AI agents are eating SaaS: the $139 billion market replacing your software stack

Something happened this past week that deserves more attention than it’s getting.

In the span of seven days, Amazon launched an AI health agent for 200 million Prime members. Shopify declared every store on its platform “agent-ready by default.” NVIDIA open-sourced an entire enterprise agent toolkit at GTC. Picsart opened an AI agent marketplace for its 130 million users. And Visa started testing AI-initiated payments with eight of the world’s largest banks.

None of these are chatbot upgrades. These companies are rebuilding their core products around autonomous AI agents that act and transact on behalf of humans.

The agentic AI market sits at roughly $9.14 billion today. Fortune Business Insights projects it will reach $139.19 billion by 2034, a 40.5% compound annual growth rate. Honestly, that number might be conservative given what shipped this week.

The OpenClaw moment everyone missed

If you want to understand why the agent revolution accelerated so fast, look at OpenClaw.

OpenClaw is an open-source framework that lets anyone deploy autonomous AI agents. We’re talking agents that complete tasks, make decisions, and take actions across Telegram and WhatsApp. Created by Austrian developer Peter Steinberger (who has since joined OpenAI), the tool proved something that caught a lot of people off guard: powerful AI agents can run on a Mac Mini sitting on your desk.

NVIDIA CEO Jensen Huang called it “the largest, most popular, the most successful open-sourced project in the history of humanity” and declared it “definitely the next ChatGPT.” Whether you think that’s hype or not, there’s a real parallel. ChatGPT made large language models tangible to regular people in late 2022. OpenClaw is doing the same thing for autonomous agents right now.

Developers are discovering that running agent fleets on home hardware is far cheaper than cloud APIs for big models. Chinese tech firms are adopting it aggressively. Meta responded by rushing out a desktop app for its Manus agent. The barrier to deploying autonomous AI agents collapsed in weeks, not years.

86% of companies are putting money behind this

NVIDIA’s State of AI 2026 report, drawing from over 3,200 respondents globally, found that 86% of enterprises plan to increase their AI budgets this year. Nearly 40% expect increases of 10% or more. In North America, that figure jumps to 48%.

The number that matters most: 88% report that AI has already increased their annual revenue, and 87% say it has reduced costs. Companies are seeing returns and doubling down. The speculation phase is over.

The top spending priority? Optimizing existing AI workflows and improving production efficiency. Not exploration or proof of concepts. That tells you where the market is: past experimentation, deep into scaling.

Gartner projects worldwide AI spending will hit $2.52 trillion in 2026, a 44% year-over-year increase. Goldman Sachs estimates AI companies alone may invest over $500 billion. The capital is flowing, and increasingly, it’s flowing toward agents.

Amazon’s health agent: what “always-on” actually looks like

Amazon’s Health AI accesses real patient data through state health information exchanges, including diagnoses, medications, and medical history. It books appointments through One Medical, reads lab results, manages prescriptions.

Prime members get five free virtual consultations for over 30 common conditions, worth up to $145. Amazon is subsidizing actual healthcare delivery through an AI agent, betting the experience will convert users to $99/year One Medical memberships.

200 million Prime members now have a 24/7 health agent in their pocket. No scheduling, no waiting rooms. The agent handles triage, the human doctor handles the complex stuff. This is what SaaS replacement looks like in healthcare. Not a better EHR interface, but an agent that makes the interface irrelevant.

Shopify bets the store on agentic commerce (literally)

Shopify president Harley Finkelstein stood at the 2026 Upfront Summit and called this “the transformation of a lifetime.” CEO Tobi Lutke followed up by announcing that every Shopify store would be agent-ready by default.

Their Agentic Storefronts initiative is worth understanding. With a single setup step, any Shopify merchant can make their products discoverable to AI agents on ChatGPT, Perplexity, and Microsoft Copilot. The agents discover, compare, and complete transactions on behalf of shoppers. Full loop.

This matters because only 18% of U.S. retail purchases happen online. The Shopify thesis is that agentic shopping changes this ratio dramatically. An AI personal shopper that knows your preferences, budget, and past purchases can surface products you’d never find through search. For the millions of Shopify merchants who struggle with product discovery, agents become the new front door.

OpenAI is already working with Shopify, Etsy, Walmart, and Amazon on agentic shopping integration after its first attempt stumbled. The race to be the default commerce layer for AI agents is on.

Picsart: the first agent marketplace for creators

Picsart launched something that looks small but has big structural implications: a marketplace where 130 million creators can “hire” AI agents for specific creative tasks.

The initial lineup includes four agents. Flair integrates with Shopify to analyze market trends and suggest product photo improvements. Resize Pro uses generative AI to intelligently extend frames when resizing content across platforms. Remix applies style transfers (vintage film, cyberpunk, watercolor) to entire photo libraries in bulk. Users interact with these agents on WhatsApp and Telegram, with adjustable autonomy levels.

