The End of the SaaS Default: How AI Coding Agents Are Reviving Custom Software

In early 2026, six of the largest SaaS companies lost $730 billion in market value in a single month. Adobe, Microsoft, Salesforce, SAP, ServiceNow, and Oracle all saw sharp declines. The cause was not a recession or a regulatory shock. It was a repricing of the assumption that enterprises will keep paying rising subscription fees for software they could build themselves. AI coding agents have made custom software fast and cheap enough to compete with off-the-shelf products. For the first time in two decades, "build" is winning arguments against "buy."

The SaaS Pricing Problem

SaaS pricing has grown 5x faster than general inflation. According to the Vertice SaaS Inflation Index, SaaS prices rose 11.4% year-over-year in 2025 compared to 2.7% general inflation across G7 countries. The average organization now spends $55.7 million annually on SaaS, an 8% increase from the prior year.

These are not isolated price adjustments. Among the top 500 SaaS companies, there were 339 pricing and packaging changes in 2025 alone. Salesforce raised Enterprise and Unlimited Edition prices by 6%. Slack jumped 20% to $15 per user per month. Salesforce's new Agentforce add-on costs $125 per user per month.

The pattern behind these increases tells a story. Up to 72% of forward ARR growth for top SaaS companies comes from price increases on existing customers, not new customer acquisition. SaaS vendors are extracting more from their installed base because new growth has stalled. And 60% of vendors deliberately obscure pricing changes through bundling and packaging shifts.

The average company now runs 106 SaaS applications. Companies with over 10,000 employees use 447. Half of all enterprises waste at least 10% of their SaaS spend, and 75% of IT teams lack clear visibility into usage or renewal dates.

What Changed: The AI Coding Agent

The tool that broke the build-vs-buy stalemate is the AI coding agent. These are not autocomplete features or code suggestion tools. They are autonomous systems that accept a task description and produce working software.

The Current Landscape

GitHub Copilot now has 20 million users and powers development at 90% of Fortune 100 companies. Cursor adds project-wide context awareness. A University of Chicago study found Cursor users merged 39% more pull requests.

Devin, built by Cognition AI, now writes 25% of Cognition's own codebase. Goldman Sachs piloted Devin alongside 12,000 developers targeting 20% efficiency gains. The tool's price dropped from $500/month to $20/month in one year. Bolt.new, a browser-based AI development platform, reached $40 million in annual recurring revenue within 4.5 months of launch. On Replit, 75% of users never write a single line of code.

The aggregate numbers confirm the shift. 92% of US developers use AI tools daily. 41% of all code written in 2025 was AI-generated. The AI code generation market was $4.91 billion in 2024 and is projected to reach $30.1 billion by 2032.

What These Tools Do Differently

Previous code generation tools helped developers write code faster. AI coding agents help non-developers build software from scratch and help developers complete projects that would have required larger teams.

Users of Claude Code report compressing three weeks of refactoring work into two days. IBM found that internal tools built with natural-language programming reduced enterprise development time by 60%. Microsoft's Q1 2025 market study shows AI development investments returning an average of 3.5x, with 5% of companies reporting returns as high as 8x.

The Klarna Effect

Klarna became the poster case for this shift. CEO Sebastian Siemiatkowski announced the company was exiting its SaaS ecosystem, targeting Salesforce and Workday specifically. Klarna replaced these tools with an in-house AI-driven stack anchored by Neo4j, with AI handling two-thirds of customer service inquiries. The company targeted $40 million in annual savings.

The reality was more complex than the headlines suggested. Klarna replaced Workday with Deel and swapped Salesforce CRM for other SaaS alternatives, not purely AI-built replacements. The story was less "build everything in-house" and more "replace expensive horizontal SaaS with cheaper alternatives plus custom AI where it adds value."

That nuance matters. The companies making this transition are not eliminating all SaaS. They are eliminating SaaS where the value delivered no longer justifies the price, and filling the gap with custom software that AI agents make feasible to build and maintain.

Other Companies Following Suit

Publicis Sapient is reducing SaaS licenses by about 50%, replacing them with AI tools. A Salesforce customer terminated a $350,000 contract to build a custom replacement using Base44, a vibe coding platform that Wix later acquired. Shopify CEO Tobi Lutke issued an internal memo declaring AI usage a "baseline expectation" for all employees. Staff must prove a task cannot be done by AI before requesting additional headcount.

