AWS accelerates its bet on autonomous agents in the cloud

Amazon Web Services has made a move to consolidate itself as a leader in autonomous agents on its cloudCombining new software services with proprietary hardware designed to scale enterprise artificial intelligence. At re:Invent 2025, the company It has presented a series of announcements that aim to enable any organization to deploy thousands or even millions of agents capable of operating continuously. on AWS.

This strategic shift relegates the mere conversation about generative models to the background and moves it towards a action-oriented agentic AISystems that plan, decide, and execute complex tasks with minimal supervision. For companies in Spain and Europe, where regulation and data protection are paramount, AWS's proposal relies on... fine-tuned security controls, governance, and energy efficiencykey aspects for being able to adopt these agents on a large scale.

A new generation of autonomous agents on AWS

At the conference held in Las Vegas, AWS defined Agentic AI as the next big step for the industry: AI agents capable of dynamic reasoning, operating for hours or days and coordinate complex tasks without the need for constant rescheduling. The company's thesis is that, in the future, Each company will have billions of internal agents covering almost any imaginable function.

These systems differ from traditional assistants because They don't just generate text or codebut also They plan workflows, orchestrate external tools, and make decisions in changing environments. For many European organizations, this approach opens the door to automating everything from customer service processes to back-office tasks, provided that strict control is maintained over risks, compliance, and privacy.

According to AWS, the Agentic AI market could skyrocket in the next decade, with forecasts already placing its value at hundreds of billions of dollarsThe company insists that its goal is to "democratize" access to these agents, allowing them to SMEs and large corporations so they can use them without needing to build their own high-cost infrastructure.

This approach is particularly relevant for regulated European sectors, such as banking, insurance, healthcare, or public administration, where automation requires traceability, clear policies, and human oversight that can be audited by regulators.

Amazon Bedrock AgentCore: the nerve center of corporate agents

The key element of AWS's approach is Amazon Bedrock AgentCore, its platform for design, deploy and govern AI agents in enterprise environments. AgentCore is conceived as an intermediary layer that connects models, corporate data, and business tools with control and safety mechanisms designed for production.

One of the main advances is Policy, available in preview, which allows teams to define limits of action using natural languageInstead of writing complex technical rules, a manager can specify, for example, that an agent Do not approve returns exceeding a certain amount without human review, or that does not access certain repositories of sensitive data.

These policies integrate with AgentCore Gateway to automatically block actions that violate the guidelinesacting as a security layer that prevents unauthorized operations with systems like Salesforce, Slack, or other critical applications. For European companies with obligations under the GDPR or the future EU AI regulation, this type of granular and auditable control It is an important piece for mitigating legal risks.

Another notable novelty is AgentCore Memory, which equips the agents with a episodic contextual memoryThis function allows systems to remember relevant information from each user or use case—such as travel preferences, project context, or past incidents—to make better decisions in the future, without having to reconfigure themselves in each interaction.

In parallel, AgentCore Evaluations It introduces 13 pre-configured evaluators that measure dimensions such as security, accuracy, proper use of tools, or quality of responsesThanks to this continuous monitoring, teams can detect performance drops or potential behavioral deviations and adjust agents without having to create their own evaluation systems from scratch.

Frontier agents: Kiro, Security Agent and DevOps Agent as new teammates

Building on AgentCore, AWS has launched a new class of agents called border agentsdesigned to function as virtual members of the development, security, and operations teamsThe idea is that they cease to be one-off tools and become permanent components of the software life cycle.

The first is Kiro Autonomous AgentKiro is geared towards software development. Unlike more basic code assistants, Kiro takes a more advanced approach. “spec-driven development”Before writing code, the agent Generates requirements, technical documentation, and work plans detailed, reducing improvisation and design errors.

Kiro can generate, update and maintain complete codebasesThis includes documentation and unit testing, maintaining persistent context across sessions, and learning from pull requests and developer feedback. This allows you to address issues from From bug classification to changes affecting multiple repositoriesalways presenting their proposals as edits or pull requests that the team can review.

For tech startups and European growth-stage companies, this type of agent shows promise. shorten delivery cycles and free developers from repetitive tasksHowever, adoption will require reviewing internal processes, risks of technological dependence, and policies on AI-generated code.

The second member of the family is AWS Security Agent, conceived as a virtual security engineerThis agent reviews architecture documents, analyzes pull requests, and evaluates applications against internal security standards and known vulnerabilities, helping to prioritize the risks that truly affect the business instead of generating endless lists of generic notices.

AWS Security Agent also transforms penetration testing into an on-demand service, which can to be executed more frequently and at a lower cost than traditional manual testing. The findings include remediation code proposals, which facilitate quickly correcting detected problems, something especially critical in regulated environments such as European banking or fintech.

The third pillar is AWS DevOps Agentfocused on operational excellence. This agent is "on call" when incidents occur, using data from tools such as Amazon CloudWatch, Dynatrace, Datadog, New Relic or Splunk, along with runbooks and code repositories, to pinpoint the root cause of problems.

In addition to reacting to incidents, AWS DevOps Agent analyzes historical failure patterns It offers recommendations for improving observability, optimizing infrastructure, strengthening deployment pipelines, and increasing application resilience. Within Amazon, this approach has already managed thousands of internal escalations, with a root cause identification rate that the company says exceeds 80%.

Trainium3 infrastructure and the path towards Trainium4 to power autonomous agents

AWS's commitment to autonomous agents is also supported by a major infrastructure overhaul. The company has unveiled the Trainium3 chip and Trainium3 UltraServers, specifically designed for train and run large AI models with lower energy consumption.

