AI Didn't Kill SaaS. It Killed the Seat.

Everyone keeps telling me software is getting cheap. They are half right, and the half they get wrong is the half that matters.

The price on your SaaS renewal is not going down. It is going up faster than it has in years. What is collapsing is the cost of the alternative: building the thing yourself. AI coding agents turned "we could just build this" from an empty negotiating threat into a real option, and that changes who holds the power in every software deal.

So no, SaaS is not dying. The seat is dying, and so is every vendor that charged for access instead of outcomes. The companies that survive will look very different from the ones that won the last decade.

I spent years on the vendor side of this argument, building customer identity infrastructure at LoginRadius. Now I run GrackerAI and I am a software buyer too. Here is how I read what is happening from both chairs.

The data says two opposite things at once

List prices are rising at record speed while buyers walk away from the purchase entirely. Both are true in 2026, and the gap between them is the story.

Start with the price tag. Vertice, which tracks more than $75 billion of software spend, measured SaaS inflation at 16.4% in June 2026, its highest reading on record. Vertice flags shrinkflation, where the same tier buys fewer features. Read a few renewal notices and the rest is easy to name: forced AI bundling and legacy plans sunset into "AI-enhanced" tiers. Little of it is new value. Atlassian's 7 to 10 percent Cloud list-price increase takes effect on October 13, on top of new usage meters arriving in December.

Now look at the buyer. McKinsey's State of AI 2026 survey of 1,719 respondents found that 32% of organizations declined to buy at least one software product or feature because they built it in-house with agentic coding tools. In tech companies, that number is 41%.

Signal	Number	Source
SaaS inflation, June 2026	16.4%, a record high	Vertice
Orgs that skipped a purchase to build with coding agents	32% (41% in tech)	McKinsey State of AI 2026
Teams that already replaced a SaaS tool with a custom build	35%	Retool 2026 Build vs. Buy report
Builders who shipped software outside IT oversight in the past year	60%	Retool
Software market cap erased in the early-2026 selloff	about $1 trillion	Yahoo Finance

Read those rows together. Vendors are raising prices on a customer base that just learned it can say no. That is not a stable equilibrium. Every renewal with a double-digit uplift is now an invitation for an operations lead to open Claude Code and see how far they get in a weekend.

The market priced some of this in, then changed its mind. The early-2026 "SaaSpocalypse" selloff erased about $1 trillion and hit seat-heavy names like Salesforce and ServiceNow hard. Software stocks bottomed in the spring, and by September a run of strong earnings had largely reversed the damage. In one week, Salesforce rose 34 percent and ServiceNow 37 percent.

I read that rally as a correction to the panic, not a verdict on the seat. Look at what the winners were selling on those calls. Salesforce used Dreamforce to push flat-fee agentic enterprise agreements alongside per-agent and outcome pricing. HubSpot announced a move to seats plus usage credits. The vendors winning back investors are the ones already moving off the seat.

Why "we'll just build it" finally stopped being a bluff

For twenty years, every SaaS sales team heard "we could build this ourselves" and smiled, because it was almost never true. Building cost months of engineering time the buyer did not have. Coding agents changed the denominator.

I wrote about this shift from the builder's side in Claude Code and the future of programming and in what actually works in AI software development. The buyer's side is the part SaaS vendors underestimated. The person deciding to build is often not an engineer at all. Retool found just over a third of its respondents were developers; the rest came from operations, finance, marketing, and data. Retool surveyed its own customers, a crowd that skews toward builders, so treat its numbers as a leading indicator rather than a census.

The categories getting rebuilt first are predictable. Retool's survey put workflow automation at the top (35% of respondents replacing it), then internal admin tools (33%) and BI dashboards (29%). CRMs, project management, and support tooling follow. These products share three traits: thin logic over the customer's own data, per-seat pricing, and a feature set that is 80% unused.

Retool's report has a clean example. A head of automation at Harmonic got tired of a $20,000-a-year tool with slow support and rebuilt it internally. The company now runs 33 internal apps on its own stack. Once a team has done that once, the default question at every renewal changes from "which vendor?" to "why buy at all?"

The sharpest signal is who is building. McKinsey found that its AI high performers, the 6% of respondents who attribute at least 5% of EBIT to AI, are much more likely to skip purchases: about half of them did, against 31 percent of everyone else. The companies getting real money out of AI are the ones most willing to cancel your contract.

The bill nobody puts in the build-vs-buy spreadsheet

Building got cheap. Owning did not. Most of the internal tools being shipped this year will cost more in year two than the SaaS they replaced, and most of the teams building them have not run that math.

The research is getting specific. A study of 302,600 AI-authored commits across 6,299 repositories found that 22.7% of the issues AI introduced were still sitting in the latest version of the code. Most were code smells, the kind of debt that makes every later change slower. A second study found that humans made 83% of the commits that later touched AI-generated files. The agent writes the code in an afternoon. People still do most of the maintaining.

Security is worse. Veracode's 2026 GenAI Code Security Report put the average security pass rate for AI-generated code at 56%, essentially flat from the year before. That is close to a coin flip on whether your weekend replacement for a $20,000 tool ships with a hole in it. I have written about how vibe coding rewrites the rules of digital security, and Retool's own number should scare every CISO: 60% of builders shipped software outside IT oversight last year. That is shadow AI with a production database attached.

The real cost of software was never the first version. It is the patch at 2 a.m., the SOC 2 evidence, the SSO request from your biggest customer, the engineer who built it leaving. I spent a decade watching companies learn this about authentication. I wrote the long version in build vs buy auth is the wrong question and priced it in the identity tax. AI does not change the lesson. It just lets more teams skip to the expensive part faster.

