Paper
Is AI coding affecting SaaS yet?
SoftHouse Advisory research. Published 26 October 2026. Public sources checked live; evidence available by this date only.
Download the PDFBottom line
Yes. The clearest direct effect is that some software purchases are being avoided and some individual tools replaced by internal builds. The evidence does not establish an industry-wide revenue collapse. The more defensible description is selective substitution, more competition, and spending moving toward AI services and the platforms used to build custom software.
Three phenomena need separating: AI helps build software; buyers build instead of buying; and AI-powered products compete with existing SaaS. They can reinforce each other, but evidence of the third is not proof of the first causing cancellations.
Direct evidence of building instead of buying
McKinsey, published 25 August 2026. Its May/June survey of 1,719 respondents across 97 countries found 32% reporting that their organization had declined at least one software product or feature because it could build the functionality with agentic coding tools. This is direct evidence of forgone purchases, not 32% of spending disappearing, nor proof that all the resulting builds work well. It is self-reported rather than an audit of cancelled contracts. Survey and methodology.
Retool, published 17 February 2026, surveyed in late 2025. Among 817 customers and builders, 35% reported replacing at least one SaaS tool with a custom build. Workflow automation and internal administration led the replacement categories. This is a relevant but selected sample: Retool sells the alternative to buying packaged tools, and its audience is unusually likely to build. Do not generalize 35% to all enterprises. Report announcement.
Stack Overflow, 2026 developer survey. Its underlying table records 18.6% building replacements for paid vendors; 54.4% of that replacement cohort used AI. The denominator matters: it does not mean 54.4% of all developers or all AI users are replacing SaaS. Scripts and internal tools were substantially more common. The page contains a confusing prose sentence, so this brief uses the explicit tables and accompanying cohort explanation. Technology results.
These samples measure different things, at different times, in different populations. They should not be averaged into a market-wide replacement rate.
Named customer cases
Harmonic: a Retool customer story describes replacing a third-party tool costing more than USD20,000 annually, and operating 33 internal applications. It still connects those applications to services including Salesforce, Gong and Slack. This is selective replacement around an existing stack, not complete software independence. The avoided subscription is not an audited net saving after labor, platform fees and maintenance. Customer case.
Nursa:Lovable's Series C funding announcement (12 August 2026) says Nursa had deployed Lovable across more than 200 employees and was in the process of retiring ten SaaS systems. That is a named AI-building example, but a vendor-reported transition, not evidence that ten cancellations had completed or that lifetime savings were positive. Lovable announcement.
What the spending and incumbent evidence says
Zylo's 2026 index:average annual SaaS spend in its dataset increased 8%, while average portfolios were essentially flat at 305 applications. AI-native application spending rose 108%. This suggests money is moving and pricing is changing, not simply leaving software. Zylo's managed enterprise sample is not the entire market, and the index should not be treated as a live economic census. Index and dataset description.
Ramp's transaction-based analysis: in business intelligence, newer vendors gained about five percentage points in average adoption while incumbents lost 1.3 points. Yet incumbents retained nearly 83% of category spend. Adoption and dollars can move at different speeds. These are Ramp-network results about AI-native competition, not a causal measurement of coding agents replacing subscriptions. Vendor adoption and spending analysis.
Salesforce, quarter ended 31 July 2026: subscription and support revenue was USD10.8 billion, up 12% year over year, including USD440 million from Informatica. These are reported financial results, not a forecast. Acquisition contributions mean the headline growth is not purely organic. They do not show immunity to future pressure, but contradict a blanket claim that major SaaS revenue is already collapsing. Q2 FY2027 results.
ServiceNow, quarter ended 30 June 2026: subscription revenue was USD3.877 billion, up 24.5%. It also reports making its Build Agent generally available, including integrations with outside coding environments. Incumbents can absorb AI coding into their own platform rather than merely be replaced by it. Neither the growth figure nor the platform announcement isolates an AI-coding effect. ServiceNow reports its Q3 2026 results after market close on 28 October 2026; this brief reflects the Q2 figures available at publication. Q2 2026 results.
