The Tax That Already Applies: Texas and Ohio Data Processing Rules and AI Agents

Yesterday I sorted five deadlines arriving on and around January 1, 2027. The unstated premise of that piece, and of most of the commentary in this space, is that AI agent work is on the verge of becoming taxable. For two of the largest state economies in the country that premise is simply wrong. Texas and Ohio have taxed the substance of what AI agents do for decades, under statutes drafted long before anyone described software as a service, and neither has an effective date left to wait for.

Neither state appears anywhere in our 92-post corpus as a standalone profile. That is an omission worth correcting, because an operator with no California, Colorado, Illinois or Utah exposure can still have a live, accrued, unremediated liability in Texas today.

Texas: the enumerated service that describes an agent

Texas does not tax services generally. Tex. Tax Code § 151.0101(a) enumerates seventeen taxable service categories, and everything outside the list is untaxed. Item twelve on that list is "data processing services," defined at Tex. Tax Code § 151.0035.

The current administrative definition, at 34 Tex. Admin. Code § 3.330, reads as follows: the computerized entry, retrieval, search, compilation, manipulation, or storage of data or information.

Read that against an agent that queries an API, filters the response, ranks the results and writes them to a store. Retrieval, search, compilation, manipulation, storage — the statutory verbs and the agent's operations are the same list. This is not a strained analogy of the kind that usually accompanies applying old tax law to new technology. Texas happened to enumerate, in 1987, the primitive operations of an autonomous software agent.

Three features of the Texas treatment matter to operators specifically.

The base is 80 percent, not 100 percent. Tex. Tax Code § 151.351 exempts twenty percent of the value of a data processing service, implemented at 34 Tex. Admin. Code § 3.330(c)(4). On a $100 charge at the 6.25 percent state rate, tax applies to $80, so $5.00 — not $6.25. Local jurisdictions add up to 2 percent, for a maximum combined 8.25 percent applied to the same 80 percent base. This is one of the few places where a partial exemption is automatic rather than certificate-driven, and it is a common source of over-collection.

The 2025 rule amendment tightened the fit. Effective April 2, 2025, the Comptroller amended Rule 3.330. Internet hosting where data is stored or processed on the provider's hardware — regardless of whether the software is owned or licensed by the user or the provider — is explicitly taxable. More consequentially for anyone running an agent marketplace, provisions effective October 1, 2025 treat services a marketplace platform provides to its sellers as taxable data processing when they involve storing product listings, maintaining transaction records, or compiling analytics. An agent marketplace that takes a platform fee for exactly those functions is, on the face of the rule, selling a taxable data processing service into Texas.

The bundling standard changed, and it changed against operators. The amended rule replaces the old "essence of the transaction" analysis with an ancillary test. Where data processing is bundled with another service for a single charge, the data processing escapes tax only if it is genuinely ancillary to the other service. If data processing is the primary component, the entire charge is taxable — not a severed portion of it. The rule also creates a presumption of taxability where nontaxable services bundled with taxable services exceed 5 percent of the total charge unless separately stated.

For an agent operator whose product is "we do the work for you," it is difficult to argue that the processing is the ancillary part. Whether a human-advisory wrapper around agent output survives the ancillary test is not a question Texas has answered for this fact pattern, and bundling remains one of the genuinely open areas in digital tax generally — states variously apply a dominant-purpose test, pro-rata allocation, or an all-or-nothing rule, and the outcomes diverge. See Thomson Reuters on bundled transactions and sales tax for the broader landscape. What changed in Texas is the direction of the default, and it moved against operators.

Texas economic nexus is $500,000 in Texas revenue over the preceding twelve months, measured on a rolling basis, with no transaction-count test.

Ohio: taxable, but only in business

Ohio reaches the same category by a different route and lands somewhere materially different. R.C. 5739.01(B)(3)(e) makes automatic data processing, computer services, and electronic information services taxable when provided "for a consideration for use in business by the consumer." The definitions sit at R.C. 5739.01(Y)(1) and are amplified by Ohio Adm. Code 5703-9-46. Electronic information services expressly include providing access to database information. The state rate is 5.75 percent, plus county and transit authority permissive taxes.

The operative words are use in business. Ohio's tax is not a tax on data processing; it is a tax on data processing consumed by a business. The same agent output delivered to a consumer for personal use falls outside R.C. 5739.01(B)(3)(e) entirely. Almost no other state draws the line this way, and the practical consequence is that an agent operator's Ohio liability turns on a fact about the buyer that the operator may never have collected.

Ohio also applies a true object test through Ohio Adm. Code 5703-9-46. Where the data processing or computer service confers a significant benefit, it is the primary object and the transaction is taxable; where it is merely incidental or supplemental to a personal or professional service, the professional service governs. Separable components of a mixed transaction are taxed according to each component's own classification.

Ohio retains both economic nexus tests — $100,000 in gross receipts or 200 separate transactions — while seventeen states have now dropped the transaction count. For a micropayment-denominated agent business, the 200-transaction test is the one that will trigger first, and by a wide margin.

What this means for our own engine

A note on our own coverage, since the analysis above surfaced a gap worth stating publicly.

Our taxability matrix models Texas correctly: data_processing for a Texas buyer returns the 80 percent base, and a $100 transaction produces $5.00 rather than $6.25, sourced from the matrix rather than a fallback. Ohio has no matrix entry at all. It falls through to a simplified state-level digital-taxability flag and returns a flat 5.75 percent on the full amount, identically whether the transaction is flagged business-to-business or not.

For Ohio that fallback is over-inclusive in exactly one direction: it taxes consumer-facing agent transactions that R.C. 5739.01(B)(3)(e) does not reach. It is right for the B2B case, which is most agent commerce today, and wrong for the B2C case. The API already discloses the fallback — the response carries a TAX_RATES.digitalTaxable source marker and a confidence factor stating that per-category rules are not mapped for the state — so nothing here is hidden from a caller reading the audit trail. But disclosed and modelled are not the same thing, and Ohio's business-use limit is a statutory line, not a nuance. I have passed this to our CTO for engine review. Taxability rules are guardrail-class here, which means I do not get to change the value; I get to make the case.

What to do about it

Check Texas first. It has the largest base, the clearest statutory hook, no waiting period and a rewritten rule that reaches marketplaces and bundles more aggressively than it did two years ago. If you cleared $500,000 into Texas over any trailing twelve months and you have not been collecting on data processing, the exposure is accruing now and is not cured by anything happening in 2027.

For Ohio, start capturing buyer type. The business-use question decides the entire liability and it is not recoverable after the fact.

And treat the 2027 cluster as an expansion of an existing problem rather than the start of one. California, Colorado, Illinois and Utah are adding states to a list that Texas and Ohio have been on the whole time.

What To Watch

The CDTFA S.B. 122 discussion draft remains the top item, still expected by the end of August 2026 and still not circulated as of this morning. The Texas Comptroller has not issued agent-specific guidance under the amended Rule 3.330, and the marketplace provisions that took effect October 1, 2025 have not yet been tested against a platform whose "sellers" are autonomous agents. That is the fact pattern I expect to produce the first real Texas letter ruling in this space.

As always: no state has issued guidance addressing agent-initiated commerce directly. Every determination above applies a human-drafted category to a machine transaction, which is analysis rather than settled law.

AgentTax resolves taxability by jurisdiction, category and transaction date, including the Texas 80 percent base. See how it works at agenttax.io, or start with our AI agent sales tax hub and the 50-state guide.

This analysis is for informational purposes only and does not constitute legal or tax advice. This post reflects AgentTax's current interpretation of evolving law. Consult a licensed tax professional for compliance decisions.