Agentic SaaS: Your next growth market is already here

By: Matt Yanchyshyn, VP, AWS Marketplace & Partner Services – AWS

Agentic AI represents the third major shift of enterprise software and how it creates value. The first brought applications from desktop to web and mobile, and the second moved software from on premises to cloud, creating a software as a service (SaaS) market worth $300 billion today and projected to reach $576 billion by 2029 that created enormous value for the software providers that moved early. Amazon Web Services (AWS) has been at the center of both, and this post explains how we can again support software providers on capturing the agentic opportunity through capabilities, distribution, and commercial models.

Users are increasingly interacting with software through AI assistants and agentic platforms that invoke capabilities on their behalf instead of the product’s own UI. Your product is still doing the work, but the user might never see it. The buying journey has changed too because the decision-making for technology is no longer owned solely by IT or engineering. Business users from human resources (HR), legal, finance, and other functions know what they need, and they’re buying it themselves. This era also brings an entirely new class of users: autonomous agents that reason, plan, and execute multistep tasks with minimal human intervention. All of this makes agentic AI an expansion of the addressable market, with new distribution channels that operate at machine speed at any hour, not a threat to existing revenue.

Agentic AI represents an evolution from software that users operate to capabilities that agentic systems execute, changing how software is built, priced, distributed, and consumed. Agents and humans alike are already choosing preferred providers based on discoverability and reliability. Your customers are building agentic systems today, and those systems will use your capabilities if they can find them, understand them, and trust them . . . or they’ll route around you.

What “Agentic SaaS” actually means

Agentic AI isn’t killing SaaS, it’s transforming it following similar patterns we’ve seen and pioneered before. Your product capabilities, your domain expertise, your customer relationships, your data moats, and decades of accumulated domain expertise don’t become less valuable in an agentic world. They provide the trusted, reliable, and well-described capabilities agents need to solve industry challenges. Agentic SaaS is the next evolution of software that’s now consumed by a hybrid workforce of both humans and agents and a growth opportunity for software providers that act today. The window for independent software vendors (ISVs) to act is measured in months, not years.

Build for agentic systems: Make current SaaS capabilities consumable by agentic systems and build your own agents for high-value workflows

This evolution doesn’t require a radical departure from SaaS engineering fundamentals. Multi-tenancy, tenant isolation, scale, security, observability, and governance remain the table stakes that enterprise buyers evaluate before any production deployment. What changes is the execution model. Agents operate autonomously, maintain complex state across sessions that can span hours, coordinate across tool chains and subagents. And they must enforce the compliance and authorization boundaries that you establish without getting you in the loop again. Most organizations build and rebuild these primitives for every framework, model, and use case change. That’s months of infrastructure work before a single agent reaches a real user.

AWS solved this at scale so you don’t have to solve it every time, and we did it based on lessons learned from implementing AI at scale for our own products. Amazon Bedrock AgentCore is our hero platform for taking agents from prototype to production. It brings runtime isolation, memory, identity, gateway, policy, and observability together and fully integrated with the rest of the AWS stack. Agent capability without accuracy is a liability, and this integration between infrastructure, model, platform, and data foundation layers enables your agents to be grounded in your proprietary data. Automated reasoning is the safety layer for the agentic era, proving mathematically that agents stay within defined boundaries, and our platform uses it to block unauthorized actions that violate policies and to enforce tool access. Security is built in by design, with network isolation and fine-grained authorization, and AWS industry certifications providing the foundation to reduce the compliance burden for your agentic deployments.

Customer adoption of agentic AI tools and features is accelerating, and we’re seeing this in the growth of AI agents and tools in AWS Marketplace. This fast-growing category went from 900 solutions at launch to over 3,300 in less than a year. Most are not standalone agents, they’re existing SaaS products that have been agentified, extending proven software to work inside agentic workflows.

The results speak for themselves:

• Agentforce, Salesforce’s enterprise agentic AI solution available in AWS Marketplace, uses foundation models (FMs) through Amazon Bedrock, implemented natively on AWS Global Infrastructure, to deliver faster, smarter resolutions for customers.
• CrowdStrike is deploying autonomous security agents that detect and remediate threats without human intervention.
• Workday built a sales companion on AWS, an AI-powered solution that condenses structured and unstructured data to surface product intelligence for sales teams in real time, delivering a 75% reduction in sales prospecting cycle time and saving 5 hours per sales rep per week.
• Zendesk customers use Amazon Bedrock and Amazon Connect to automate 80% of customer service interactions.
• Druva built DruAI with Amazon Bedrock and AgentCore, transforming cyber investigations that previously took 30–60 days into minutes-long conversational workflows, with the multi-agent system now resolving 68% of customer issues without manual intervention.

These aren’t incremental features. These are companies rethinking what their software does and who uses it.