Pricing and delivery are changing too
Intercom combines human-seat pricing with Fin outcome charges. Its help documentation lists USD0.99 for a resolution and several other outcomes, with USD9.99 for a sales qualification. Outcome definitions and pricing. The combined seat-plus-outcome structure is explained alongside current plan pricing. Intercom pricing explained. This is a concrete shift toward charging for work performed, but belongs to the broader AI-agent story, not evidence that vibe coding caused the price model.
In Retool's builder survey, 72% of those who had shipped software used AI for code pieces integrated into larger projects; 31% used prompting for complete applications. These uses can overlap. Much of the practical change is AI-assisted engineering, not unaudited one-prompt software. Detailed builder report.
AI coding: chronology and measurement
AI-assisted coding predates the vibe-coding label. GitHub Copilot entered technical preview on 29 June 2021. GitHub's announcement. Vercel announced v0 beta, prompt-based UI generation, on 11 October 2023. Vercel's announcement. Andrej Karpathy's first-person 2025 review describes coining the term vibe coding that year. Karpathy's review.
Stack Overflow respondents using or planning to use AI in development rose from 70% in 2023 to 76% in 2024 to 84% in 2025, a 14 percentage-point increase, not a 14% relative increase. These are separate annual samples, not a fixed longitudinal panel, and planned use is not actual adoption. Actual use rose from 44% to 62% between 2023 and 2024, according to the 2024 survey. 2024 survey, 2025 survey.
The 2025 survey's separate vibe-coding question found 72% said it was not part of their professional workflow, with another 5% emphatically rejecting it.
The 2026 survey changed its question to ask about AI tools or agents in respondents' role at work. Coding assistants or coding agents were selected by 65.9% of 17,464 respondents, and 73% of coding-tool users reported daily use. The question and denominators differ from the 2023 to 2025 series, so these 2026 results stand on their own rather than extending that series. 2026 survey data.
No consistent multi-year primary-source series quantifying SaaS revenue lost specifically to vibe coding has been established. These figures describe adoption of development tools, not measured SaaS displacement.
WordPress and Vercel website-share trends
W3Techs measures share of all surveyed websites, not WordPress share among detected content-management systems. January figures are single-day snapshots taken on 1 January, not annual averages.
| Date | WordPress CMS | Vercel hosting |
|---|---|---|
| 2021-01-01 | 39.5% | 0.1% |
| 2022-01-01 | 43.2% | 0.2% |
| 2023-01-01 | 43.1% | 0.3% |
| 2024-01-01 | 43.1% | 0.4% |
| 2025-01-01 | 43.6% | 0.6% |
| 2026-01-01 | 43.0% | 1.0% |
| 2026-10-26 | 39.6% | 2.3% |
WordPress history, Vercel hosting history.
Over the exact five-year interval from January 2021 to January 2026, WordPress gained 3.5 percentage points, about plus 8.9% relative to its 2021 share. Vercel gained 0.9 percentage points over the same span, roughly a tenfold increase from a small base; the multiple is approximate because the published shares are rounded to one decimal place. The October endpoint is a later, separate update, not part of that five-year calculation.
The current WordPress dip is not its first: the annual table already falls from 43.2% in January 2022 to 43.1% in January 2023, before climbing back to 43.6% in January 2025. A decline in share alone does not prove a decline in the absolute number of WordPress sites. CMS and hosting are also different technology layers, not mutually exclusive competitors; WordPress and Vercel can coexist, including through headless setups. Vercel's headless WordPress guide. W3Techs changed its website sample in November 2022. W3Techs sample change. These trends do not demonstrate that sites migrated from WordPress to Vercel, or that vibe coding caused WordPress's decline.
