SaaS Risk Factors and AI Agents: 8 of 30 10-Ks Now Admit the Threat

We read the most recent 10-K of 30 public SaaS companies and searched every 10-K and 10-Q they have filed for agent language. Only 8 of the 30 name AI agents or autonomous AI as a competitive or demand risk in their risk factors. Yet 26 of the 30 now use the words somewhere. The gap between those two numbers is where SaaS risk factors on AI get interesting.

The short answer: 8 of 30 SaaS 10-Ks name AI agents as a risk

Which SaaS companies admit in their 10-K that AI agents threaten their business? In our sample of 30 filers, 8 do it explicitly in Item 1A. Another 5 describe AI replacing their product or shrinking demand without using the word agent. The remaining 17 mention AI only as general competition, a security issue or a regulatory risk.

Item 1A is the risk factors section of an annual report. It is where a company must set out the material risks to an investment, in writing, under liability. Since the SEC's 2020 amendments to Regulation S-K, the test for inclusion is materiality, and any section running past 15 pages needs a summary of up to two pages, according to SEC Release 33-10825.

That legal setting is why the count matters. An earnings call is marketing with a microphone. A risk factor is what the lawyers would point to if the risk came true and shareholders sued. Naming agents there means the company judged the risk worth admitting.

Our count is also in line with a wider tally. A March 2026 summary relayed an analysis by The Information, which used the research platform AlphaSense. It reported 27 software companies listing AI agents as a competitive risk in 2026 filings, against 7 in the same period a year earlier. We could not open the original, so treat that figure as directional. The direction matches what the EDGAR record shows.

How the 10-K risk factor count was built

Every number in this post can be reproduced from public data in about an hour. A ledger of who admits what is only useful if you can check it yourself and update it next quarter.

The 30 filers and the three phrases

We picked 30 US-listed SaaS companies that file 10-Ks with the SEC, chosen to cover the main seat-priced, usage-priced and platform categories. Foreign private issuers filing 20-Fs, such as Monday.com, are excluded because their risk disclosures follow a different form. Companies taken private before 2026 are excluded too.

For each filer we ran three exact-phrase queries against EDGAR full-text search, scoped by company CIK and limited to 10-K and 10-Q forms. The phrases were "AI agents", "agentic" and "autonomous agents". For each phrase we recorded the total number of matching filings and the earliest filing date. The example link runs the first query for Salesforce.

What counts as an admission

Phrase hits show the vocabulary is present, not where it sits. So for each company we opened the most recent 10-K and read the competition and technology risk factors in Item 1A, recording the closest sentence about agents, AI substitution or customers building their own tools.

A filer counts among the 8 only if Item 1A names agents or autonomous AI as a threat to demand, competition or pricing. Describing your own agent products as a liability risk does not count. Neither does a sentence in Item 1, the business description, however bold.

The ledger: 30 SaaS filers and their AI agent risk language

The table below is the whole dataset. "First agent phrase" is the earliest 10-K or 10-Q matching any of the three phrases, anywhere in the document. "Item 1A verdict" is our reading of the latest annual report's risk factors.

Dates are EDGAR filing dates. Fiscal years differ: Adobe's 10-K covers the year to 28 November 2025, Intuit's the year to 31 July 2026. ZoomInfo writes "AI-powered agents" and Paylocity writes "purpose-built agents", which is why an exact-phrase search misses both.

The 8 SaaS filers that name AI agents as a competitive risk

The eight admissions split into three distinct mechanisms, and the mechanism says more about a vendor's exposure than the admission itself.

Seat compression, in ZoomInfo's own words

ZoomInfo's 10-K is the most direct filing in the sample. It says that "as AI-powered agents become more sophisticated, individual users may become substantially more productive, enabling customers to achieve similar outcomes with fewer seats accessing our platform". The next sentence names the mechanism outright: "This AI-driven seat compression could result in customers reducing seat counts at renewal".

That is the scenario this site has been modelling in the analysis of seat compression in SaaS pricing, now written into a filing. The same 10-K adds that some customers may "rely on internal and homegrown databases" instead. Ironically, ZoomInfo does not register in an exact-phrase search at all, because it writes "AI-powered agents".

