AWS AI Chief On Making AI ‘Cheap And Easy,’ OpenAI Strategy And AI Products Partners Need To Be Selling

“What we’re trying to do at AWS is create an environment where building one more really valuable AI system is as quick and as cheap and as easy as just building another website,” says Matt Wood, chief AI and technology officer at AWS.

Amazon Web Services envisions a future in which AI systems are as cheap and easy as building a website.

“What we’re trying to do at AWS is create an environment where building one more really valuable AI system is as quick and as cheap and as easy as just building another website,” said Matt Wood, chief AI and technology officer at AWS.

Additionally, the $169 billion global cloud market share leader has several specific AI products it wants AWS partners to be selling to unlock all the AI magic for customers.

In an interview with CRN, Wood takes a deep dive on AWS’ AI vision, its OpenAI partnership plans and what AI products partners should be selling to clients right now.

What’s one big AI investment that AWS is making to win customer mindshare?

We invest in making sure that the core learning loop for big AI systems operates as quickly and as efficiently as possible by putting in custom-designed silicon to enable very large models to be trained and also models to be fine-tuned based on customer need.

We invest in making sure that once those models are trained, we can run inference against them, predictions, generations, videos, text, images, whatever it might be as quickly and as efficiently and critically with the levels of privacy and security that our customers have come to expect from AWS.

And we’re building a set of capabilities for agents, which allow our customers working with AWS and our partners to be able to build autonomous systems that drive meaningful efficiency and reinvention of their core businesses as quickly and as cheaply as possible.

And to do that, we found that there’s a couple of building blocks which turn out to be really, really important.

Click through to read about AWS’ key building-block products, OpenAI partnership plans and key AI strategies.

What are these AWS “building blocks” that partners should be selling and leveraging in the market right now?

The first is you really want an effective context layer, which contains all of the information from across your organization, which is agent accessible.

So we made available AWS Context.

This is a service from AWS that a partner can turn on with a customer, which allows your agents to be able to understand the information inside your organization—whether that is structured information or unstructured information, KPIs, dashboard data—and use them to build agentic workflows.

These are quick and efficient and actually allows you to do work that is meaningful much more quickly, much more efficiently than was ever possible before.

Get through that backlog, understand an email, but also reinvent what it means for your own organization to interact with your employees, your customers and to deliver entirely new customer experiences.

That all lives in an agentic context layer, which is delivered to be as broad as possible, but without requiring large-scale data migration or data modernization.

So that’s what AWS Context allows.

What are the other two AWS AI building blocks?

We also have capabilities like Amazon Quick. Amazon Quick is a chat assistant.

It looks like you can just interact with it, but it’s much deeper than that.

It actually interacts with our context layer from AWS Context and includes its own custom-designed harness, which allows everyday workers—whether those are builders or knowledge workers, whether they are lawyers or project managers—to be able to understand all of that contextual data and build their own agents to be able to run their own workflows completely automatically.

And we found inside Amazon that this was just a tremendous boost.

It was a night-and-day difference once we made Amazon Quick available and how our own teams were using artificial intelligence.

Finally, we have capabilities for building AI systems which are quick, understandable, efficient, verifiable, and which are secure. And that’s what AgentCore allows us to do on Bedrock.

AgentCore allows us to bring all of these remarkable models together with their own harnesses and their own security capabilities to be able to build very, very long-running, managed agents—which don’t just run over tens of seconds—but can run over days, weeks, or even months, in order to be able to complete very complex tasks on behalf of humans.

How is AWS trying to make AI easier and cheaper for customers?

AI can be very fast-moving, it can be kind of intimidating, it can be somewhat technical, but the opposite has been true.

What we’re trying to do at AWS is create an environment where building one more really valuable AI system is as quick and as cheap and as easy as just building another website.

If you rewind 30 years and just think of the internet, there were maybe 10 or 12 really big websites that were just emerging and growing then. There were lots of different emergent use cases, but there were probably about only 10 or 12 really big websites. And creating one more big website was an intimidating technical prospect.

But if you fast-forward to current day, building another website is incredibly cheap and incredibly quick and incredibly easy and carries the opportunity of delivering economic growth and prosperity of the scale of some of those very earliest internet sites.

Our job at AWS is to make that’ s true for AI as well.

Today we have 6 or 12 really meaningful, economically valuable workloads for AI.

We envision a world where there are going to be millions and millions and millions of equivalently large AI systems operating inside organizations and reinventing those organizations’ customer experiences for their own customers.

Now today, it will require more than just AWS to be successful to get there.

We need to work with our customers and with our partners to enable that vision and bring it as close to tomorrow as possible.

Talk about AWS’ partnership with OpenAI.

We were very happy to bring the latest OpenAI models to customers on Bedrock on day one.

We’ll continue to do that as more and more models become available, as the latest versions become available.

Making those models available inside a secure, private environment allows our customers to be able to run agents which are increasingly working with more sensitive data, whilst retaining and actually raising the bar on customers’ expectations when it comes to security and privacy for artificial intelligence workloads.

We partnered with OpenAI on model training, and we got a deep partnership and ongoing engineering project with them in building a next-generation harness for fully managed agents.

So we expect that agents are going to run for hours, days, weeks, years, and are going to increasingly become much more like an elastic resource.

Whereas instead of having a large agent which tries to do a bit of everything, we see more and more smaller, composable specialized agents which are fully managed and which have a very, very low marginal cost and which you only pay for as and when you use them.

Overall, what’s the goal with OpenAI?

So bringing together the best of OpenAI’s models, the best of our infrastructure and our experience with developers and fully managed services—we’re bringing all of that together to create a set of fully managed agents which are designed to have very, very low marginal cost and to run for very long periods of time with elastic pricing and composable connectivity.

That enables you to bring lots of different agents together and to have one agent per task, or per employee or per customer or per AWS resource or per transaction.

There really isn’t a limit on how specialized an individual agent can become. So we’re pretty excited about bringing that to more developers going forward.

It’ll be a transformative new way of delivering very long-running agents in an elastic fashion, bringing the best of our knowledge of elasticity with the best of OpenAI’s models to AWS.