The VC and Analyst Perspective

The investor class is not treating this as a cyclical downturn. They are pricing in a structural shift.

Dean Shahar of DTCP, a $3 billion fund, put it bluntly: "The SaaS world is dying. Not software itself, but SaaS as a business category. AI has turned software into a commodity."

Andreessen Horowitz (a16z) frames the shift as moving from "Software as a Service" to "Service as Software." The traditional SaaS model charged per seat for tools that organized human work. AI agents replace the labor itself. Anish Acharya, General Partner at a16z, notes over $1 billion in new revenue in the coding tools space alone in 2025. His prediction: "As coding agents work with increasing accuracy and longer time horizons, the hard problem moves from how to build to what to build."

Bessemer Venture Partners argues that vertical AI represents a 10x larger opportunity than vertical SaaS. Their reasoning: vertical SaaS addresses the 1% of US GDP spent on IT, while vertical AI addresses the 13% spent on business labor.

Cathy Gao at Sapphire Ventures predicts winning companies will look less like SaaS providers and more like "managed labor." Agents get job titles, budgets, and limits. Pricing shifts from per-seat subscriptions to payment per task completed.

Bain's Framework: Five Scenarios

Bain and Company published a framework mapping SaaS workflows against two dimensions: automation potential and penetration risk. They identified five outcomes.

• AI enhances SaaS. The SaaS vendor embeds AI features and increases value. CRM platforms adding AI-powered lead scoring fit here.
• Spending compresses. AI makes existing SaaS more efficient, so companies need fewer seats. Productivity tools where one AI-assisted worker replaces three fall into this category.
• AI outshines SaaS. Purpose-built AI solutions outperform the SaaS tool at its core function. AI customer service agents replacing help desk SaaS is the clearest example.
• AI cannibalizes SaaS. Custom AI solutions replace the SaaS product entirely. Internal tools and workflow automation fit here.
• No AI impact. Compliance, regulatory, and deeply embedded systems of record remain untouched.

Bain projects that within three years, routine rules-based digital tasks will shift from "human plus app" to "AI agent plus API." The emerging pricing model is outcome-based: tasks completed, tickets resolved, records processed. Per-seat licensing becomes an artifact.

The Case Against Overreaction

The data does not support abandoning SaaS entirely. Several factors keep SaaS entrenched for specific categories.

Determinism vs. Probability

Accounting, ERP, compliance, healthcare records, and payment processing require 100% accuracy. LLMs are probabilistic systems. They do not guarantee correct outputs. Charles Betz at Forrester notes that regulatory compliance across 20,000 legal jurisdictions keeps enterprises dependent on established vendors like SAP. No enterprise will replace their general ledger with AI-generated code.

The AI Wrapper Problem

90% of AI wrappers are predicted to fail by 2026 due to unsustainable economics. Traditional SaaS achieves 70-90% gross margins. AI-first products reach only 50-60% because API costs consume 15-30% of revenue. Building custom AI solutions requires ongoing compute costs that SaaS subscriptions bundle and amortize.

Enterprises Are Still Buying, Not Building

76% of AI use cases in 2025 were purchased rather than built internally, up from 53% in 2024. This suggests enterprises are buying AI SaaS, not replacing SaaS with custom AI. 42% of companies abandoned most AI initiatives in 2025, up from 17% in 2024. Only 19% of executives report revenue increases above 5% from generative AI.

The Productivity Paradox

A METR randomized controlled trial found experienced open-source developers took 19% longer on tasks when using AI tools, despite believing they were 20% faster. Only 30% of AI-suggested code gets accepted. 48% of AI-generated code contains security vulnerabilities. Speed gains are real for certain tasks, but the gap between perceived and actual productivity should temper expectations.

Where Build Beats Buy in 2026

The clearest opportunities for custom software over SaaS fall into specific categories.

Internal Tools and Workflows

Admin dashboards, approval workflows, reporting tools, and data pipelines are prime candidates. These tools are specific to each organization, rarely used to full capacity in SaaS form, and straightforward enough for AI agents to build in days rather than months.

Customer-Facing AI Applications

Chatbots, support agents, recommendation engines, and personalization layers deliver more value when built on proprietary data and tuned to specific business logic. A custom AI agent trained on your product documentation and customer history outperforms a generic SaaS chatbot.