Customers building their own agents: DocuSign and Atlassian

DocuSign's risk factor heading warns that "Rapid and unpredictable advances in AI and other technologies could reduce demand for our solutions." The body is blunter. Advances in AI may let companies "quickly and cheaply create agents or other homegrown alternatives that perform some or all of the functions they currently obtain from Docusign".

Atlassian, in its August 2026 10-K, describes AI-native competitors offering "highly specialized, autonomous, or automated solutions that may bypass traditional business process workflows or displace established user interfaces". It also says some customers "have elected, and more may in the future elect, to develop or acquire their own internal collaboration and productivity software tools", naming AI-assisted development tools as the route.

Both filers are describing the build-it-yourself threat, not a rival vendor. In our view this is the more dangerous version for a seat-priced business. A competitor can be out-sold. A customer's internal tool has no sales cycle to lose.

Competitor agents and pricing: Adobe, Klaviyo, Okta, HubSpot and Workday

Adobe's 10-K for fiscal 2025 says it faces "increasing competition from companies offering generative and agentic AI solutions", a list that ends with "purpose-built AI agents". Klaviyo's 2025 10-K states that "generative AI and autonomous AI agents" are "rapidly changing the competitive dynamics of our industry".

Okta's risk is about price rather than product. Its fiscal 2026 10-K warns that if competitors "adopt preferable pricing models for evolving technologies, such as AI agents, we may be unable to attract new customers or retain existing customers based on our historical pricing". That is a pricing admission, and it is rarer than it should be.

HubSpot and Workday share something stranger: the same sentence. Both 10-Ks warn of "greater competition from non-specialist solutions relying on generic large language models", generative AI "and general-purpose agents to address a broad range of business needs". Both also name software built through natural language prompts, "referred to as vibe coding". HubSpot filed first, on 11 February 2026. Workday filed on 6 March.

Shared risk language is common, since risk factors are often drafted from templates. But it should lower your confidence that either sentence reflects a company-specific assessment. Workday's choice to hedge with credits is covered in the breakdown of Workday Flex Credits as a seat hedge.

5 more SaaS filers admit AI substitution without saying agent

Five filers describe AI replacing their product or shrinking their revenue base without naming agents. We count them separately because the vocabulary differs, but a buyer should read them as the same risk.

Five9's is the clearest. A risk factor heading in its 2025 10-K says "AI solutions will likely perform an increasing proportion of contact center interactions". It warns that if Five9 is "unable to replace decreases in subscription revenue from licenses with revenue from the sale of additional AI solutions", its "revenue, results of operations and business will be harmed". It adds that "Some customers may also use AI solutions offered by other companies". Elsewhere it flags a possible market shift "from agent-based pricing to interaction-based pricing", where agent means a human contact-centre seat.

Dropbox's 2025 10-K warns of customers "reducing their number of seat licenses over time, including as a result of increased adoption of AI tools". Snowflake says "companies may use AI to develop their own software, reducing their need to purchase third-party solutions". Appian worries that "AI technologies for code generation or application development" could "reduce the demand for our process automation platform".

ServiceNow rounds out the group with "Cloud-based and AI native vendors may build more business applications or AI powered automation solutions that compete with our products and services". That is a competition risk rather than a substitution confession, which is why it sits at the edge of this group.

Put the eight and the five together and 13 of 30 filers, or 43%, now concede in Item 1A that AI could replace their product or shrink the base they bill. Which vendors carry the most revenue exposure to that outcome is mapped in the scorecard of SaaS revenue billed per seat.

When AI agent language entered SaaS filings

The phrase search dates when agent vocabulary arrived, filer by filer, though not the risk factor itself. The chart below plots the cumulative number of filers using any of the three phrases.

Two filers used agent vocabulary in 2024, both in 10-Qs filed on 4 December: Salesforce and Box. Eleven more arrived in the first quarter of 2025, almost all in annual reports. By the end of 2025 the count was 18. The 2026 annual report season added 6 more, and two 10-Qs brought the total to 26 by September 2026.