Trainium3 is made with 3 nanometer technology and integrates into servers capable of grouping up to 144 chips in a single unitAccording to AWS, these UltraServers offer more than four times the speed and four times the memory compared to the previous generation, as well as a 40% higher energy efficiency, a key factor in containing electricity costs in data centers.

Architecture allows connection thousands of UltraServers on a network to achieve configurations with up to one million Trainium3 chips working togetherThis capability is designed for organizations that need to train frontier models and deploy high-volume agents, something that could be especially useful for large European providers of digital services, banking, or telecommunications.

Among the first customers who have already tested Trainium3 are Anthropic, LLM Karakuri, SplashMusic or DecartThese companies have sought to reduce inference costs and accelerate training times. Although these cases are primarily concentrated in the United States, AWS's strategy involves bringing these capabilities to global customers as well, including those in Europe.

In the longer term, AWS has confirmed that Trainium4 is already in developmentThis next generation promises substantial improvements in computing performance—with multiplied growth in FP4 and FP8—and higher memory bandwidth for the next wave of models and agents. A relevant aspect is their Expected compatibility with Nvidia NVLink FusionThis should make it easier to combine Nvidia GPUs with Trainium chips in the same infrastructure.

This interoperability is aimed at attracting developers who work with CUDA and Nvidia ecosystemsallowing them to deploy applications already optimized for these GPUs on a hybrid infrastructure that combines Amazon and third-party hardware, potentially reducing costs without losing access to established libraries and tools.

Enterprise AI ecosystem, partners, and model expansion

To bolster the deployment of its autonomous agents, AWS is expanding its ecosystem of partners and complementary servicesIn its AWS AI Competency Partners program, the company has introduced new categories focused on agentic AI that recognize providers specializing in autonomous solutions at enterprise scale.

The digital catalog AWS Marketplace It also incorporates AI-based innovations, such as a agent mode for conversational searches, express private offers to automate price negotiation and multi-product solutions that group services from different providers, including AI agents ready for deployment.

In the area of ​​customer experience, Amazon Connect adds 29 new features that rely on autonomous agents to offer automated voice, real-time assistance, and predictive analytics. This type of capability is especially relevant for call centers and customer service providers distributed across Europe that are looking to reduce waiting times and improve service quality without increasing the workforce at the same rate.

In addition, AWS has incorporated 18 new open weight models on Amazon Bedrock...in what it describes as the largest expansion of models to date. These include: Mistral Large 3 and Ministral 3 from Mistral AI —a European company with a strong presence in the EU—, as well as Google's Gemma 3, MiniMax's M2, Nvidia's Nemotron, and OpenAI's GPT OSS SafeguardAmong others. This range allows companies to choose the model that best suits their needs, compliance requirements, and data sovereignty preferences.

For customers who need dedicated infrastructure, the AWS AI Factories They offer AI deployments in their own data centers, combining Nvidia GPUs, Trainium chips, and services such as Amazon Bedrock vs Amazon SageMaker AIAlthough these solutions are designed for large organizations, they may be attractive to European entities with strong regulatory or data residency restrictions.

Security, governance and corporate adoption of agents in Europe

Beyond technical capabilities, AWS is trying to respond to the security and compliance concerns which accompany the deployment of autonomous agents. In this area, it is already generally available AWS Security Hub, which unifies signals from services such as GuardDuty, Amazon Inspector or Amazon Macie to offer near real-time risk analytics and coordinate cloud security operations.

The solution Amazon GuardDuty Extended Threat Detection expands its scope to Amazon EC2 and Amazon ECSproviding a broader view of sophisticated attack sequences and facilitating faster remediation. This type of tool aligns with the goals of many European companies automate part of the incident response without losing the traceability required by regulators and audits.

At the same time, AWS insists that its agents do not replace human oversight, but rather act as extension of existing equipmentFrontier agents are conceived as shared resources that learn from each organization's context, adapting to its standards of quality, security, and compliance—something especially sensitive in markets like Spain, where SMEs tend to have limited security and DevOps resources.

The strategic collaborations that AWS has signed with global companies—such as BlackRock, Nissan, Sony, Adobe or Visa—reinforcing their message that autonomous agents can be integrated into large-scale critical operations. Although many of these deals have been announced in other markets, it is expected that Its effects extend to subsidiaries and operations in Europe, accelerating the adoption of similar architectures in local companies.

For European businesses, a key issue will be how to balance the benefits in productivity and speed of deployment with the demands of the new EU AI regulations, which will require impact assessments, transparency and risk management in systems that make automated decisions with significant effects on people.

With this combination of new frontier agents, advanced capabilities in Amazon Bedrock AgentCore, and an infrastructure reinforced with Trainium3—and the future Trainium4—AWS is trying to position itself as a reference platform for build, govern, and scale autonomous agents in the cloud. For companies in Spain and the rest of Europe, the key will be to assess whether this ecosystem allows them to accelerate their digital transformation without losing sight of the security, compliance and efficiency requirements that define the current regulatory and economic context.

Related article:

AI for freelancers and SMEs: all the processes you can automate without knowing how to program

Alberto Navarro

I am a technology enthusiast who has turned his "geek" interests into a profession. I have spent more than 10 years of my life using cutting-edge technology and tinkering with all kinds of programs out of pure curiosity. Now I have specialized in computer technology and video games. This is because for more than 5 years I have been writing for various websites on technology and video games, creating articles that seek to give you the information you need in a language that is understandable to everyone.

If you have any questions, my knowledge ranges from everything related to the Windows operating system as well as Android for mobile phones. And my commitment is to you, I am always willing to spend a few minutes and help you resolve any questions you may have in this internet world.