When AI Agents Buy and Sell: Who Pays the Tax?
Something unprecedented is happening in commerce. AI agents — autonomous software systems acting on behalf of businesses — are buying and selling services from each other at enormous and rapidly growing scale. An agent purchases GPU compute from another agent's endpoint. A coding assistant buys API access from a data provider. A research bot pays for storage from a cloud service. All without a human in the loop.
These aren't hypothetical scenarios. They're happening right now, thousands of times per day, and the market behind them is accelerating fast. The global AI agents market was valued at roughly $7.6 billion in 2025 and is projected to exceed $10.9 billion in 2026, growing at over 45% annually (Grand View Research, 2025). By 2030, that figure could top $50 billion (BCC Research, January 2026).
Every single one of these transactions carries a tax obligation that almost no one is tracking.
The New Reality of Autonomous Commerce
When a human buys software online, the tax flow is well-understood. You visit a website, add something to your cart, and the platform calculates sales tax at checkout based on your location. Stripe or Shopify handles the math. You pay, the seller remits the tax. Done.
Now remove the human entirely. Your AI agent needs 10,000 GPU-hours to train a model. It discovers a compute provider via API, negotiates a price, and initiates payment — all autonomously, in milliseconds. The transaction completes before you've finished your morning coffee.
Who calculated the sales tax? Who's remitting it to the state? Who even knows this transaction occurred?
For the vast majority of agent-to-agent transactions happening today, the answer is: nobody.
The Tax Obligation Doesn't Disappear
Here's the core principle every AI builder needs to internalize: the tax obligation exists whether or not anyone is tracking it.
If your AI agent sells digital services to a buyer in Texas, you owe Texas 6.25% sales tax on that transaction. It doesn't matter that your agent completed the sale autonomously. It doesn't matter that the buyer was also software. The tax code doesn't contain a "robots are exempt" provision.
The obligation falls on the business entity that operates the agent. If you're a startup in California and your agent sells API access to buyers across 30 different states, you potentially have sales tax collection obligations in every state where you've established economic nexus — which, thanks to the 2018 Supreme Court decision in South Dakota v. Wayfair, can be triggered by as little as $100,000 in annual sales into a state, with no physical presence required (Avalara; Sales Tax Institute).
And it gets worse on the buying side. If you're purchasing services and the seller didn't collect sales tax, you may owe use tax — a self-assessed obligation that buyers are responsible for remitting directly to their home state. Most businesses don't know use tax exists. Most AI builders have never heard of it. But the liability accumulates silently with every agent-to-agent purchase.
Why This Is Different From Regular E-Commerce
You might be thinking: isn't this just the same problem Shopify sellers have? Not quite. Agent-to-agent commerce introduces several unique complications:
Volume and velocity. A human might make a few purchases a day. An AI agent can execute thousands of transactions per hour across dozens of jurisdictions. The compliance surface area is orders of magnitude larger.
No checkout flow. Traditional e-commerce has a cart and a checkout page — natural points where tax calculation happens. Agent-to-agent transactions happen via API calls with no built-in tax step. Unless someone deliberately inserts a tax calculation into the agent's workflow, it simply doesn't happen.
Ambiguous classification. When an AI agent buys "compute," is that a digital service? A data processing service? A lease of tangible personal property? The answer determines whether it's taxable, and it varies by state. Texas taxes data processing services at 80% of the sale price. California doesn't tax SaaS at all. The same transaction, taxed completely differently depending on the buyer's location (TaxJar; TaxCloud, January 2026).
The 1099 problem. When your buyer agent pays 50 different vendors in a single day, each payment counts toward the IRS's $600 threshold for 1099-NEC reporting. At agent speed, you can blow past that threshold with a single vendor in hours — and the IRS expects you to track every dollar.
Who's Actually Liable?
Let's be precise about where the legal obligation falls:
For sellers: If you operate an AI agent that sells services, you are the seller for tax purposes. You must determine whether you have nexus in the buyer's state, whether the transaction type is taxable in that state, and what rate applies. You must collect the tax at the point of sale and remit it to the state on the required filing schedule. This is true even if your "storefront" is an API endpoint and your "customer" is another bot.
For buyers: If you operate an AI agent that purchases services, and the seller does not collect sales tax, you owe use tax to your home state on that purchase. You are also responsible for tracking cumulative payments to each vendor for 1099 reporting purposes.
