Want smart insight into your inbox? Enterprise AI, only what matters to data and security leaders, sign up for our weekly newspapers. Subscribe now
Cloud veteran Amazon web services (AWS) Assuming AI agents will change how we all work and interact with information, and the enterprises require a platform that allows them to create and deploy agents on a scale – all at the same place.
Today at its New York Summit, AWS unveiled Amazon bedock agentcore, A new enterprise-grade platform that is designed to manufacture, deploy and operate AI agents safely and on a scale.
Swami Shivsubramanian, Vice President of Agentic AI, said that during the keynote speaker, the agentcore “helps the organizations to move beyond the experiments of the production-taiyar agent system that can be trusted with your most important business processes.”
Agentcore is a modular stack of services – available in preview – provides core infrastructure required to develop developers from prototype to production from prototypes, including runtime, memory, identity, observation, API integration and web browsing and codes.
AI Impact series returns to San Francisco – 5 August
The next phase of AI is here – are you ready? Leaders of Block, GSK and SAP include how autonomous agents are re-shaping the enterprise workflows-from the decision making of time-to-end and automation.
Now secure your location – space is limited:
“We believe that agents are going fundamentally to change how we use equipment and internet,” said Dipk Singh, Vice President of Database and AI. “The line between an agent and an application is getting blurred.”
Agentcore manufactures on the existing bedock agents framework, which was launched in late 2024, but any agent expands dramatic abilities by supporting the framework or foundation model by supporting the Foundation model – not only hosted within the bedrock.
He is also involved Compatibility with Open-SOS toolkit such as Cravei, Langchen, Lamanadex, Laminadex, Langgraph and AWS’s strands agents SDK.
What is involved in AWS bedock agentcore
- Agentcore runtime: A server-free, low-sequelment execution environment that supports long-lasting sessions with multimodal workload and session isolation.
- Agentcore memory: Long and short -term memory services that allow agents to learn from previous interaction and maintain relevant knowledge in sessions.
- Agentcore identification: Oauth-based identity and access management allows agents to function on behalf of users in systems such as Github, Slack, or Slesforce.
- Agentcore observation: Support for Opentalemetry, Langsmith and Detaidogue with built -in dashboard, debugging, and telemetry tools.
- Agentcore gateway: The model converts internal API, lambda tasks and third-party services into agent-compatible devices using reference protocol (MCP).
- Agentcore browser: The headless browser provides access to the agents to interact with websites.
- Agentcore code interpreter: A safe environment to execute the code generated by agents for analysis and visualization.
Agentcore AWS also integrates with marketplace, enables teams to discover and deploy pre-made agents and equipment.
According to Singh, Agentcore has been designed keeping in mind the interoperability. It supports emerging industry standards such as MCP and Google’s agent -2 -agent (A2A) protocol. Ensure features such as agentcore identity and gateway ensure that agents have clear permission and can safely interact with internal systems and third-party APIs.
The launch of AWS has included it in the center of becoming one of the most competitive segments in Enterprise AI.
Openai agent SDK and Gemini-based agent SDK of Google Both end-to-end agents are carrying forward the same vision as development platforms.
The author’s AI headquarters and startups such as cognition (manufacturer of Dewin) are also making equipment for the management of autonomous software agents.
“Agents are the most impressive changes we have seen at age,” said Shivsubramanian. “There is a change for service as a software with agents. It is a tectonic change of how software is made, deployed and operated.”
Customer adoption and initial use cases
Many companies have provided early access to the agentcore, already producing production-grade agent applications in industries including finance, healthcare, marketing and material management.
Cloud document and file storage company is looking for ways to expand its content management tool using box strands agents and bedock agent runtime.
CTO Ben Kus stated that integration box gives customers “top level safety and compliance”, while enterprises increase AI abilities in the environment.
Etau Unibnco, Brazil, has used Agentcore to support the development of hyper-parsnel, safe digital banking experiences. Chief Technology Officer Carlos Eduardo Mazzi said the new platform “will help us to provide a spontaneous banking experience with the efficiency of automation and privatization customers.”
In the Healthcare Space, Innovasar has created a new protocol – HMCP (Healthcare Model Reference Protocol) -on top of a agentcore gateway. CEO and co-founder Abhinav Shashank called the gateway a “game-changer”, which allows the company to convert the existing API into agent-compatible equipment while maintaining the trust, compliance and operational efficiency.
The marketing firm is taking advantage of agentcore to accelerate the time of construction of the Epsilon campaign and improve the engagement. SVP of Software Engineering, Prashant Ethota said that the company hopes that construction time will reduce the time by 30% and the customer will increase the travel privatization.
Availability and pricing
Agentcore is now available in previews in selected AWS areas including US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), and Europe (Frankfurt). It is free to try by 16 September, 2025, then with pricing.
The pricing for agentcore is completely consumed-based, with no upfront commitment or minimum fee. Each module- precious, memory, identity, observation, gateway, browser, and code interpreter- is bills independently and can be used together or used together.
The price of runtime, browser, and code interpreter services is priced per second, based on CPU and memory use, with $ 0.0895 per VCPU-hour and $ 0.00945 per GB-hour rates.
The Gateway fees $ 0.005 per 1,000 tool API invocation, $ 0.025 per 1,000 search query, and a 100 tool per month per month.
Memory cost is based on data volume: $ 0.25 per 1,000 short -term memory event, $ 0.75 per 1,000 long -term memories stored (or $ 0.25 with custom strategies), and $ 0.50 per 1,000 repayment.
The cost of agentcore identification is $ 0.010 per 1,000 tokens or API costs the cost of major requests, although it does not include no additional fees when used via runtime or gateway.
The observation is bills through Amazon Cloudwatch rates.
To learn or start, AWS directs developers for this Agentcore documentation, Github samples, and a dedicated Discord server,