Starting at $10/month, this is the first real test of an agent marketplace model in the creator economy. Picsart built the initial agents, but the marketplace will include agents from external creators too. If this model works, with creators browsing and hiring specialized AI agents the way they currently subscribe to SaaS tools, it previews how entire software categories get replaced.

NVIDIA’s open-source agent infrastructure

At GTC on March 16, NVIDIA released the infrastructure to build agents. Not slides about agents. Actual tools.

The NVIDIA Agent Toolkit includes four components: Nemotron (open models optimized for agentic reasoning), AI-Q (a blueprint for agents to perceive, reason, and act on enterprise knowledge), OpenShell (a runtime with policy-based security and privacy guardrails), and cuOpt (an optimization skill library).

The AI-Q Blueprint already tops DeepResearch Bench accuracy leaderboards while cutting query costs in half through a hybrid approach mixing frontier and open models. Seventeen enterprise partners — Adobe, Salesforce, SAP, Cisco, CrowdStrike, ServiceNow among them — are building with it.

NVIDIA is making a specific bet: the agent infrastructure layer should be open. That’s a sharp contrast to the closed API approach that dominated the chatbot era. And yes, it’s self-serving. Open agent infrastructure means more agents running on more hardware, which means more GPU demand. But for builders, the result is the same: enterprise-grade agent tools are now free.

Visa: when agents start spending money

This is the part that really got my attention. Visa isn’t testing whether AI agents should initiate transactions. They’re past that. They’re testing how to make it safe at scale.

The Visa Agentic Ready program has enlisted Commerzbank, Revolut, HSBC UK, Nexi Group, DZ Bank, Bank Leumi, Nationwide, and Barclays to test agent-initiated payments in controlled production environments. The approach uses tokenization (substituting card numbers with unique digital codes) and biometric authentication (fingerprints or face scans) to link agent transactions to verified humans.

On March 18, Visa Crypto Labs released Visa CLI, a command-line tool that lets AI agents initiate card payments without storing API keys. The same day, Stripe-backed Tempo went live on mainnet and published the Machine Payments Protocol for standardized agent-to-service payments. Coinbase’s x402 protocol embeds stablecoin payments directly into HTTP requests, so an agent can hit a paywall, pay in USDC, and continue its task without human intervention.

McKinsey estimates agentic commerce will generate $3 to $5 trillion in global revenue by 2030. The infrastructure for that is being built right now, this week, across card networks and crypto rails simultaneously.

The SaaS replacement thesis

Bessemer Venture Partners put it bluntly: vertical AI “represents a fundamentally larger opportunity than vertical SaaS ever did” because “unlike vertical SaaS, which typically captures a fraction of Fortune 500 IT spend, Vertical AI taps directly into the labor line of a P&L.”

That distinction matters. Traditional SaaS charges for software tools. AI agents charge for work done. When one person with AI agents does what five employees used to handle, per-seat pricing starts to look like a relic.

The market has noticed. The “SaaSpocalypse” of early 2026 wiped roughly $285 billion from software stock valuations. Y Combinator’s Winter 2026 batch is heavily weighted toward startups building AI agents that replace entire SaaS categories. The investment thesis has shifted from “AI-enhanced software” to “AI that replaces software.”

Multi-agent specialization is the emerging architecture. Rather than one general-purpose chatbot, companies deploy fleets of specialized agents. Harvey for legal, Jasper for marketing, domain-specific agents for healthcare and supply chain. Research shows this approach reduces process handoffs by 45% and improves decision speed by 3x compared to monolithic systems.

The shift is from software-as-a-service to service-as-software. Agents don’t help you use tools. They replace them.

What this means, concretely

SaaS companies: The threat is not that agents compete with your features. Agents will eliminate your product category. Bolting a chatbot onto your existing UI won’t cut it.

Enterprises: The 86% budget increase is a starting gun, not a trend report. Companies deploying domain-specific agent fleets will outperform those still evaluating chatbot pilots. The productivity multiplier changes headcount economics in ways that are hard to ignore.

Developers: The barriers are gone. OpenClaw runs on your desk. NVIDIA’s toolkit is open source. Picsart just proved you can build agents for a marketplace. The real question is whether you’re building one that solves a specific problem better than the SaaS tool currently doing it poorly.

Fintech: A split is forming. Traditional card rails for regulated human commerce, stablecoin-based payments for autonomous agent commerce. Visa and Coinbase are both building for this. If you’re not thinking about agent-initiated transactions, you’re planning for a future that won’t arrive.

The $139 billion projection for 2034 assumes gradual replacement. What we saw this past week was not gradual. Amazon, Shopify, NVIDIA, Visa, and Picsart didn’t coordinate their announcements. They all arrived at the same conclusion independently.

The agents are here, and the software stack you’re paying for is in their sights.