AWS has dived headfirst into the agentic AI hype cycle, but old tricks will help it chart new waters

Agents, agents, and more agents. It’s the big trend that’s taken the tech industry by storm over the last year, and truth be told, I’m at the point where I’ve covered the topic so much that I’m dealing with a severe case of semantic satiation.

AWS announced a new class of agents at its annual re:Invent conference last week. Known as “frontier agents”, these are more powerful, intuitive, and better equipped to deal with extended periods of operation than the first generation of agents.

From start to finish, the event reeked of hype and hyperbole, with CEO Matt Garman placing agentic AI as on par with the web and cloud computing in terms of its potential long-term impact on businesses.

I’ve no doubt that agentic AI will mark a step change in the evolution of the technology, but I was at least expecting AWS’ old guard reputation in the industry to temper expectations. On that front, I was way off the mark.

As with generative AI, the agentic trend is nowhere near the point of delivering on the bold promises made by major industry providers. An MIT study in August, for example, found enterprises have invested upwards of $35 billion in generative AI projects.

Yet despite these huge investments, they still haven’t reached the productivity-related promised land predicted by big tech.

On the agentic AI front, the situation is roughly the same, with a similar pessimistic outlook on success rates. Analysis from Gartner found that over 40% of projects will be cancelled by 2027, underlining the high stakes involved.

A key factor here, the consultancy noted, lies in “escalating costs, unclear business value, or inadequate risk controls”.

That gives AWS and its industry counterparts roughly two years to drive home the messaging before enterprise IT leaders start wondering where all the money has gone – and that’s if we’re being generous and assuming patience will last.

AWS' old tricks could work with new trends

To AWS’ credit, it has the experience and know-how to potentially deliver on this promise. Indeed, the company is pursuing a similar strategy to the early days of the cloud shift.

Start with the hardware and infrastructure and build from there. In this regard, there’s no denying the company has the capability to deliver: it has strong, deep roots and still holds the title of the largest of the three hyperscalers.

The hyperscaler covered all bases in this regard across the four-day event: from new Trainium chips to Amazon S3 upgrades and an array of AI-related announcements aimed at lowering the barrier to entry for enterprises.

Chief among these was the launch of Amazon Nova, a new system that allows customers to build frontier models. The platform enables customers to combine in-house data with AWS’ open training models, with the result being a bespoke model curated specifically for an enterprise’s individual needs.

This will undoubtedly bridge what has typically been a gap for many enterprises since the advent of generative AI: the lack of financial muscle required to build their own models.

Notably, these custom models will be hosted through Bedrock, the firm’s one-stop shop for in-house and third-party AI tools and a key focus for the company over the last two years.

There is an argument to be made that AWS is building a walled garden here. It obviously wants to keep customers operating within its ecosystem, and platforms like Bedrock and the AgentCore service – a similar setup for sourcing and deploying agents – will help it achieve that.

However, this is less of a walled garden approach akin to the cloud era and more clearly aimed at cultivating an ecosystem that covers all the bases for customers. The company has made no secret of its willingness to simplify integration with other hyperscalers and its relationship with AI providers like Anthropic further highlights this.

Getting projects off the ground

Shiny new tools might get customers excited, but whether enterprises can actually capitalize on these is another question. That's where partners will come in handy for the hyperscaler.

The AWS partner ecosystem is going to play a key role in driving adoption rates, particularly on the agents front. This thriving ecosystem of systems integrators and independent software vendors (ISVs) was critical during the halcyon days of cloud adoption.

As ITPro learned at re:Invent 2025, Rohan Karmarkar, managing director of AWS’ partnership solutions architecture, said partners have been “critical from day one” – the same rules apply with agentic AI.

With 140,000 partners to draw from, AWS customers have no shortage of expertise and support if they’re dabbling in agents. From the hyperscaler’s perspective, this presents a huge opportunity to simplify adoption processes, which are going to be a challenge if it's going to hit the ground running with this strategy.

Tangible business use-cases

One recurring gripe enterprises and industry stakeholders alike have had with generative AI has been the pie-in-the-sky claims from providers.

We’ve all heard the talking points about productivity improvements and efficiency gains, but the reality is that there simply hasn’t been enough concrete evidence to support the idea that the technology will be the game changer it’s framed as.

Agentic AI has - and will continue to - suffer the same problem. In this sense, AWS made a big statement in highlighting its own internal use of agents - and so far the signs are promising.

As AWS CISO Amy Herzog told ITPro, the company has been using agents in security operations for some time now, enabling teams to drastically reduce manual toil and speed up incident response capabilities.

During the same panel session where Herzog touted the technology’s potential, Hart Rossman, VP, Office of the CISO, also detailed the company’s success with an internal “security response agent”.

AWS’ reputation as a reliable, no-nonsense hyperscaler precedes it. If the company draws on its cloud credentials to push agentic AI and continues showcasing tangible business use-cases, then it might finally stake its claim as the go-to AI shop after what was a rocky start in the early days of the generative AI race.