Goldman Sachs Research projects that AI agents will account for more than 60% of the software market's economics by 2030, a shift that would redirect the industry's profit pool away from traditional SaaS subscriptions and toward autonomous workflow systems. The bank estimates the application software market could reach $780 billion by that date, growing at a 13% compound annual rate.
The report, led by analyst Gabriela Borges on Goldman's emerging software team, frames AI agents as the next unlock for business productivity. "We believe agents will drive productivity, and software companies will capture a portion of this value," Borges writes, which is the sort of statement that sounds inevitable until you look at what's actually deployed.
Most 'agents' aren't agents
Goldman's own research undercuts the enthusiasm. The team spent six months conducting industry diligence and found that the majority of deployed systems are, in Borges' words, chatbots with basic LLM integrations. Select examples of more advanced AI exist, but these are mostly proof-of-concept projects or tools trained for internal use at software companies. The gap between the 2030 projection and the 2026 reality is enormous.
The industry can't even agree on what "agent" means. Goldman notes that companies use the term loosely, though a consensus is forming around autonomy: agents should be non-deterministic, react to environmental changes, and maintain context across tasks. By that definition, most enterprise AI deployed today doesn't qualify. It's pattern matching and decision assistance, not autonomous task execution.
So why the bullish number?
Goldman's $780 billion figure rests on a specific logic. The team modeled the customer service software market and found it could expand 20% to 45% beyond a no-AI baseline by 2030. They then argue this is actually the conservative scenario, since customer service is a cost center with a fixed number of interactions. Markets tied to revenue generation (sales, marketing, developer tools) have more room to grow. The 20% overall expansion figure is what Goldman calls a "low-end proxy."
That framing is clever but worth questioning. The estimates come from conversations with industry experts and pricing specialists, not from observed deployments at scale. Goldman is projecting the market size of a product category that barely exists in production. The methodology is value-based and cost-based pricing analysis, which tells you what companies could charge if agents work as advertised. Whether they will is another question entirely.
The timing problem
This report lands in the middle of what traders have dubbed the SaaSpocalypse. Software stocks have been in freefall since late January 2026, with the S&P 500 Software & Services Index dropping roughly 20% year to date. Salesforce, ServiceNow, and Adobe have each shed 25% to 30%. The immediate trigger was Anthropic's launch of Claude Cowork automation tools in early February, but the underlying anxiety runs deeper: if AI agents perform the work, you need fewer humans, and fewer humans means fewer software seats.
Goldman itself seems conflicted. Borges' team argues agents will expand the total software pie. But Goldman strategist Ben Snider has drawn parallels to the newspaper industry's decline in the 2000s, warning that the selloff could be the beginning, not the end. Meanwhile, a separate Goldman economics note from early March found "no meaningful relationship between productivity and AI adoption at the economy-wide level," with meaningful gains limited to software development and customer service.
Borges acknowledges the infrastructure isn't ready. Platform standardization is at least 12 months away, she writes. Developers and customers still face concerns around data integrity, security, and authentication. History suggests application adoption follows platform standardization, not the other way around. Until that platform layer stabilizes, the 60% figure is a forecast about a market that doesn't yet have the plumbing to exist.
Who wins if Goldman is right?
The report contains one observation that feels genuinely important. Companies that wrap workflows in AI agents become, as Borges puts it, the new user interface for knowledge workers. That's a significant claim: the agent layer wouldn't just automate tasks, it would become the primary way people interact with business software. The vendor that owns that interface captures the productivity gains instead of passing them to customers.
This is the real tension. SaaS companies spent two decades building per-seat pricing models. If agents compress seat counts (10 agents doing the work of 100 reps, as SaaStr's Jason Lemkin has argued), existing vendors need to reprice entirely, something Goldman compares to Adobe's painful subscription transition around 2012 to 2014. The bank expects SaaS incumbents to claim a large share of the new agent market, but the transition won't be linear, and AI-native competitors are already showing up.
Goldman's latest podcast episode, recorded March 10, is titled "Can Software Survive AI?" The fact that the bank is even asking that question, months after projecting a $780 billion market, tells you something about how fast sentiment is shifting. The numbers in the Borges report may prove prescient. But right now, they describe a future that's several infrastructure layers and at least one pricing revolution away from reality.
