AWS Logo
Menu
Innovating with AI from Building to Selling at Scale

Innovating with AI from Building to Selling at Scale

At our event "Building Great AI Applications on AWS" (May 6, 2025 in the San Francisco AWS Office). We set ourselves the task of building an AI agent, evaluating what we've built, and distributing it as SaaS. Complete with a control plane, tenant isolation, metering, and even listing on the AWS Marketplace.

Bill Tarr (AWS)
Amazon Employee
Published May 21, 2025
Last Modified May 23, 2025
"There are two sides of the innovation cycle that have to go in parallel. One is the technology or the app stack, where you have to really build your app, find the product market fit, grow, and drive the adoption from there. But then there's the other part where: how do you get to distribute this thing to your customers?"
In a room full of founders and builders, Kamal dropped this on us, perfectly capturing the theme of our event Building Great AI Applications on AWS (May 6, 2025 in the San Francisco AWS Office). GenAI isn’t just a model or a feature, it’s a transformational shift in how we build and deliver software. We set ourselves the task of building an "Agent-as-a-Service", fine-tuning what we've built, and distributing it as SaaS. Complete with a control plane, tenant isolation, metering, and even listing on the AWS Marketplace.

App Stack  - Build and Grow

As Kamal suggested, innovation still requires the ability to create applications that deliver value to our customers. So we turned to Nawar Alsafar, founder of Bytez, who demonstrated how with the right toolset, we can unlock the power of AI quickly and without wasted effort.
"In one line, bam, we've integrated with a bunch of different models and actually focused on building our GenAI app instead of getting bogged down with integrations."
- Nawar Alsafar, founder of Bytez
His application demonstrated analyzing and comprehending text, generating images that matched their sentiment, and generating changelog text. The challenge? These tasks are best served by a variety of models. But using Bytez unified API, Nawar accessed dozens of the over 70k models to which Bytez provides access. Check out the blueprint Nawar demonstrated for building powerful AI apps fast, while keeping integration overhead low.

Find your Fit - Tune and Manage

With the power of AI also comes additional responsibility. That includes evaluation of models for cost and performance, full observability, and governance of our agents. Romi Datta of DataRobot explained how to take AI apps from the lab to production using their agent lifecycle management platform. DataRobot helps teams evaluate and deploy GenAI and predictive models side by side. This can help us build efficient, performant applications that we need to find product market fit and grow. They also provide tools to operate and govern agents across clouds or on-prem with full observability, access controls, and compliance.

Distribute - Make it SaaS

So what about the other part of the innovation cycle Kamal mentioned?  The one where we distribute the stuff we built? Well, fortunately that's where Omnistrate comes in. Kamal's co-founder Alok Nikhil demonstrated the power of the SaaS control plane they've built using Omnistrate to launch a SaaS version of our AI app in days—complete with pricing tiers, tenant controls, and onboarding. AI distribution is also driving more diverse deployments, like Hybrid and Bring Your Own Account (BYOA) where we deploy into our customers’ AWS account.

Secure It

Now that we can distribute, we know that there's some enterprise customers out there that would be very interested in this application that we've created. How do we handle their concerns about securing AI? We brought in Vinay Mamidi, the CEO of Whiteswan Identity Security to help us understand the world of AI security and governance and how it affects both us and our customers. He explored how prompt injection, over-permissive agent access, and sensitive data handling can be mitigated with identity-aware access control, audit trails, and just-in-time authorization for agents—especially in BYOA and hybrid deployments.

Sell It

Now that we are ready to sell, and we have enterprise customers ready to buy. So Sridhar Adusumilli, CEO of Labra explained how we can commercialize and scale your AI SaaS product through AWS Marketplace and co-sell channels. 
"It's not just about selling to one single customer or a second customer. It's more about how you're using this as your primary go-to-market channel."
Sri demonstrated how Marketplace makes enterprise procurement easier, and explained how your customers can benefit by spending down their own spend commitments to AWS by purchasing our application.

What did we learn?

With the right partners, we can cut our innovation cycle of our application to just weeks or days, and we don't have to build technical debt that will keep us from building secure applications ready to distribute and sell to enterprise customers.

Final Reflection: SaaS x AI Is Not a Feature—It’s a Stack

To build great AI apps on AWS, builders must consider a a number of architectural choices:
  1. Agent Layer: Modular tools with reasoning, orchestration, and tool use
  2. Model Choice and Evaliation: Multi-LLM abstraction via SDKs like Bytez or Bedrock
  3. Lifecycle Management: Tools like DataRobot to iterate, test, deploy, and govern agents
  4. SaaS Control Plane: Omnistrate to scale, meter, secure, and commercialize the product
  5. Security Policies: Identity, access, and compliance integrated via tools like White Swan
Together, this stack enables the Agent-as-a-Service delivery model: multi-tenant AI systems, operating across customer accounts, backed by scalable control planes, and governed end-to-end.

I missed it!

Thanks to Myers Media, we have video that will make you feel like you were in to room https://www.youtube.com/watch?v=v3qEqlf64SM - check it out!
 

Any opinions in this post are those of the individual author and may not reflect the opinions of AWS.

1 Comment