{"id":16594,"date":"2025-12-03T10:53:26","date_gmt":"2025-12-03T09:53:26","guid":{"rendered":"https:\/\/haimagazine.com\/uncategorized\/amazons-new-venture-ai-in-companies-is-set-to-pay-off\/"},"modified":"2025-12-08T15:10:50","modified_gmt":"2025-12-08T14:10:50","slug":"amazons-new-venture-ai-in-companies-is-set-to-pay-off","status":"publish","type":"post","link":"https:\/\/haimagazine.com\/en\/ai-news-2\/amazons-new-venture-ai-in-companies-is-set-to-pay-off\/","title":{"rendered":"\ud83d\udd12 Amazon\u2019s new venture \u2013 AI in companies is set to pay off"},"content":{"rendered":"

Amazon didn’t announce the most powerful model on the market. But they didn’t need to. Instead of another “benchmark race,” we got something more fundamental. During re:Invent 2025 <\/mark><\/a>(the annual Amazon Web Services conference), the company unveiled a complete ecosystem where AI is no longer just a chatbot but is starting to serve as a tool for heavy lifting. Which means AI in businesses might finally start to pay off.<\/p>

At the heart of this change<\/a><\/mark> are frontier agents \u2014 new autonomous systems capable of working for hours or days without supervision. Alongside them are four Nova 2 models with multimodal capabilities and a million-token context, the modular Nova Forge platform for training your own models, and Trainium3, an infrastructure designed to remove the “unprofitable technology” label from AI. All this leads to one conclusion: Amazon wants AI to shift from being a cost to being genuinely productive.<\/p>

An AI that doesn’t need hand-holding<\/h4>

The biggest buzz was about the launch of frontier agents \u2014 systems capable of performing complex tasks over extended periods without constant user support. Previous agent tools often required confirmation at every step or would stop working when facing a more complicated task. Amazon is aiming to solve this issue.<\/p>

The presented solutions are:<\/p>

\u2022 Kiro \u2013 an agent that assists developers\u2022 Security Agent \u2013 a security analytics tool\u2022 DevOps Agent \u2013 a system for automating operational processes<\/p>

Each of them uses episodic memory, stable workflows, and mechanisms that allow them to continue tasks despite interruptions or changes in context. According to Matt Garman, CEO of AWS, “it’s crucial to give agents autonomy so they can carry out time-consuming and consistent processes.”<\/p>

Work in difficult conditions<\/h4>

Alongside the agents, Amazon also unveiled the Nova 2 family. It comprises four models with different capabilities but a shared logic: they’re designed to be useful in environments where large, complex data are processed. Specifically:<\/p>

\u2022 Nova 2 Lite and Pro let you adjust the depth of analysis with the extended thinking option\u2022 Nova 2 Omni combines text, image, video and sound, and additionally generates text and images\u2022 Nova 2 Sonic supports natural voice communication<\/p>

Importantly, all models support a context up to a million tokens. This capability allows them to analyze multi-page documents, large code repositories or video recordings. The Amazon AGI team points out<\/mark><\/a> that the Nova 2 models are \u201ctools for operational work, not just for creative tasks,\u201d which is a response to shifting priorities among corporate clients.<\/p>

A solution to failed implementations?<\/h4>

In many organizations, the biggest challenge is that the available models don’t quite fit the company data and processes. Building a model from scratch is pricey, and tweaking a general model can result in losing its core abilities.<\/p>

Nova Forge<\/mark><\/a> is designed to help solve this dilemma. Amazon provides a starting point with partially trained models and allows companies to customize them with their own data in a secure environment. Garman emphasizes that the approach known as “open training” aims to “enable combining the company’s specifics with the skills of the baseline model.”<\/p>

This approach is already used by companies such as Sony, Booking.com, and Reddit, where models support content moderation or the handling of internal processes.<\/p>

