SaaS Is Dying — AI Agents Are Eating Software, and Here's What I See Happening

The Shift Nobody Wants to Admit Is Already Happening

I've spent over 20 years building automation systems — at Amazon, Ring, Tinder, and now Motorola Solutions. And in all that time, I've never seen anything move as fast as what's happening with AI agents in 2026.

The pattern I'm seeing is this: the instruction used to be "Open Salesforce and update the opportunity." Today, it's becoming "Handle this renewal risk." An AI agent gathers context, checks permissions, executes across systems, logs its actions, and moves on. The SaaS app is still there — but the human workflow that surrounded it is dissolving.

Business Insider recently called it the "2026 software crash." I think that's a little dramatic, but the underlying signal is real. Power is shifting from the apps to the agents on top of them.

What Agentic AI Actually Means for How We Work

Silicon Valley is calling this the "centaur phase" — a term borrowed from chess, where a human paired with a computer outperformed either alone. The idea is that an engineer or QA manager paired with an AI agent is now the most powerful unit in the knowledge economy.

As someone who has championed AI-driven quality engineering for years — building test generators, self-healing Playwright frameworks, LLM-assisted automation — I've felt this shift up close.

Here's what's changed: a year ago, AI tools in my workflow were assistants. They helped me write tests faster, triage failures smarter, summarize logs. They still needed me to orchestrate every step.

Now? I'm running agents that plan, execute, evaluate, and retry — with minimal intervention. The cognitive load has shifted from doing the work to reviewing the work. That's a completely different skill set.

Three Real Signals I'm Watching in 2026

1. Observability platforms are pivoting to agent monitoring. New Relic just launched a dedicated AI agent platform. The fact that observability companies — whose entire job is to watch software — are now pivoting to watch agents tells you everything about where the market is heading. Agents are becoming first-class infrastructure.

2. Non-technical users are building automation. Google's Opal platform now lets non-engineers create automated workflows via text prompts powered by Gemini. This is huge for QA: it means business stakeholders can prototype their own acceptance criteria as runnable workflows. As a Suneet Malhotra QA automation leader, I see this as both a threat and an opportunity — QA needs to own quality governance as execution becomes democratized.

3. Enterprise leaders are rethinking the SaaS stack. 72% of CEOs now see themselves as the primary AI decision-maker in their organizations, according to a recent Forbes analysis. That means budget conversations are happening at the top. And the question isn't "which SaaS tools do we renew?" — it's "which workflows can we replace with agents?"

What This Means for QA Engineering

Let me be direct about something: AI agents executing across enterprise systems without robust quality gates is a disaster waiting to happen.

I've seen what happens when automation moves faster than validation. Flaky tests, silent failures, misrouted actions — these compound fast at agent scale. A bad test suite costs you a few hours of debugging. A bad agent suite costs you data integrity, compliance exposure, and customer trust.

This is the argument I'm making to every QA leader I talk to right now: agentic AI needs agentic QA.

That means:

• Test frameworks that can evaluate agent behavior, not just function outputs
• Observability hooks that capture agent decision chains, not just API calls
• Quality gates baked into agent orchestration layers (LangGraph, CrewAI, custom pipelines)
• Human-in-the-loop checkpoints for high-stakes agent actions

The Suneet Malhotra AI testing philosophy has always been: automate the repetitive, guard the critical. That principle applies to agent systems just as much as it did to Selenium suites in 2015.

How I'm Adapting My Own Practice

Here's what I'm doing differently in 2026:

I've started treating AI agents as test subjects, not just test tools. When I build an agent workflow, I build an adversarial testing layer alongside it — prompts designed to expose edge cases, unexpected inputs, and failure modes the happy path never surfaces.

I'm investing in prompt engineering as a core QA skill. The way you structure a system prompt determines how an agent fails. Understanding that failure mode is now a quality engineering competency.

I'm pushing for agent observability in every project I touch. If you can't trace what an agent decided and why, you can't debug it. Full stop. New Relic's new platform is interesting for exactly this reason.

The Bottom Line

SaaS isn't dying tomorrow. But the workflows built on top of SaaS — the human-operated, screen-by-screen, tool-by-tool execution — those are evaporating fast.

The winners in this next phase won't be the companies with the best apps. They'll be the companies with the best agents, the best quality systems watching those agents, and the engineering leadership smart enough to build both.

That's the opportunity I see for quality engineering in 2026. And honestly? It's the most exciting moment I've seen in my entire career.

If you're thinking about how to build quality systems for AI agent workflows, I'd love to connect. Find me on LinkedIn or explore more at suneetmalhotra.com/blog.