The Only SaaS Moats Left in 2026

Executive Summary

Rob Walling argues that AI is not making every SaaS company obsolete, but it is weakening many traditional defenses—especially feature advantages, custom code, integrations, and switching friction. AI lowers the cost and time required to build competing products, while also changing whether customers can afford software or still need it.

Walling estimates that 10–20% of SaaS companies may decline or disappear as these pressures accumulate. His recommendation is to identify which of eight more durable moats a company already has—or could deliberately build—and use them to strengthen its long-term resilience.

Ten Ways AI Can Pressure SaaS Businesses

Walling groups the risks into supply-side and demand-side shocks.

Supply-Side Pressures

1. Customers build their own alternatives. They may not need to reproduce all 50 features in a product—only the three they actually use.
2. Competitors reach feature parity faster. AI makes it easier for new products to match existing functionality, reducing the time that features provide an advantage.
3. Open-source alternatives improve. AI speeds development and makes deployment more accessible to nontechnical buyers.
4. Platforms add competing features. A CRM or accounting platform may build an add-on feature itself, using its existing customer base and distribution to displace a specialist.
5. Agencies offer customized software at SaaS-like prices. Faster AI-assisted development can make bespoke solutions economically viable, often with ongoing support.
6. AI platforms absorb standalone products. Features such as generating ad copy or images may become built-in capabilities rather than separate paid apps.

Demand-Side Pressures

1. Customers’ businesses become less viable. Industries exposed to automation—such as some translation, transcription, and copywriting agencies—may lose revenue and stop buying software. Walling suggests some especially exposed agency categories could see 80–90% of firms fail; this is an illustrative estimate, not a general forecast for all agencies.
2. Customer headcount shrinks. If a customer’s workforce contracts, its need for per-seat software can fall. Walling notes that some customer-support work is 80–90% automatable, potentially reducing a team of six to eight support staff to one or two people supervising AI agents.
3. The underlying activity disappears or changes. For example, software managing take-home programming tests may become less useful if hiring shifts toward pair-programming interviews.
4. Customers buy the automated outcome instead of the tool. A buyer may prefer an autonomous service that replaces a role over software that helps a person do that role.

Traditional SaaS Moats Under Pressure

The presentation reviews familiar defenses: unique features, brand, owned traffic such as SEO, proprietary code or algorithms, UX familiarity, integrations, switching costs, coordination among users, and being the system of record.

AI does not make all these defenses vanish, but it weakens several of them. Features and integrations are faster to reproduce; AI can help extract data even when an app lacks a clean export; and familiar UX matters less if work shifts from people to agents. Walling sees brand, coordination, and system-of-record status as more durable, while emphasizing that moats are matters of degree—not simply present or absent.

Eight Moats That May Remain Durable

1. Buyers know and trust you. Brand, customer understanding, partner recommendations, and an internal champion can make it difficult for an unknown competitor to displace a product—even with better software.

Examples: QuickBooks benefits from accountants’ familiarity and recommendations; Shopify has a large partner and agency ecosystem. Diagnostic: Who has built a career or business around recommending or protecting your product?

1. You are the system of record and part of daily operations. A product becomes harder to replace when it holds important business data and workflows, including statuses, approvals, roles, rules, history, and connected systems.

Examples: Jira’s value extends beyond ticket tracking because teams, audit history, and integrations are embedded in it. Gymdesk expanded from gym-management software into billing, payments, and access control. Diagnostic: If a customer left, what data, history, or workflow would be difficult to move or recreate?

1. Your product can act, not just advise. Tools that execute transactions or intervene in workflows can be more defensible than products that only report or recommend.

Examples: Stripe Connect transfers funds; Reimbi handles candidate-expense reimbursements; Churnkey can intervene in a cancellation flow with retention offers. Diagnostic: Does your software take an authorized action, or only provide information for someone else to act on?

1. Customers trust your results enough not to check them. This is a high bar: customers rely on the provider’s work without routinely verifying it themselves. Trust accumulated over time can make a replacement risky.

Examples: Gusto for payroll and tax documents; Avalara for sales-tax calculations. Diagnostic: Do customers independently verify your output, or do they rely on it as correct?

1. You have approvals, certifications, or regulatory expertise that take time to earn. The moat is not merely serving regulated customers; it is having qualifications or operational capabilities that competitors must acquire.