The filers that moved from nothing to explicit risk

The more telling movement is inside the 8 admitters. HubSpot's first use of any of the three phrases is its 10-K filed 11 February 2026, and that filing already names general-purpose agents as a competitive threat. DocuSign's first use is its 10-K of 18 March 2026, which arrives with the homegrown-agents warning fully formed. Klaviyo's first "autonomous agents" hit is its 10-K of 10 February 2026.

The shift matches the timing of the software selloff examined in the explainer on the SaaSpocalypse selloff. We cannot say from filings alone whether the market moved the lawyers or the lawyers saw what the market saw. Either way, the language changed within weeks of the share prices.

Agent sellers describe agents in Item 1, not in their risk factors

Salesforce and ServiceNow were among the first in the sample to use agent language, from 4 December 2024 and 30 January 2025. Neither lands among the 8. Their 10-Ks mostly discuss agents in Item 1, the business description, as products to sell.

ServiceNow's Item 1 says customers can "create custom AI agents on our Platform using natural language". Salesforce's Item 1A does mention agentic AI, but as a tool used by threat actors and as a monetisation question. It notes that "the markets and monetization strategies for certain offerings, including Agentforce and Data 360, remain relatively new and uncertain". GitLab's Item 1A follows the same pattern: agents appear as its own product carrying "regulatory, litigation, ethical, reputational, and financial risks".

We think this split is the most useful thing the ledger shows. Companies that sell agents describe agents as an opportunity and a liability. Companies exposed to agents describe them as a competitor. When a vendor's filings only ever treat agents as its own feature, it has not yet told investors what happens if the customer's agent replaces the vendor's seat.

That is not proof of concealment. A platform vendor may genuinely believe agents increase demand for its system of record. But a buyer evaluating a multi-year contract should notice which story the vendor tells under liability, and which it tells on stage. That gap recurs in the count of AI mentions on S&P 500 earnings calls.

Where this SaaS risk factor count is weakest

This ledger has real limits, and some of them cut against the headline.

Phrase searches both miss and overcount

Exact-phrase search missed ZoomInfo, which has the sharpest agent risk in the sample, because it writes "AI-powered agents." It missed Paylocity's "purpose-built agents" for the same reason. In the other direction, Five9's filings use "agents" for human contact-centre staff, which a looser search would count as AI language. The 26-of-30 figure is a floor for vocabulary, not a measure of concern.

Our reading of Item 1A was not exhaustive for every filer

Risk factor sections in this sample run to dozens of pages. For the 13 filers classified as admitting a risk, we confirmed the quoted sentences verbatim in the filing text. For several of the 17 marked generic, our extraction covered the competition and technology risk factors but not every page of Item 1A. A sentence we missed would move a company up a row. It would not change the phrase counts or the dates.

Risk factors are drafted by lawyers, not strategists

The strongest counter-argument is that risk factors are defensive by design. Lawyers add remote risks to protect against future litigation, so an admission may signal caution rather than belief. The HubSpot and Workday shared sentence supports that reading. On this view, the 8 are simply the most cautious legal teams, and the 17 generic filers are no safer.

We think that argument is half right. Risk factors overstate remote risks in general. But no lawyer adds "AI-driven seat compression" to a 10-K without a business reason, and three filers tying AI to their own seat or licence counts is a specific, falsifiable statement. Thirty filers is also a sample, not the market. A different 30 could produce a different count, and we would publish that too.

Frequently asked questions

How to use this ledger before your next renewal

Pick your three largest SaaS contracts and search each vendor's latest 10-K for "agent", "seat" and "build their own". It takes ten minutes per vendor on EDGAR. If the vendor admits in writing that AI could cut your seat count, you have a negotiating position: ask for usage-based or outcome-based terms at renewal, or a seat-reduction right mid-term.

If the vendor's filing only ever describes agents as its own feature, ask the account team the question the 10-K avoids. What happens to your price when your own agents do the work your seats do today? Write their answer into the contract, not the call notes. Then check next year's 10-K to see whether the language moved.

Related reading