has chosen. It also makes specific assumptions about the world:

One identifiable taxpayer per transaction. A company with an EIN, a state registration, and a chart of accounts.

Transactions aggregated and reconciled in batches. Daily, weekly, or monthly close processes feed the filing engine.

Human-reviewable taxability decisions. "Is this product a digital good, prewritten software, or a SaaS service?" gets reviewed by Zamp's tax team and codified.

A return as the unit of output. The deliverable is a Form ST-1, ST-100, or equivalent, filed on a state's calendar.

When the buyer and the seller are companies — even very automated companies — those four assumptions hold. The Zamp product is well-suited to that world.

The Layer Zamp Does Not Address

Now consider an AI agent acting as a transactional principal. An autonomous procurement agent operating on behalf of a Delaware C-corp pays an x402 invoice for a research deliverable. A user-facing agent built on a CrewAI stack pays per-token for a model API. A merchant's customer-service agent issues a refund and a partial credit to a buyer in another state. Each of those is, or may be, a sales-and-use-tax event.

Each of them also breaks at least one of Zamp's four assumptions:

Identity attribution is unsettled. The principal is not always the operator. A multi-tenant agent platform may execute payments on behalf of dozens of underlying businesses in a single TCP session. The "seller" on the receipt is not necessarily the registered taxpayer.

Aggregation does not match the cadence. x402 invoices settle in seconds. Daily-close reconciliation is a tax-engineering choice, not a feature of the rails. If a state requires destination-based determination at the moment of sale, a batch process is the wrong primitive.

Taxability is decided at machine latency. A human tax analyst cannot review every taxability call when the call is happening 400 times per day per agent. The taxability rule needs to be encoded, versioned, and queryable inline.

The unit of output is a real-time decision, not a quarterly return. Returns still get filed — but the upstream determination has to be correct at the moment of payment, or the data feeding the return is already wrong.

This is not a criticism of Zamp's roadmap. It is a description of two different problems that share the words "sales tax." One is the operating system for a company's compliance lifecycle. The other is a tax-determination service that runs at the same latency and identity layer as the payment rail itself.

Why the Distinction Matters Now

Two trends make this distinction load-bearing rather than academic.

First, the AI tax-tooling market is consolidating around the compliance-workflow framing. Avalara's Avi agent, announced at NEXT in March and re-emphasized for the September CRUSH conference, plays in the same lane: an embedded AI workforce that automates the existing tax-compliance pipeline. Zamp is doing the same with a managed-services twist. KPMG's GenAI Tax Stack, which we covered on March 19, is the same pattern at the Big Four scale. None of these products is wrong. All of them assume a recognizable taxpayer with a quarterly cadence.

Second, the substrate underneath them is changing. The x402 protocol, launched by Coinbase in May 2025 and now embedded in payment rails like Visa's Agent Card and the Coinbase Developer Platform's Agent Kit, settles individual invoices between machines in seconds. Bedrock AgentCore Payments and Stripe's agent-payment SDK move in the same direction. The transactions these systems generate are not simply "more transactions" — they are transactions where the buyer is software, the seller may also be software, and the tax determination has to happen inline or it does not happen at all.

If you are a CPA or controller picking a sales-tax operating system today, Zamp is on a real shortlist. If you are a builder shipping an agent that transacts on behalf of users, you need a different layer — one that lives next to the API call, not next to the close process.

Practical Implications for AI Agent Operators

For teams building agent-driven products:

Map your transaction layer before you pick a tax tool. If your agent makes per-call payments through x402 or a similar rail, a quarterly-close compliance product cannot determine tax for those payments. It can only summarize them after the fact.

Identify the registered taxpayer. Whoever's EIN and state registration are on the return is the taxpayer. If your platform is the registered seller, your agents' transactions are your transactions. If you intermediate without registering, marketplace facilitator analysis applies and the rules vary by state — see our 50-state AI agent sales tax guide for jurisdiction-by-jurisdiction posture.

Pick a tax-determination API, not just a returns-filing platform. These are different products. AgentTax provides the determination layer that runs at the moment of the agent transaction. Zamp, Avalara, and Vertex provide the returns-filing layer that consumes determination output. Both are necessary; they are not substitutes.

Watch the consolidation. Zamp's Thomson Reuters Ventures investment is a signal that the compliance-workflow market is being capitalized for scale. The agent-transaction layer is earlier and more open. That window is finite.

What to Watch Next

Zamp has signaled international expansion. Watch how their identity-attribution model handles VAT, GST, and digital-services-tax regimes that are more aggressive about real-time invoicing than the U.S. system. Several jurisdictions — Mexico's SAT, Brazil's NF-e, India's e-invoicing portal — already require sub-second invoice clearance. Those are the regimes where the compliance-workflow framing and the transaction-layer framing collide first. The architecture choices Zamp makes there will preview how the U.S. market eventually handles agent-driven commerce.

For now: if you are a tax professional, take the Zamp launch seriously. If you are building an AI agent that transacts, recognize that this is not your tooling — and plan accordingly. AgentTax is built for the layer Zamp leaves unaddressed. Try the API playground or read the developer guide for an end-to-end example.

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