Data Integration and Orchestration

Companies running 100+ SaaS applications need integration layers. Rather than paying for another SaaS to connect your SaaS tools, AI agents can build and maintain custom integrations. Anthropic's Model Context Protocol (MCP) and Google's Agent2Agent (A2A) protocol are emerging as standards for this kind of orchestration.

Vertical-Specific Applications

Industry-specific tools where generic SaaS forces awkward workarounds. A logistics company's dispatch system, a clinic's patient intake workflow, or a manufacturer's quality inspection dashboard are better served by purpose-built software that matches the actual process, not a configurable platform that approximates it.

The New Build-vs-Buy Framework

The old framework asked: "Is it cheaper to build or buy?" That question assumed building was slow and expensive. AI coding agents changed both variables.

The new framework requires four questions.

• Does this require deterministic accuracy? If the system handles financial transactions, regulatory compliance, or health records, buy from an established vendor. Probabilistic AI outputs are unacceptable here.
• Is your use case generic or specific? If 80% of companies use email the same way, buy email SaaS. If your workflow is unique to your industry or organization, build.
• What is the maintenance burden? SaaS vendors handle updates, security patches, and infrastructure. Custom software requires ongoing investment. AI agents reduce the build cost but do not eliminate the operate cost.
• Where is your competitive advantage? Build the software that differentiates your business. Buy the software that runs commodity operations. AI agents make it practical to build more of the former without the traditional team and timeline.

What Gartner and McKinsey Project

Gartner predicts 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. By 2026, more than 80% of enterprises will use generative AI APIs or deploy AI-enabled applications, up from 5% in 2023.

McKinsey reports 23% of organizations are scaling agentic AI, with 39% experimenting. Nearly two-thirds have not yet begun scaling AI across the enterprise.

AI funding represents about half of all venture capital in 2026. For non-AI SaaS companies, the environment is hostile. Median SaaS revenue expansion rates fell from 17% in 2023 to 14% in 2024, heading toward 12%. VC investors openly state it will be "very difficult for a SaaS company without native AI or agentic capabilities to find funding at any stage."

Implications for Enterprise Leaders

This transition does not require you to replace your SaaS stack overnight. It requires a deliberate shift in how you evaluate software decisions.

Audit Your SaaS Spend

Identify applications where you pay full price but use a fraction of the features. Internal tools, reporting dashboards, and simple workflow automation are the first candidates for replacement. The 50% of SaaS spend that companies currently waste is the starting point.

Build AI Development Capability

Equip your engineering team with AI coding agents. GitHub Copilot, Cursor, and Claude Code are production-ready. The 39% increase in merged pull requests from Cursor users represents real throughput gains. These tools compound: developers who use them daily build institutional knowledge about where AI works and where it fails.

Start with Internal Tools

Internal tools carry lower risk than customer-facing applications. Build an admin dashboard or a reporting tool with AI assistance. Measure the cost and time against what you would have paid a SaaS vendor. Use the results to calibrate your build-vs-buy model.

Renegotiate Before You Renew

SaaS vendors know the market is shifting. Use that pressure at renewal time. If you have a credible build alternative, you have negotiating power that did not exist two years ago.

Watch the Outcome-Based Pricing Shift

The move from per-seat to per-outcome pricing changes the economics of buy. A SaaS tool that charges per ticket resolved instead of per agent seat aligns cost with value. Some vendors will adapt their pricing models to compete with custom alternatives. Evaluate these new models on their merits.

Key Takeaways

• SaaS pricing rose 11.4% year-over-year while general inflation sat at 2.7%, pushing enterprises to find alternatives
• AI coding agents produced 41% of all new code in 2025, with tools like Devin dropping from $500/month to $20/month
• Klarna, Publicis Sapient, and Shopify are leading the shift, though the reality is "selective replacement" rather than total SaaS elimination
• Internal tools, customer-facing AI applications, and vertical-specific workflows are the strongest candidates for custom builds
• Systems requiring deterministic accuracy (finance, compliance, healthcare) remain firmly in SaaS territory
• 76% of enterprise AI use cases were purchased rather than built in 2025, showing SaaS still dominates for now
• Gartner projects 40% of enterprise apps will have AI agents by end of 2026, up from less than 5% in 2025
• The new build-vs-buy framework turns on four questions: accuracy requirements, use case specificity, maintenance burden, and competitive advantage