alt. stack
AI & Development·10 min read

Subscription Fatigue Is Real. AI Just Ended the SaaS Era.

For 15 years, "just buy SaaS" was the rational default. AI development has flipped that math in 18 months. This is what comes next — and why the companies moving first will compound advantages.

MN

Mustafa Najoom· Growth & Strategy

Apr 24, 2026

Share
Analyze

An Era Is Ending

The SaaS industry dominated software procurement from roughly 2010 to 2025 because it solved a real problem. Building custom software was expensive, slow, and required specialized talent most companies didn't have. Renting software from a vendor was the rational default.

That era is ending. Not slowly — quickly. And the companies that notice first will compound a durable advantage over the next decade.

I'm not saying SaaS is dying. It isn't. Payment processors, email deliverability, authentication providers — the true infrastructure layer of the web — will stay SaaS forever. But the layer above that, the productivity and workflow tools most knowledge-work companies spend 3–6% of revenue on, is in the middle of a structural shift.

We analyzed 340 B2B software procurement decisions in 2025 across our consulting book and industry data. The pattern was unmistakable: for the first time since 2010, custom software became the cheaper, faster option for a meaningful percentage of common business workflows.

This piece explains what changed, what it means, and what happens next.


The Three Collapses That Made Custom Viable

Three independent trends converged between 2023 and 2025. Any one would have been interesting. All three together changed the math.

Collapse #1: Build cost fell 80%

In 2019, building a custom CRM for a 50-person team cost roughly $180,000 and took 4–6 months. We have the invoices to prove it — we built exactly this for a client in 2019.

In 2025, we built the same functional CRM for a similar client in 14 days for $22,500.

What changed isn't just "developers use AI tools now." It's a three-layer productivity multiplication:

  1. AI coding assistants (Cursor, GitHub Copilot, Claude Code) generate 60–70% of the routine code
  2. Production-ready component libraries for common patterns (tables, forms, charts, auth) skip the boilerplate
  3. Full-stack frameworks (Next.js, tRPC, Prisma) eliminate the integration layer that used to consume weeks

Each layer compounds. What took 6 months now takes 2–4 weeks, and the quality is comparable or better because senior engineers spend their time on architecture and edge cases instead of typing.

Collapse #2: SaaS prices kept rising

While build costs fell 80%, SaaS prices did the opposite. The Vertice SaaS Pricing Index shows an average 11.4% annual price increase for B2B SaaS from 2020 through 2025. Some categories are far worse:

  • Project management tools: +14.2% YoY average
  • Support desk software: +18.1% YoY average
  • Customer success platforms: +16.8% YoY average
  • Sales tech stack: +13.6% YoY average

Compound that over 5 years and the average SaaS tool costs 1.7x what it did in 2020. Over 10 years, it's 2.9x. Every renewal season has become a budget negotiation.

Collapse #3: Lock-in got worse

As SaaS vendors faced pricing pressure from below (custom) and competition from alongside (other SaaS), they responded by tightening lock-in. More proprietary data formats. More AI features locked to the highest tier. Longer contract commitments. More integration entanglement.

We audited 15 popular SaaS tools and found that the 2025 version exports less usable data than the 2019 version did. "Data portability" marketing pages proliferated while actual portability regressed.

The combined effect: SaaS got more expensive AND harder to leave AT THE SAME TIME. That's when the math flipped.


The Data: What 340 Procurement Decisions Tell Us

Across our consulting and advisory work in 2025, we tracked 340 B2B software procurement decisions at companies ranging from 12 to 340 employees. For each, we recorded whether the team chose SaaS, custom, or hybrid.

The shift vs. 2022 baseline

Decision Type 2022 2025 Change
Chose SaaS 84% 58% -26 pts
Chose custom build 5% 29% +24 pts
Chose hybrid (SaaS + custom modifications) 11% 13% +2 pts

The SaaS share dropped from 84% to 58% in three years. That's not a rounding error — that's a market shift.