My rule for buyers is simple. Build the things that make you different and that you are willing to own forever. Buy anything where failure is a security incident, a compliance finding, or a 2 a.m. page. If you cannot name the person who will maintain it in 2028, you are not building. You are borrowing technical debt at a rate you have not seen yet.

Who gets eaten, and who gets stronger

The dividing line is not old versus new or big versus small. It is whether the product is a thin layer over the customer's own data or a system the customer cannot afford to get wrong.

Category	Exposure	Why
Internal admin panels, form builders	Very high	Thin logic, customer's own data, easy to describe in a prompt
Workflow automation, BI dashboards	High	Top of Retool's replacement list; the value is the workflow, which the customer knows best
Lightweight CRM and project tracking	High	Seat-priced, mostly unused features, switching cost collapsed
Tier-1 support tooling	Medium	Agents do the work, so seat counts shrink even when the vendor survives
Identity, auth, and access	Low	Failure is a breach; standards, audits, and enterprise SSO demands keep growing
Security, compliance, and regulated records	Low	Liability transfer is the product; nobody wants to own the audit
Payments, payroll, ERP, core banking	Very low	Regulation and switching cost no coding agent removes

The bottom half of that table is getting stronger, not weaker. Every internal tool built this year needs authentication, authorization, audit logs, and a security review. Every AI agent those tools spawn needs an identity of its own, which is why nearly every identity giant just bought an AI agent security company. The more software customers build, the more they need the boring infrastructure underneath it.

The real losers are vendors in the top half who respond to the threat by raising prices. Forced AI bundling on a product the customer can now rebuild is the fastest churn strategy I have seen. You are charging more at the exact moment your customer's walk-away alternative got cheap.

What I would do if I were running a SaaS company today

Stop competing with the customer's coding agent on features. You will lose that race, because the agent ships the exact feature the customer wants and nothing else. Compete on the things an agent cannot generate.

1. Price the outcome, not the seat. If AI shrinks your customer's headcount, seat pricing turns their success into your churn. I made the full case in the pricing model that proves you're actually AI-native. Hybrid pricing, a platform fee plus usage or outcome components, is now the most common model in Kyle Poyar's 2026 monetization survey, and the big vendors are moving the same way.
2. Sell accountability. The customer can build the feature. They cannot easily build the SOC 2 report, the uptime SLA, the breach liability, or the on-call rotation. Make those the product, not the fine print.
3. Become the platform their builders build on. Open APIs, clean data export, and first-class support for agents turn the internal-build wave into usage of your platform instead of a replacement for it. Fight the builder and you lose them. Host the builder and you keep them.
4. Kill the AI tax. If you raise prices by bundling AI features nobody asked for, you are funding your customer's rebuild project. Charge for AI when it does measurable work.
5. Win the shortlist before procurement starts. Buyers now ask ChatGPT and Perplexity whether to build or buy, and which vendor to pick. If your product is not in that answer, the coding agent is. That is the shift I track in AI search becoming the default for B2B discovery and why product content wins the AI shortlist.

None of this is new advice. It is the same set of hard truths about SaaS in 2026 I wrote in May, now with a procurement number attached. Product was never the moat. AI just made that impossible to ignore.

The bottom line

Software is not getting cheaper. The excuse for overpriced software is. Buyers now have a credible exit, and vendors who keep raising prices on thin products will learn what that exit costs them.

The winners will charge for outcomes, own the risk their customers do not want, and make themselves the foundation the builders build on. Everyone else is a weekend project waiting to happen.

Frequently Asked Questions

Is AI killing SaaS?

No. AI is killing per-seat pricing and thin products, not SaaS as a model. SaaS list prices rose 16.4 percent in June 2026, a record per Vertice, while McKinsey found 32 percent of organizations skipped at least one software purchase because they could build it with agentic coding tools. Products that are a thin layer over the customer's own data are most exposed. Identity, security, compliance, and payments are getting stronger.

How many companies are building software instead of buying it?

McKinsey's State of AI 2026 survey of 1,719 respondents found 32 percent of organizations declined to buy at least one software product or feature because they could build it internally with agentic coding tools, rising to 41 percent in technology companies. Retool's 2026 survey found 35 percent of its respondents had already replaced at least one SaaS tool with a custom build.

Which SaaS categories are most at risk from AI-built internal tools?

Internal admin panels, form builders, workflow automation, BI dashboards, and lightweight CRM and project tracking are most exposed, because they are thin logic over the customer's own data and are usually priced per seat. Identity, security, compliance, payments, payroll, and ERP are least exposed, because failure means a breach, an audit finding, or a regulatory problem.

Should my company build or buy software with AI coding tools?

Build what makes you different and what you are willing to own for years. Buy anything where failure is a security incident, a compliance finding, or a 2 a.m. page. AI-generated code still passes only about 56 percent of security tests in Veracode's 2026 report, and most of the cost of software is maintenance, not the first version.

Related reading on guptadeepak.com

• The hard truths about SaaS in 2026
• The pricing model that proves you're actually AI-native
• Build vs buy auth is the wrong question
• Software development with AI: what actually works in 2026
• The post-agentic organization
• When your SaaS vendor goes dark

Sources

• Vertice: SaaS inflation rate
• McKinsey: The State of AI 2026
• Retool: 2026 Build vs. Buy report
• Liu et al.: Debt Behind the AI Boom (arXiv 2603.28592)
• Sawada et al.: To What Extent Does Agent-generated Code Require Maintenance? (arXiv 2605.06464)
• Veracode: 2026 GenAI Code Security Report
• Growth Unhinged: The State of B2B Monetization in 2026
• Yahoo Finance: software ETFs as the SaaSpocalypse bottoms out