Where the pressure falls
This is our interpretation of the evidence above, not a measured ranking.
| Software proposition | Likely exposure | Reason |
|---|---|---|
| Narrow internal forms, dashboards and approval workflows | Higher | Bounded scope and company-specific requirements make a custom alternative plausible. |
| Premium add-ons for a few missing features | Higher | The buyer may keep the core subscription and build the missing piece instead of upgrading. |
| Broad CRM, finance, identity and compliance systems | More defended, not immune | Recreating screens does not replace operations, migration, permissions, integrations and accountability. |
| App-building platforms, hosting, databases and AI APIs | Potential beneficiaries | The internal replacement usually depends on another paid software stack. |
The first commercial symptom may be a missing new sale or expansion, not a cancellation. A customer can retain a core system while declining its extra modules. Consequently, growing revenue and real competitive pressure can coexist.
What it means for buyers and builders
Our recommendation is to treat this as a selective custom-software opportunity, not a mandate to replace sound systems. Offer a bounded workflow with an explicit owner, integrations, tests, access controls, rollback and ongoing maintenance. Compare total operating cost with the subscription and its implementation cost, not just the time needed to create the first demo.
For SaaS products, a feature list is becoming a less convincing reason to pay. Deep domain knowledge, trusted data, distribution, embedded workflows and operational responsibility are stronger places to invest. AI can also improve an incumbent's delivery speed; lower coding cost does not belong exclusively to new entrants.
The practical question for any buyer is which expensive, ill-fitting part can be safely replaced while retaining the reliable systems underneath. That is a more defensible plan than rebuilding an entire CRM because a prototype looks similar.
Limits
- No defensible industry-wide percentage of SaaS revenue lost specifically to AI coding was established in this research.
- Surveys establish reported behavior, not durable return on investment or causality at the whole-market level.
- Vendor case studies have selection and commercial incentives; they are examples, not prevalence estimates.
- Rising spend can conceal churn, price increases, acquisitions and substitution. It does not disprove displacement of particular vendors.
- Public earnings, survey results and vendor snapshots are dated evidence, not interchangeable measures of the current quarter.
References
Public primary sources, numbered to match the PDF. Checked 15 October 2026; freshness reviewed 26 October 2026.
- 1.The state of AI in 2026: On the road to ROI · McKinsey, 25 Aug 2026
- 2.2026 Build vs. Buy Report announcement · Retool, 17 Feb 2026
- 3.2026 Developer Survey: Technology · Stack Overflow, 2026
- 4.Harmonic customer case · Retool, 2026
- 5.Series C announcement and customer examples · Lovable, 12 Aug 2026
- 6.2026 SaaS Management Index · Zylo, 2026
- 7.AI-native software is quickly climbing the ranks · Ramp, 2026
- 8.Second Quarter Fiscal 2027 Results · Salesforce, 26 Aug 2026
- 9.Second Quarter 2026 Financial Results · ServiceNow, 22 July 2026
- 10.Fin AI Agent outcomes · Intercom, 2026
- 11.The Build vs. Buy Shift · Retool, 2026
- 12.Intercom pricing explained · Intercom, 2026
- 13.Introducing GitHub Copilot · GitHub, 29 June 2021
- 14.Announcing v0: Generative UI · Vercel, 11 Oct 2023
- 15.2025 LLM Year in Review · Andrej Karpathy, 19 Dec 2025
- 16.2024 Developer Survey: AI · Stack Overflow, 2024
- 17.2025 Developer Survey: AI · Stack Overflow, 2025
- 18.2026 Developer Survey: AI data · Stack Overflow, 2026
- 19.CMS usage: yearly historical trends · W3Techs, snapshot 26 Oct 2026
- 20.Web hosting usage: yearly historical trends · W3Techs, snapshot 26 Oct 2026
- 21.Our website sample set has been extended · W3Techs, 14 Nov 2022
- 22.Using Headless WordPress with Next.js and Vercel · Vercel, 2026
- 23.Technology survey methodology · W3Techs, 2026