Is AI eating SaaS? Nine filed lines to read before the October–December prints.

Every figure below was read from the company's own filing on sec.gov (10-Q, 10-K, or for SAP the 20-F and 6-K exhibits) and is cited with its accession. Guidance is each company's own, from its last earnings exhibit, and is labelled as company guidance. We show no analyst estimates and no consensus, because none of it is filed. Growth rates the companies do not print, ratios to free cash flow and 12-month shares are our arithmetic on filed values, marked (Ours.) Report dates are labelled "announced" only where the company itself published the date; the rest are "expected". Research, not investment advice.

The argument on X this autumn is that AI agents are about to replace the software that companies rent by the seat. The counter-argument is that the products are too embedded to rip out. Both sides mostly argue from share prices.

The filings answer a narrower question, and they answer it ten times: is anything in the revenue, the retention metric, the pricing language or the risk factors showing substitution yet? We picked the ten names below by how much they were discussed on X over the seven days to 5 October (cashtag post counts, method at the end), took each one's latest filing, and read the same nine lines in each.

Short version: subscription revenue grew at every one of them. The changes are in the wording. Two companies added the same new sentence about AI-native competitors to their 10-Ks, four rewrote how they describe pricing, and the retention metrics that would show seat cuts first are, in several cases, built not to show them.

1. Revenue and backlog are still growing everywhere

Three numbers in that table say less than they seem to. Salesforce's +12% includes $440 million from Informatica; (Ours: without it, +7.1%.) MD&A adds that the growth was "volume-driven" and "Pricing was not a significant driver". Atlassian's fiscal 2026 includes Data Center revenue its own shareholder letter says was "pull-forward of customer purchasing from future periods" after the end-of-life announcement. Adobe's RPO grew 8%, against 14% for subscription revenue, and fell $0.36 billion from the fiscal year end. (Ours.)

Check: the 12-month backlog line at each print. It moves before revenue does.

2. The retention metric is often built not to catch seat cuts

If AI shrinks seats, a net retention rate would show it before revenue does. Only two of the ten file one. Snowflake's net revenue retention was 126%, against 125% a year ago, with the definition paragraph word-for-word unchanged. Datadog's was "in the low-120%'s" against "about 120%", followed by a new sentence: "we saw a reduction in usage from our largest customer starting in the third quarter of 2026."

The others file something narrower. ServiceNow's renewal rate has been 98% in every period, but its own definition says "a customer that reduces its subscription upon renewal is not considered a lost customer" unless the reduction is "the majority of the customer's ACV". A customer cutting 40% of its seats is invisible in that number. Salesforce's attrition rate is "approximately eight percent" three years running, while the list of exclusions grew each year ("Slack self-service", then "current year acquisitions", now also "Informatica"). Workday files gross revenue retention, which moved: approximately 98% in the FY2025 10-K, approximately 97% since. SAP, Palantir, Adobe, Intuit and Atlassian's 10-K file no retention rate.

3. Large-customer cohorts are growing; small ones slowed

ServiceNow had 658 customers above $5 million of annual contract value, against 533 (Ours: +23.5%). Snowflake had 828 customers above $1 million trailing product revenue, against 654 (Ours: +26.6%). Datadog, about 4,720 customers above $100,000 ARR, up 23%.

The slower lines are at the small end. Atlassian's customers above $10,000 of Cloud ARR grew 10.3% in fiscal 2026 against 13.4% the year before (Ours). Intuit's Online Ecosystem paying customers grew 3%, against 5%, while revenue per customer rose 15%. Growth there is price, not customers.

And one disclosure went away: SAP's 2025 quarterly statements carried the share of cloud orders above €5 million (71% in Q4 2025) and below €1 million. Neither 2026 quarterly statement does.

4. Four companies rewrote how they describe pricing

ServiceNow's FY2024 10-K (0001373715-25-000010) said its digital workflow products "are generally priced on a per user basis". The FY2025 10-K (0001373715-26-000007) drops that sentence and the revenue split that went with it, and adds: "certain AI and data solutions include a consumption-based pricing component that governs when customer usage exceeds the fixed number of service credits".

Datadog replaced "priced per host" with "Usage is measured on a per-unit basis, with the unit of measure differing for each product". Salesforce's latest 10-Q adds "our shift toward consumption-based pricing models" and "AI-driven consumption models" to the attrition risk. Workday's Q2 10-Q, and neither earlier filing, says: "to address the rapidly changing technology market and AI landscape, we are addressing shifts in customer buying behavior and demand through actions such as new pricing models".

