For twenty years, SaaS won by turning bespoke software into subscriptions. The bet was simple: configuration beats customization; buy, don't build. That cost curve is flipping. Generative-AI tooling is compressing the time (and therefore cost) to design, code, and maintain narrow, business-specific software. Developers complete coding tasks dramatically faster with AI pair-programmers, and product teams ship sooner when gen-AI is woven into the lifecycle.
This doesn't mean "SaaS is dead." It means the market share of broad, horizontal suites that try to be everything for everyone will erode as more companies build small, sharp tools tightly fitted to their workflows.
Controlled studies and field data show substantial speed-ups (~55% faster task completion for common coding tasks) and shorter time-to-merge, which directly lowers the cost of custom feature work.
Enterprises now juggle hundreds of SaaS apps; BetterCloud's 2025 report shows ~106 apps on average overall, while Zylo finds large enterprises manage ~275, with spend rising again. CFOs are consolidating and questioning seat-based waste.
Package ecosystems and frameworks already made assembly faster: npm passed ~2.1M OSS packages by late 2024; PyPI lists ~660k+ projects; Rails and peers normalized "convention over configuration." AI now accelerates the assembly of those parts.
CRM suites: Sales Cloud's list pricing today climbs from $175/user/month (Enterprise) to $350/user/month (Unlimited), with add-ons charging per user or per login. A custom CRM that's purpose-built for your pipeline deletes Viewer/Agent seat taxes.
Analytics/BI: Role-based licensing (Creator/Explorer/Viewer) monetizes read access; "just to look" still costs. Internal dashboards over your warehouse avoid per-viewer tolls entirely.
Helpdesk: Zendesk prices per agent and upsells AI/WFM/privacy bundles. A product-integrated support tool removes per-agent metering and the "second system" context switch.
Work tracking: Jira scales, but admin overhead rises with custom fields, schemes, and workflow permutations. A minimal internal tracker with three workflows and repository-native links does the job with orders of magnitude less complexity.
CRM administration: These platforms often need specialist administration and paid "success" tiers. When the data model is simple and local to your go-to-market motion, a small internal app with direct warehouse integration can cut both admin time and subscription load.
Horizontal SaaS keeps data in product silos, then charges you (directly or indirectly) to move and model it. A custom app can sit on the same store as analytics (e.g., the warehouse), eliminating sync jobs, API rate limits, and "ETL frictions."
Teams routinely ignore 80% of a suite's features yet still pay for the privilege. Bespoke tools can be intentionally narrow: a single purpose, one or two killer automations, and exactly the fields that matter.
Common pain: high per-user cost at upper tiers; paid add-ons; complexity around fields/objects; ongoing admin.
What to remove: seat taxes for read-only roles; "universal" object model; paid success plans; separate AI upsells. Build a slim app over your data model with programmatic rules and only the automations you actually use.
Common pain: configuration sprawl (workflows, schemes, custom fields) and performance/migration headaches.
What to remove: the combinatorial config surface. Keep three states, one SLA, repo-native links, and auto-generated release notes.
Common pain: per-agent pricing and a separate tool for agents to live in; rigid macros; data duplication.
What to remove: agent seat metering and siloed ticket data. Add in-app help, unify identity, and let an LLM draft replies from product context.
Common pain: role-gated licenses, vendor-specific modeling layers.
What to remove: viewer licensing and external semantic layers; generate dashboards directly over your warehouse tables, keep business logic in code and tests.
Yes, the aggregate SaaS market is large and still expanding, but the mix is shifting. Enterprises report rising SaaS spend alongside app-portfolio consolidation, and several major vendors announced price increases in 2025. That combination - consolidation plus higher list prices - pushes teams to ask where a small custom tool would serve better than another enterprise seat.
Open-source frameworks (Rails, then Django, Spring, React, Node, etc.) standardized patterns and flattened the long tail of web app plumbing. Package ecosystems put batteries on the shelf: npm alone passed ~2.1M OSS packages by late 2024; PyPI counts ~660k+. Gen-AI takes the next step by automating the glue - the repetitive scaffolding, boilerplate, tests, mappings, and docs that previously made "build" the slow option. The result is a practical crossing point where a tight internal app is cheaper and better-fitting than a broad external suite.
SaaS will keep winning where: (1) compliance/regulatory guarantees or certifications are non-negotiable; (2) network effects or ecosystems dominate (e.g., ad platforms, payments at global scale); (3) the problem is genuinely horizontal and rarely changes. But the "do-everything platform with per-seat pricing" is losing its default status as AI reduces the cost of building exactly what you need.
SaaS won the 2010s by turning customization into configuration. AI is reversing that advantage. As the effective cost of bespoke software drops, the rational strategy for many teams is to buy fewer general-purpose seats and build more narrowly focused tools that live right on the company's data and workflows. The result is less seat tax, less configuration drag, fewer integration tolls, and software that finally fits.