Examples: SOC 2, HIPAA, and ISO 27001 requirements; Vanta’s compliance-related business; the operational and legal burdens of US SMS, including registration and consent rules. Diagnostic: What approvals, licenses, or regulatory hurdles would a new entrant have to overcome?

1. Replacing you requires a coordinated group to move. Network effects and group coordination make switching harder when many users, teams, or external participants depend on the same system.

Examples: Slack; Commit Swimming, where teams, coaches, families, and parents rely on shared schedules and notifications. Diagnostic: Would a competitor need to move one user—or an entire team, company, or community at once?

1. Each customer makes the product smarter for other customers. A defensible learning loop turns customer activity into improvements that benefit the wider customer base. It requires ongoing learning, not just accumulating data.

Examples: Stripe Radar learns from fraud patterns across merchants; Mailchimp’s Omnivore system learned to detect spam; Ramp can use customer corrections to improve expense categorization. Diagnostic: Does one customer’s experience improve the product for others, and can customers take their data without taking the insights learned from it?

1. You handle an operational burden others do not want to maintain. Software may be relatively easy to build but difficult to keep reliable because of infrastructure, changing APIs, abuse, or ongoing operational work.

Examples: ScrapingBee manages the infrastructure and changing IPs needed to reduce blocks when scraping websites; Cloudflare operates complex internet infrastructure; email providers such as Drip had to dedicate substantial effort to spam prevention. Diagnostic: What do you support or maintain—even if you dislike doing it—that a customer or competitor would find painful to take over?

Metrics and Business Implications

• 10–20%: Walling’s estimate of SaaS businesses that may decline or disappear amid the combined AI pressures.
• 80–90%: His estimate of how much of some customer-support work can be automated; he illustrates the impact with a team shrinking from six to eight support staff to one or two AI supervisors.
• 80–90%: An illustrative estimate for failure among certain exposed agency categories, such as some translation or transcription businesses—not a prediction for all agencies.
• $30,000–$50,000 and roughly three months: Example of traditional agency pricing and delivery time for custom software. AI may reduce development time to roughly one-third, one-quarter, or one-fifth of the original.
• $99 versus $299–$999 per month: Illustrative pricing comparison between an all-in-one SaaS product and a more customized, built-and-supported agency solution.
• $50 versus $300–$500 per month: Hypothetical example of customers paying more for a largely autonomous agent if it replaces the need for an employee.
• 70%: Walling’s estimate of the share of acquisitions in this SaaS segment made by private equity.
• $2 million–$20 million in revenue: The range he gives for companies served by the M&A advisory firm he references.
• Three to five years: The durability horizon buyers may assess when deciding whether a SaaS business can sustain its value.
• Eight figures: ScrapingBee’s reported cash exit, cited as an example of the value of a business built around difficult-to-maintain operations.
• Walling also contrasts public-market examples: he says HubSpot’s stock fell 21% after earnings while Atlassian’s MCP usage doubled and the company grew 35%. These are presented as examples of market reactions, not as a general performance forecast.

Actionable Recommendations

• Audit the business against all 10 AI attack vectors: consider both how AI could increase the supply of substitutes and how it could reduce customers’ ability or need to buy.
• Score existing defenses on a spectrum rather than labeling them simply “moat” or “no moat.”
• Use the eight diagnostic questions above to identify the strongest current moat and the most achievable next one.
• Move beyond a point solution where practical: for example, evolve a vertical tool into an ecosystem, a reporting product into an action-taking product, or a single-user tool into a coordinated network.
• Treat operational pain as a possible advantage. Reliability, compliance, fraud prevention, deliverability, and infrastructure may be difficult to copy precisely because they require continuous work.
• Build learning loops only when customer activity creates durable, shared product improvements—and keep investing so the system continues to learn.
• Plan for AI-driven changes in customer headcount and business viability, not just the possibility that customers will recreate your software.

Presenters and Sources

The talk is presented by Rob Walling and was co-written with Interval Set, which Walling says reviewed 89 papers from think tanks, private-equity groups, Gartner, and other sources, and used Claude and ChatGPT during its analysis.

The presentation also draws on an M&A perspective from Einar at Discretion Capital and cites examples from companies including QuickBooks, Shopify, Jira, Gymdesk, Stripe, Reimbi, Churnkey, Gusto, Avalara, Vanta, Slack, Commit Swimming, Stripe Radar, Ramp, Mailchimp, ScrapingBee, Cloudflare, and Drip.