It’s supposed to be cheaper<\/h4>

Amazon also has an answer to the problem that model training costs remain one of the most significant constraints for companies. Trainium3<\/mark><\/a> processors are designed to partially solve this problem. This latest generation of specialized processors from Amazon is exclusively for training and deploying large AI models. They offer significantly higher performance and energy efficiency than their predecessors, shortening the training time for large models from several months to a few weeks.<\/p>

AWS reports that in some cases, costs could be slashed by half. Companies like Anthropic, Karakuri and Metagenomi are already taking advantage of these arrangements. Importantly, the high performance is intended to apply to both training and deploying production models.<\/p>

AI Factories for Business<\/h4>

Another element of the announcements is AI Factories, i.e., installations that Amazon deploys directly at customers’ sites. This solution addresses the needs of organizations operating in regulated environments and data sovereignty requirements.<\/p>

As part of AI Factories<\/a><\/mark>, clients receive an integrated infrastructure that includes NVIDIA chips, Trainium, AWS network and Bedrock services. Initial deployments include a project to build an AI zone in Saudi Arabia. Amazon emphasizes that such an architecture is intended to accelerate the adoption of large AI systems in sectors that have until now operated primarily on local resources.<\/p>

Agents under control<\/h4>

Amazon also introduced extensions to Amazon Bedrock AgentCore<\/mark><\/a>, which allow monitoring and controlling the operation of agents. The tools include policies defined in natural language, a set of evaluation metrics, and episodic memory functionality.<\/p>

AWS representatives point out that “transparency in agent decisions is a condition for their safe large-scale deployment.” This allows companies to test agents, analyze how they operate and respond to drops in quality.<\/p>

Implementations that work<\/h4>

AWS emphasizes that the showcased technologies are already proving effective in real-world applications. Lyft is using agents to streamline driver support, reducing ticket resolution time by 87%. Nissan uses AWS to develop software for software-defined vehicles. Visa is working on agent-based payments, while Adobe relies on AWS infrastructure for some of its creative solutions, including Firefly.<\/p>

Each of these examples points to a single direction of development: AI should support operational, analytical and technical tasks, not just generate content. The key is the ability of systems to carry out long, multi-step tasks. Amazon is clearly shifting the emphasis: from the impressive capabilities of models to their practical application in business processes, which\u2014finally\u2014are expected to deliver real economic benefits.<\/p>","protected":false},"excerpt":{"rendered":"

Instead of focusing on model specs, Amazon has unveiled an ecosystem of agents and tools designed to have artificial intelligence do real work, not just spit out responses. And it’s supposed to earn money too.<\/p>\n","protected":false},"author":465,"featured_media":16503,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_lock_modified_date":false,"footnotes":""},"categories":[813],"tags":[],"popular":[],"difficulty-level":[],"ppma_author":[892],"class_list":["post-16594","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news-2"],"acf":[],"authors":[{"term_id":892,"user_id":465,"is_guest":0,"slug":"kmironczuk","display_name":"Krzysztof Miro\u0144czuk","avatar_url":{"url":"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/10\/awatar-2.png","url2x":"https:\/\/haimagazine.com\/wp-content\/uploads\/2025\/10\/awatar-2.png"},"first_name":"Krzysztof","last_name":"Miro\u0144czuk","user_url":"","job_title":"","description":"Od lat zajmuj\u0119 si\u0119 nowymi technologiami w biznesie, edukacji i codziennym \u017cyciu. W centrum mojej uwagi pozostaje cz\u0142owiek \u2013 i to, by technologia wyr\u00f3wnywa\u0142a szanse, zamiast tworzy\u0107 bariery."}],"_links":{"self":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16594","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/users\/465"}],"replies":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/comments?post=16594"}],"version-history":[{"count":1,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16594\/revisions"}],"predecessor-version":[{"id":16595,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/posts\/16594\/revisions\/16595"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media\/16503"}],"wp:attachment":[{"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/media?parent=16594"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/categories?post=16594"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/tags?post=16594"},{"taxonomy":"popular","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/popular?post=16594"},{"taxonomy":"difficulty-level","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/difficulty-level?post=16594"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/haimagazine.com\/en\/wp-json\/wp\/v2\/ppma_author?post=16594"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}