The seat line that runs the other way is Atlassian's, identical in both 10-Ks: "The increase in subscription revenues was primarily attributable to paid seat expansion from our existing customers and price increases." That is the clearest filed evidence in the set that seats are not shrinking yet.

5. Headcount: flat or falling at the seat sellers

Prior-year growth rates are ours. Two of those counts are already stale. Intuit's 20 May Item 2.05 8-K (0000896878-26-000024) commits to reduce "its full-time workforce by approximately 17%", substantially complete by 31 October. Workday's 29 September 8-K (0001327811-26-000048) adds "a reduction of approximately 2.5% of Workday's current workforce". Atlassian's restructuring "to accelerate building the future of teamwork in the AI era" affected "approximately 10% of our workforce".

6. The risk factors: the same new sentence, twice

This is where the filings changed most. Salesforce's FY2026 10-K (0001108524-26-000060) and Atlassian's FY2026 10-K (0001650372-26-000036) each add, word for word, competition from:

AI-native companies and emerging startups that leverage generative AI and large language models as the core foundation of their architecture, offering highly specialized, autonomous, or automated solutions that may bypass traditional business process workflows or displace established user interfaces

Neither prior-year 10-K (0001108524-25-000006, 0001650372-25-000036) has it. Atlassian also widened its build-your-own risk from "particularly large organizations" to all customers, "including useing [sic] AI-assisted development tools", and dropped the FY2025 sentence "our revenue is dependent on the number of users of our products."

ServiceNow went from "Competition from cloud-based vendors may increase as they build business applications or AI powered automation solutions" to "Cloud-based and AI native vendors may build more business applications or AI powered automation solutions". Workday added "non-specialist solutions relying on generic large language models … and general-purpose agents" and replaced "we expect this competition to intensify" with "our ability to maintain market differentiation may affect our competitive position". SAP's FY2025 20-F (0001104659-26-020058) inserts six words into its product-strategy risk that the FY2024 20-F (0001104659-25-017815) lacks: customers "may leverage AI to produce their own solutions". Adobe's competition paragraph grew in three steps, to solutions "embedded within or operating across third-party AI platforms" in the Q3 10-Q. Intuit added "providers of general purpose AI technologies" to its competitor list.

Palantir and Snowflake added AI language too, but neither added a sentence describing AI as a substitute for its own product. Datadog's AI risk is the other direction: its "AI-native cohort, which cohort includes our largest customer" contributed "high single digit percentage points" of growth, and that customer cut usage.

7. Buybacks against free cash flow

Free cash flow is each company's own definition where filed. Three bought back more than they generated (Ours): Salesforce 3.6 times, funded by "$25.0 billion aggregate principal amount of unsecured Senior Notes"; Workday 2.7 times; Atlassian 1.36 times. At Snowflake, buybacks plus net-settlement taxes came to $627.7 million against $316.6 million. Snowflake's stock-based compensation was 2.6 times its free cash flow and Atlassian's 1.22 times (Ours). On 2 October Snowflake priced $3.75 billion of 0% convertible notes (0001640147-26-000043).

8. The guidance, as each company filed it

Sources: the last earnings exhibits, ServiceNow 0001373715-26-000072, SAP 0001104659-26-087251, Palantir 0001321655-26-000039, Snowflake 0001640147-26-000033, Adobe 0000796343-26-000147, Salesforce 0001108524-26-000187, Datadog 0001628280-26-053829, Intuit 0000896878-26-000029, Atlassian 0001650372-26-000031, Workday 0001327811-26-000042. Six of the ten guide a slower rate than they just printed on the line they guide: ServiceNow's cRPO (19.5% against 21%), SAP's current cloud backlog, Adobe's ARR, Intuit's revenue, Atlassian's ARR and Workday's subscription revenue (12% against 13.9%). Snowflake guides faster. None of the ten outlooks names AI substitution as a reason; ServiceNow's names AI adoption as a reason for lower gross margin.