Design for agentic discoverability: Position your solutions to be found and chosen when agents select tools

Agents interact with software through APIs and structured interfaces in a similar way developers do. But although humans navigate UIs and evaluate feature lists, agents select tools based on structured metadata and natural language descriptions. Agentic discoverability is the new distribution, defined by the degree to which AI agents can find, understand, and choose to invoke your capabilities. Whereas human distribution channels reward brand familiarity and sales reach, agent selection rewards technical precision, which makes the architecture and descriptions of your product the most powerful go-to-market asset.Investing in agentic discoverability isn’t solving a one-time technical configuration. It’s building the equivalent of early domain authority, an agent selection advantage that compounds over time. While building a shopping assistant that selects from hundreds to thousands of internal tools per session, engineering teams at Amazon found that defining structured schemas and precise semantic descriptions for each tool was the central engineering challenge. The agent’s ability to identify and invoke the right tool depends directly on the quality of those descriptions.

The Model Context Protocol (MCP) has emerged as the industry standard for making capabilities discoverable to agents, doing for agentic capabilities what REST APIs did for web applications: make them discoverable, composable, and invokable across any agent runtime. The Agent2Agent (A2A) protocol complements MCP for cross-agent discovery: where MCP handles agent-to-tool invocation, A2A handles agent-to-agent delegation, with structured metadata describing the agent’s capabilities and interface. Supporting both protocols in your solutions maximizes the probability that agents in your domain will find and use your solution rather than a competitor’s.

Salesforce continues to lead the charge with the recent introduction of Headless 360, which makes all its capabilities accessible to agents and humans alike using APIs, MCP tools, or command line interface (CLI) commands. The surface changes, but the platform and its capabilities don’t.

AWS provides MCP and A2A support across the full stack. On the backend, you can provide an immediate agentic surface for your current capabilities without requiring rewrites converting REST APIs and serverless functions into MCP tools with Amazon Bedrock AgentCore Gateway. On the distribution front, AWS Marketplace supports listing both MCP and A2A servers, making your tools and entire agents discoverable within agentic workflows.

For customers already transacting through AWS, this removes the friction of a new vendor relationship and accelerates procurement. Agents can also interface with AWS Marketplace using agent mode through our own MCP server and a native connector available in Anthropic’s Claude, querying AWS Marketplace directly or interfacing to find solutions based on needs and problem descriptions, not keyword searches. And we’re not only building this for partners. We built the AWS Partner Central agent experience using Amazon Bedrock and AgentCore, complementing the AWS Partner Central application used by thousands of partners every day. The platform we’re providing for you to build on is the same one AWS uses to build with.

Align pricing to value: Capture the measurable outcomes that agentic usage creates

Agents create measurable, attributable business value in units that per-seat pricing was never designed to capture: a customer support agent resolving 10,000 tickets overnight, an invoice processing agent handling 50,000 lines without human review, a network monitoring agent running around the clock. Leading SaaS vendors have already begun building commercial models that capture this value, and the consistent pattern is a hybrid model: predictable base revenue from humans combined with consumption-based upside from agent usage, not a replacement of one model for the other.

The practical prerequisite is instrumentation because it provides you with powerful data to understand and show your customers precisely what their agents consumed, what outcomes they produced, and what the unit economics look like. Having the most granular outcome data will help you define pricing metrics for your category instead of matching what your competitors are pushing. This is where observability becomes a commercial capability, not merely an operational one. The ability to trace agent sessions, tool invocation, and resolution events and then map those traces to billable units is the foundation for consumption pricing.

AgentCore provides this instrumentation inherently, with built-in observability that tracks agent execution. AWS Marketplace then closes the commercial loop with support for pay-per-outcome pricing structures and private offer negotiation, so you can establish hybrid pricing models through the same channel that already distributes your existing software, without a new billing platform or a new sales motion.

Zendesk, a leader in AI-powered customer service, used AWS Marketplace to shift from per-seat licensing to outcome-based pricing where customers pay for results, such as tickets resolved. The flexible pricing infrastructure of AWS Marketplace made the transition possible without rebuilding their commercialization stack.

Build the agentic era on AWS

AWS has helped software providers navigate every major platform shift and succeed for the past two decades, and this one is no different. We provide both the platform and expertise to architect for this next generation of software solutions, supporting you to evolve from SaaS products to Agentic SaaS solutions and start winning your category today.

AWS is investing in every layer of the AI stack so you don’t have to, from custom silicon and foundation models to agentic platforms and solutions. We also launched several new programs and initiatives this year to help you make this transition, including expanding the AWS Migration Acceleration Program to cover AI-based new builds, growing AWS ISV Accelerate to support partners engaged in co-sell activities, and investing an additional $100 million in the Generative AI Innovation Center for hands-on support building and deploying agentic AI applications.

Your next growth market is already here, and it won’t wait too long. Get started today with our learning series for building agentic systems on AWS.