AI Agents Are Starting to Eat SaaS: A Hacker News Discussion
A vigorous debate erupted on Hacker News about whether AI agents will replace SaaS products. CTOs and developers shared real experiences replacing subscriptions, while others defended traditional software’s staying power.
The Core Tension
The original article claimed AI agents enable customers to build internal alternatives to SaaS products. A CTO running vertical SaaS immediately challenged this: “The threat model assumes customers can build their own tools. Our end users can’t.”
His company serves users whose current system is Excel. Two large enterprises tried cloning his product internally. One gave up. The other’s users called it “crap.” Zero paying subscribers lost.
This reveals the central conflict: coding speed versus domain expertise.
What Actually Gets Replaced
Several concrete examples emerged of cancelled subscriptions:
Retool topped the list. Multiple commenters reported ditching it. One team found writing code with AI cleaner than Retool’s interface. Another spent two hours building static HTML pages that replaced TeamRetro ($240/year).
Niche ERP/CRM systems face pressure. One commenter knew two people actively replacing industry-specific products costing over $100k/year with agent-built alternatives. These weren’t Salesforce—they were expensive vertical solutions that had outgrown their usefulness.
Dashboard and internal tool budgets dried up. Teams building custom analytics dashboards with AI avoided new subscriptions entirely.
The pattern: simple, expensive tools doing one job poorly get replaced first.
Why Most SaaS Survives
Experienced CTOs and developers identified what AI agents can’t replicate:
Domain expertise remains king. “The bottleneck is still knowing what to build, not building,” wrote the vertical SaaS CTO. “A lot of the value in our product is in decisions users don’t even know we made for them.”
Maintenance costs compound. One developer noted, “All those non-technical users have to do is approve that app, manage to deploy and run it themselves somehow, and wait for the security breach to lose their jobs.”
Companies tried building internal LOB apps before AI. These projects consistently failed because maintenance costs exceeded subscription fees.
Organizational inertia runs deep. Large companies won’t risk mission-critical systems on vibe-coded solutions. ISO compliance, security requirements, and cloud restrictions create barriers no LLM overcomes.
One consultant’s friend at a dental clinic wanted custom staff check-in software—but she’d rather pay $5k than spend 100+ hours building it herself, even with AI help.
The Spreadsheet Analogy
Several commenters compared AI-generated tools to spreadsheets: useful for five minutes, personal to their creator, hated by everyone else. Organizations accumulate hundreds with no deletion plan.
“These new vibe-coded tools are essentially the new spreadsheets,” wrote one developer. “They are useful… for 5 minutes. They are also easily forgettable.”
The parallel extends: nobody knows how legacy Excel sheets work, yet companies depend on them. Will AI-generated apps become tomorrow’s undocumented technical debt?
Where SaaS Loses Ground
The threat isn’t total displacement but systematic erosion at the edges:
Feature bloat creates vulnerability. When customers use one feature from an expensive product, AI makes custom alternatives attractive. Several commenters mentioned maintaining subscriptions “only because of one feature.”
Integration glue becomes commoditized. One team used Claude to write connectors between services rather than paying for middleware. “We don’t want to build and support a full survey tool, but API glue is fine.”
Shallow-but-wide products suffer most. Products trying to serve many workflows poorly lose to focused, custom solutions. Deep-but-narrow products with genuine expertise retain value.
The Real Shift
Smart observers noted AI changes vendor economics, not buyer economics:
Boutique software becomes viable at smaller scales. A two-person team can now maintain products previously requiring ten developers. This enables custom solutions for niche markets ($20M TAM) that venture capital wouldn’t touch.
The result isn’t customers building everything themselves—it’s specialized vendors serving smaller segments better.
“Before AI, the $20MM TAM for this product would have made it a non-starter for VC backed startups,” one commenter explained. “But now, a two person team can build and maintain a product that previously took ten devs.”
What This Means for SaaS Companies
Successful SaaS products share characteristics that resist AI disruption:
- Deep domain expertise built over years
- Strong network effects (marketplaces, collaboration)
- Proprietary datasets
- Complex regulatory compliance
- Mission-critical uptime requirements
Products that are “just a SQL wrapper on a billing system” face thousands of competitors with spare Friday afternoons.
The uncomfortable truth: if an internal developer can replicate your product in a weekend with Claude, your pricing was always unsustainable.
Next Steps
For SaaS founders: accelerate development with AI while doubling down on domain expertise and user relationships. Your moat isn’t code—it’s understanding problems customers don’t know they have.
For development teams: expect more requests to build internal tools. Set clear expectations about maintenance costs before writing the first line.
For businesses: remember that free alternatives existed before AI. Companies chose SaaS for reliability, support, and focus. Those needs haven’t disappeared.
The software industry has cycled between build-and-buy for decades. AI accelerates the cycle without fundamentally changing it. The winners will be those who understand which problems deserve custom solutions and which need proven, maintained products.