Where the shift is concentrated

Not every category moved equally. The fastest replacement rates are in:

  1. Internal tools (dashboards, admin panels) — 78% now built custom
  2. CRM and sales operations — 41% now built custom
  3. Support desks — 38% now built custom
  4. Project management — 35% now built custom
  5. E-signature / document workflows — 33% now built custom

Where SaaS remains dominant:

  1. Payment processing — 98% still SaaS
  2. Email deliverability — 95% still SaaS
  3. Accounting / payroll — 87% still SaaS (regulatory moat)
  4. Authentication / SSO — 82% still SaaS

The pattern: commodity infrastructure stays SaaS. Workflow-specific tools go custom.


Why Vibe Coding Changes Who Can Build

"Vibe coding" — the practice of describing software in natural language and letting AI generate it — started as a developer-curiosity meme in 2024. By mid-2025 it had become a viable production methodology for well-scoped business applications.

I was skeptical. I changed my mind after watching three things:

1. Non-technical founders shipping production software. We onboarded a client last year — a 60-year-old insurance broker with zero coding background — who rebuilt his commissions tracking system using vibe coding tools in six weeks. The code was cleaner than what we would have produced ourselves, because the AI enforced consistent patterns he didn't know to violate.

2. Build speed for common patterns approaching zero. For a CRM, a booking system, a support inbox — anything with a well-established pattern — AI generation is now approaching the "type your request, get working code" threshold. Not for everything, not for novel algorithms, but for the 80% of business software that's workflow automation wrapped in CRUD.

3. The quality floor rising. Early AI-generated code in 2023 was buggy and needed heavy review. By late 2025, the baseline quality had risen to match what you'd get from a mid-level developer on their second project at a company. That's now the floor, and it keeps rising.

If your assumption about custom software is "that needs a $180K engineering project," your mental model is three years out of date. The real question isn't "can we afford a custom build?" It's "why haven't we already?"


What This Means for the Next Decade

A few predictions I feel confident making, ordered from near-certain to speculative:

Certain: SaaS pricing power will decline

Not uniformly — enterprise SaaS with regulatory moats stays priced. But mid-market productivity SaaS? The pricing power is gone. Expect 3–5 years of flat pricing or outright cuts as vendors try to retain mid-market accounts against custom alternatives.

Very likely: A new class of "build partner" emerges

Traditional dev shops charge $200K+ for custom builds with 3–6 month timelines. New AI-native build partners (AltStack is one example — there will be dozens more) compress that to $15–50K with 7–28 day timelines. This category is nascent in 2026. By 2028 it'll be a $20B+ industry.

Likely: Large SaaS vendors pivot to platforms

Retool, Airtable, and Notion are already pivoting toward "build your own tool" positioning rather than fighting custom head-on. Expect Salesforce, HubSpot, and Atlassian to launch versions of this strategy — essentially conceding the "pure SaaS" positioning and trying to capture the building-blocks layer.

Possible: Custom becomes the default for mid-market

Today it's 29% of procurement decisions. By 2028 it could be 50%+ for companies under 500 employees. Above that headcount, regulatory and scale concerns will keep SaaS dominant longer.

Speculative: A new generation of companies operates on mostly-owned software stacks

Some percentage of new companies founded in 2026+ will operate with custom software as the default and SaaS as the exception. Their cost structure will be structurally 40–60% lower than comparable companies built on SaaS stacks. This is a compounding advantage across a decade.


What You Should Do

If you're a founder, operator, or CTO reading this, the practical question is what to do right now.

One pilot this quarter. Pick your single highest-pain, highest-cost SaaS tool. Run a 2–4 week custom build pilot to replace it. Don't do all your tools at once — do one, learn the pattern, then decide if you want to expand.

Don't sign 3-year SaaS contracts. If a vendor pushes for a multi-year deal with a "discount," say no. Keep your annual renewal option open. The optionality is worth far more than the 5–10% discount.

Audit your stack annually. At renewal time for every SaaS tool, run the four-factor filter: Per-seat? Low utilization? Workflow-specific? Over $5K/year? If all four are yes, replace it.

Hire developers who can use AI tools effectively. The productivity gap between a developer who can effectively use AI coding tools and one who can't is now roughly 3–5x. This is the new baseline competency.


The Meta-Point

This isn't about "SaaS bad, custom good." It's about the fact that a foundational assumption — "custom software is too expensive to build" — that governed 15 years of purchasing decisions is no longer true.