9. The one MD&A line on AI, per company

Verbatim, from each latest 10-Q, 10-K or 20-F:

• ServiceNow: revenue "primarily driven by increased purchases by new and existing customers." AI appears on the cost side, in the 8-K: margin guidance "reflects … an acceleration of customer AI adoption".
• SAP: "more than two-thirds of our cloud order entry containing SAP Business AI". Containing, not revenue from.
• Palantir: growth from "increased adoption of our products and services within their organizations". AI is named only in the AIP overview.
• Snowflake: cloud costs "including those related to AI inference and GPUs … mainly as a result of increased customer consumption".
• Adobe: ARR "driven by strength in Creative Cloud flagship apps, Acrobat, and Adobe Experience Manager and agentic web apps, including the contribution from Semrush."
• Salesforce: "strong momentum in Agentforce Apps and Data 360, bolstered by the acquisition of Informatica."
• Datadog: "We saw a reduction in usage from our largest customer starting in the third quarter of 2026, which may cause a deceleration in revenue growth."
• Intuit: QuickBooks Online growth "due to the interrelated factors of higher effective prices, customer growth, and mix shift." No AI attribution.
• Atlassian: gross margin to decline on "their related AI usage". AI as a cost.
• Workday: the new-pricing-models sentence in line 4.

Not one of the ten MD&As attributes a revenue decline to customers replacing the product with AI. Where AI appears, it is either demand, mix, or a cost line.

The calendar

Run it yourself

Pick one company and read lines 2 and 6 yourself; it takes ten minutes. /filings/NOW takes you to ServiceNow's 10-K; search it for "renewal rate" and read the definition through to "majority of the customer's ACV". Then /financials/NOW for the quarterly series, and /earnings/NOW to set the guided $3,975–3,980 million beside the print on the 28th. The same three routes work for every ticker above, and /holders/CRM shows who reported owning Salesforce at 30 June.

/valuation/WDAY lets you run your own growth inputs on Workday's filed numbers.

Then write the thesis down before 21 October. Our Thesis Monitor reads each new filing against it: a renewal rate that leaves 98%, a gross retention rate that leaves 97%, a cRPO line below guide, a new risk-factor sentence. It tells you which moved. The same filings are on the EvidInvest MCP server if you work inside an agent. It is free to start, credit packs from $10, and there is no subscription.

The first thing we will read on 21 October is not SAP's cloud revenue. It is whether the current cloud backlog still grows faster than cloud revenue, and whether the order table comes back.

Sources. ServiceNow: 10-Q 0001373715-26-000076, 8-K Ex 99.1 0001373715-26-000072, 10-K FY2025 0001373715-26-000007, 10-K FY2024 0001373715-25-000010. SAP: 6-K (Quarterly Statement Q2 2026, Half-Year Report 2026) 0001104659-26-087251, 6-K Q4 2025 0001104659-26-009465, 20-F FY2025 0001104659-26-020058, 20-F FY2024 0001104659-25-017815. Palantir: 10-Q 0001321655-26-000041, Ex 99.1 0001321655-26-000039, 10-K 0001321655-26-000011, 0001321655-25-000022. Snowflake: 10-Q 0001640147-26-000037, Ex 99.1 0001640147-26-000033, 10-K 0001640147-26-000008, 0001640147-25-000052, 8-K 0001640147-26-000043. Adobe: 10-Q 0000796343-26-000156, Ex 99.1 0000796343-26-000147, 10-K 0000796343-26-000003, 0000796343-25-000004. Salesforce: 10-Q 0001108524-26-000190, Ex 99.1 0001108524-26-000187, 10-K 0001108524-26-000060, 0001108524-25-000006. Datadog: 10-Q 0001628280-26-054458, Ex 99.1 0001628280-26-053829, 10-K 0001628280-26-008819, 0001561550-25-000025. Intuit: 10-K 0000896878-26-000037, 0000896878-25-000035, Ex 99.1 0000896878-26-000029, 8-K 0000896878-26-000024. Atlassian: 10-K 0001650372-26-000036, 0001650372-25-000036, Ex 99.1/99.2 0001650372-26-000031. Workday: 10-Q 0001327811-26-000044, Ex 99.1 0001327811-26-000042, 10-K 0001327811-26-000014, 0001327811-25-000056, 8-Ks 0001327811-26-000048 and 0001104659-26-112898. Names were selected by X cashtag post counts, seven days to 5 October 2026; $NET excluded as a shared cashtag. No analyst estimate or consensus figure appears on this page. Growth rates, ratios to free cash flow, 12-month shares and ex-acquisition growth are arithmetic on filed values, labelled as ours where they appear. Research, not investment advice.