AWS re:Invent 2025
How New Capabilities Unlock Real Customer Outcomes
AWS re:Invent 2025 delivered one of the most transformative sets of announcements in recent years. The event highlighted major advances in generative AI, agentic AI, unified data infrastructure and long-running workflow automation. For organisations aiming to improve operational insight, accelerate product development and automate decision making, these capabilities create opportunities to deliver outcomes that were difficult to achieve even twelve months ago.
This post highlights the announcements that matter most and explains how they can help customers move from experimentation with AI to genuine operational impact. It also reflects on the recent Tiger Data article describing their one-year journey building a unified Postgres data infrastructure on AWS. Their experience shows how a modern data foundation can support real time operational insight and prepare the ground for autonomous, agentic workflows.
Read more on the AWS announcements referenced.
Generative AI: From models to practical value
AWS expanded its model family with the release of Amazon Nova 2 and the introduction of Nova Forge. These services allow organisations to train or refine models using domain specific data and support multimodal inputs such as text, images, audio and video. The result is a set of tools that can be deployed quickly without the overhead of managing model infrastructure.
This matters because most customers are now looking beyond proof of concept work. They want to shorten engineering cycles, improve customer interactions and automate documentation and analysis tasks. The improvements to Amazon Bedrock and the associated tooling deliver on this requirement. Bedrock now provides more robust evaluation, improved guardrails and simplified model deployment patterns. These capabilities allow teams to experiment confidently and then promote successful prototypes into production environments without friction.
Agentic AI: Moving from assistance to autonomous workflows
One of the clearest themes at re:Invent was the shift toward agentic AI. AWS announced new agent frameworks within Amazon Bedrock, including AgentCore and Nova Act, together with more autonomous frontier agents. These frameworks provide memory, policy enforcement, quality controls and orchestration patterns that allow AI systems to handle multi step tasks independently.
This evolution matters because it pushes AI beyond generating answers and into taking action. For example, an operations agent can observe live metrics, identify a performance degradation and initiate a remediation workflow. A support agent can summarise customer interactions, create tickets and suggest solutions. A compliance agent can verify documents and request further details automatically.
AWS also launched S3 Vectors, a high performance vector store integrated directly with S3. This simplifies retrieval augmented generation patterns and ensures agentic systems can access contextual data at scale.
Lambda Durable Functions: Reliable orchestration for long running processes
AWS Lambda now includes Durable Functions, a major enhancement for organisations that run long lived or multi stage processes. Traditional Lambda functions are stateless and short lived. Durable Functions introduce reliable state management, event handling and resumable execution without keeping compute resources active while waiting for external input. Customers only pay for active execution time.
This capability enables a wide range of workflows, including approvals, onboarding flows, lifecycle events, resource provisioning and asynchronous integration across systems. It is particularly powerful when combined with agentic AI. Agents can observe changes in operational data and orchestrate extended workflows using Durable Functions, ensuring predictable delivery across complex processes.
A unified data foundation: The Tiger Data approach
In their article on building unified Postgres infrastructure on AWS, Tiger Data describe how they created a single operational store that handles relational, time series and vector workloads. Their platform supports high ingest throughput, efficient storage and real time analytics. Crucially, it integrates tightly with core AWS services for security, observability, streaming and machine learning.
The significance of this approach is clear. Organisations increasingly generate operational signal at high volume. This includes logs, metrics, sensor data, customer interactions and workflow events. Traditional architectures often require multiple databases to handle these workloads. Migrating, synchronising and querying across these systems introduces delay and limits visibility.
A unified store with efficient time series and vector capabilities solves these challenges. It supports low latency interrogation of historical and live data, giving teams the insight needed to detect anomalies, predict failures and understand operational behaviour. When used alongside agentic AI and services such as Bedrock, S3 Vectors and Durable Functions, it lays the groundwork for closed loop automation.
Read Tiger Data's blog post about their journey with AWS.
How does this help customers?
Below are practical outcomes customers can expect when combining the new AWS capabilities with a unified operational data foundation.
Autonomous incident detection and recovery
Operational data flows into a unified Postgres platform with time series and vector support. Agentic AI monitors this data for anomalies. When required, the agent triggers remediation workflows orchestrated by Durable Functions. Human operators remain in control but rely on the system to catch issues early and execute well understood recovery steps.
Intelligent customer support
Models built or refined using Amazon Nova or Bedrock can summarise interactions, classify issues and propose solutions. Agents can route tickets, trigger workflows or escalate intelligently. Context is enriched with retrieval from S3 Vectors or unified Postgres embeddings.
Real time manufacturing and supply chain visibility
Time series ingestion from machines, sensors and edge devices becomes immediately actionable. AI models identify patterns such as early signs of equipment degradation, throughput changes or safety risks. Workflows can trigger maintenance tasks or adjust production sequencing.
Long lived business processes with reliable automation
Processes such as customer onboarding, compliance checks, identity verification or contract lifecycle management can run for days or weeks. Durable Functions manage the orchestration and state, while agentic AI evaluates data, drafts documents or guides human approvers.
Looking ahead
re:Invent 2025 demonstrates that AI adoption is entering a new phase. Organisations are no longer experimenting with isolated use cases. They are building operational systems that detect, decide and act. The combination of generative and agentic AI, unified data infrastructure and stateful serverless orchestration provides a practical route to automation that delivers measurable value.
As these capabilities mature, customers will be able to move more quickly from insight to action. They will reduce manual effort, improve reliability and unlock innovation across products and services. For teams looking to modernise operations or accelerate their digital roadmap, the announcements at re:Invent represent a significant opportunity.