For both: The business entity behind the agent bears all liability. Not the agent. Not the framework. Not the cloud provider. You.
The Scale of the Problem
Consider a realistic scenario: You run a small AI company with three agents. One sells API access to ML models. One buys compute. One buys and resells data. In a single month, these agents might execute 10,000 transactions across 40 states.
That's 40 potential nexus registrations. Thousands of individual tax calculations. Hundreds of vendors to track for 1099 purposes. Multiple filing frequencies — some states want monthly returns, some quarterly, some annually. And every state has its own rules about what's taxable, at what rate, and with what exemptions.
No human can manage this manually. The traditional tax compliance tools — built for retailers selling physical goods through shopping carts — weren't designed for it either. Enterprise solutions like Avalara serve Fortune 500 companies with complex ERP integrations. They're not built for a solo developer with three agents and a Stripe account.
This is exactly the gap that needs filling: tax compliance infrastructure purpose-built for AI agents, accessible via a simple API call that fits naturally into agent-to-agent workflows.
What the Solution Looks Like
The answer isn't to bolt traditional tax software onto agent workflows. It's to build tax compliance into the transaction protocol itself. The same way an agent calls an API to process a payment, it should call an API to calculate its tax obligation — before the payment settles.
That means:
- A single POST endpoint that accepts a transaction and returns the tax obligation
- Automatic nexus determination based on buyer/seller jurisdictions
- Digital goods taxability rules applied per state
- 1099 threshold tracking across all vendor relationships
- Use tax detection when the seller hasn't collected
- Dashboard visibility into all obligations without manual data entry
This is what AgentTax was built to do. One API call per transaction. The agent gets back a tax result, routes the payment accordingly, and moves on. The compliance data lands in a dashboard automatically.
What Happens If You Do Nothing
The states aren't sleeping on this. New York, California, and Michigan are already deploying AI-powered audit systems that cross-reference marketplace data against filed tax returns in real time (OurTaxPartner, 2026). States are actively expanding the definition of taxable digital products — Maine added digital audio and audiovisual services to its taxable categories for 2026, and Illinois eliminated its transaction-count threshold, moving to a pure revenue test (TaxCloud, December 2025).
The enforcement apparatus is getting smarter at exactly the moment the compliance gap is getting wider. If you're running AI agents that transact commercially and you're not tracking tax obligations, you're accumulating liability that compounds with every transaction.
The window for proactive compliance is now. The cost of getting ahead of it is a fraction of the cost of an audit.
Getting Started
If you operate AI agents that buy or sell services, here's what you should do today:
- Identify your transactions. Map every agent-to-agent transaction your systems execute. What's being bought/sold? Where are the buyers located?
- Determine your nexus. If you're selling into states where your annual sales exceed $100,000, you likely have economic nexus and a collection obligation.
- Classify your products. Is what you're selling a digital service, a digital good, or a general service? The classification determines taxability state by state.
- Integrate tax calculation into your agent workflows. Add a tax API call before payment settlement. This is a few lines of code — and it's the difference between compliance and accumulating liability.
- Start tracking vendor payments. If your agents buy from vendors, track cumulative payments for 1099 purposes from day one.
The agent economy is here. The tax obligations are real. The tools to handle them finally exist. The only question is whether you'll get compliant proactively — or wait for a state to tell you how much you owe.
AgentTax is the first tax compliance protocol built for AI-to-AI commerce. Calculate your tax obligation with a single API call: POST /api/v1/calculate. Get your free API key →
Sources:
- Grand View Research, "AI Agents Market Size and Share," 2025
- BCC Research, "AI Agents: Technologies, Applications and Global Markets," January 2026
- South Dakota v. Wayfair, Inc., 585 U.S. ___ (2018)
- Avalara, "Economic Nexus and South Dakota v. Wayfair, Inc."
- Sales Tax Institute, "South Dakota v. Wayfair FAQ"
- TaxJar, "Software as a Service Sales Tax by State," December 2025
- TaxCloud, "SaaS Sales Tax by State," January 2026
- TaxCloud, "Sales Tax Changes 2026," December 2025
- OurTaxPartner, "2026 State Sales Tax Rates: Economic Nexus Rules," February 2026