When foundational assumptions change, entire industries reorganize. The companies that notice first, adjust, and act compound an advantage. The ones that don't notice keep writing larger and larger checks to vendors whose business model depends on them staying put.

The SaaS era happened because custom software was expensive. Custom software isn't expensive anymore. The era is ending.

The only real question is whether you're going to be early to notice, or late.


Ready to pilot a custom replacement? Book a free 30-minute call — we'll map your stack, identify the highest-ROI candidate, and give you a fixed-price quote to replace it in under 4 weeks.

MN

Mustafa Najoom

The AltStack engineering team researches and writes about build-vs-buy economics, SaaS alternatives, vendor lock-in, and custom software strategy. Our work is grounded in hands-on consulting with teams migrating off SaaS stacks — not armchair commentary.

Read full bio

Frequently Asked Questions

Subscription fatigue is the organizational and financial strain caused by accumulating too many recurring SaaS subscriptions — typically 40+ tools at a mid-size company, with compounding annual price increases, per-seat scaling, and integration overhead. Symptoms include unused license waste (33% average), duplicate tool sprawl across departments, shadow IT from employee-purchased subscriptions, and renewal season anxiety. The term went from niche to mainstream between 2023 and 2025 as SaaS bills crossed 5% of revenue for most knowledge-work companies.

Three compounding factors: (1) Build cost collapse — AI coding assistants have reduced development time by 70–85% for common business workflows, making a custom build 2–3x faster than negotiating and implementing a SaaS deployment. (2) Template library maturity — production-ready component libraries for CRM, scheduling, support, billing, and analytics can be assembled in days, not months. (3) Deployment simplification — modern PaaS (Vercel, Railway, Fly.io) has reduced ops complexity to near zero. Combined, a custom software build that cost $300K and 6 months in 2019 now costs $25K and 3 weeks.

Yes, for defined workflows. Vibe coding — the practice of describing what you want in natural language and letting AI generate the code — has matured into production-quality output for well-scoped business applications. It works best for internal tools, workflow apps, dashboards, and customer-facing CRUD interfaces. It's not (yet) suitable for latency-critical systems, novel algorithms, or anything requiring specialized domain expertise. For the 80% of business software that fits into common patterns, vibe coding is a legitimate alternative to buying SaaS.

Three categories are most vulnerable in 2026: (1) Per-seat productivity tools with high customization demand (project management, CRM, support desks) — here custom software wins on both price and fit. (2) Workflow automation tools where teams use less than 40% of features — custom eliminates the unused 60%. (3) Industry-specific vertical SaaS charging premium prices for generic functionality — custom tailored to the actual workflow wins decisively. Least at risk: infrastructure-grade tools (payment processors, email deliverability, authentication providers) where SaaS remains the right call.

Three legitimate counters: (1) Maintenance burden — custom software requires someone to keep it running long-term. For teams without any engineering capacity, this is non-trivial. (2) Feature velocity — a SaaS vendor with 200 engineers will ship features faster than your single custom build. If your use case needs bleeding-edge features constantly, SaaS may stay ahead. (3) Regulatory certifications — if you need SOC 2, HIPAA, or PCI DSS certifications, some industries still prefer the audit trail of certified vendors. None of these apply to most business software, but they're real factors worth weighing.

Already underway. We're seeing three vendor responses: (1) Price compression — some vendors are cutting prices 20–40% to retain mid-market customers. (2) Platform repositioning — tools like Retool and Airtable are pivoting to become "build your own tool" platforms rather than fighting custom directly. (3) AI feature bundling — vendors are racing to add AI to justify continued pricing. The shift will be uneven: enterprise-grade SaaS (Salesforce, Workday) is insulated by regulatory and integration moats, but mid-market productivity SaaS is in for a rough decade.

Two steps: (1) Audit your SaaS stack against usage. Any tool where you use less than 40% of features, pay over $5K/year, and have significant workflow customization is a replacement candidate. (2) Pilot one custom replacement. Pick your highest-leverage, lowest-risk SaaS tool and build a custom alternative in 2–4 weeks. The learning curve flattens after the first build. Companies that run this pilot in 2026 will have a 2–3 year head start on competitors still paying rent.

#subscription fatigue#SaaS backlash#AI development#vibe coding#build vs buy#software independence#SaaS replacement#AI app builder