Replay and Pause AWS Lambda Functions with Kiro

AWS launched Agent Plugins yesterday. Production-ready. You type "deploy to AWS" and get a complete CI/CD pipeline with architecture diagrams. Bedrock AgentCore now handles memory, identity, and tool integrations as standard cloud services. The infrastructure layer for AI agents is settling. We're moving from "experiment with AI" to "run AI as operations." Three things changed: 𝟭. 𝗗𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁 𝗹𝗼𝗼𝗸𝘀 𝗹𝗶𝗸𝗲 𝗗𝗲𝘃𝗢𝗽𝘀 𝗻𝗼𝘄 CloudFormation templates for agent architectures. IAM policies for agent permissions. Familiar patterns for infrastructure teams. 𝟮. 𝗠𝗲𝗺𝗼𝗿𝘆 𝗶𝘀 𝗮 𝗺𝗮𝗻𝗮𝗴𝗲𝗱 𝘀𝗲𝗿𝘃𝗶𝗰𝗲 Vector databases and conversation persistence come as configuration options now. Less plumbing to maintain. 𝟯. 𝗧𝗼𝗼𝗹 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻𝘀 𝘀𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝗶𝘇𝗲 The deploy-on-aws plugin generates pipelines automatically. Third-party tools get consistent interfaces. At LFG Labs, we've deployed agents on VPS, Mac Mini fleets, and cloud instances. (We've seen the full spectrum from "works great" to "why is this so expensive?") AWS's move validates what we've observed: agent infrastructure is maturing fast. But managed services have costs that creep. One client's bill grew 4x in 90 days because they didn't model token volume correctly. AWS lowered the barrier to entry. That's great for experimentation. For production at scale, you still need to pick your infrastructure deliberately. Start with the problem. Match the tool. Then scale.

🚀 Just shipped: TravelEase: AI-Powered Serverless Contact Form on AWS I recently built a production-ready serverless application that solves the following business problem: travel agencies losing customer inquiries through unreliable email links with no tracking, no confirmation and no way to prioritize leads. What the solution does: → Captures customer travel inquiries via a CloudFront-hosted HTTPS form → Generates AI-powered insights using the Anthropic Claude API; giving TravelEase employees instant context on every submission before they respond. → Persists data to DynamoDB and sends automated confirmations via Amazon SES. All within ~3 seconds, at ~$0.50/month What I built:  → Serverless architecture on AWS (Lambda, API Gateway, DynamoDB, SES, Secrets Manager, CloudFront) → 100% Infrastructure as Code using Terraform → CI/CD pipeline with GitHub Actions using OIDC authentication = no static credentials stored anywhere → Human-in-the-loop deployment gates means that Terraform plan is reviewed on every Pull Request before apply runs → Active monitoring with CloudWatch Alarms and SNS for real-time alerting → Defense in depth security across every layer of the architecture What this means: Serverless solutions reduce operational overhead so TravelEase can focus entirely on what matters most; the customer relationship. AWS managed services handle the scalability behind the scenes, and the Claude AI integration gives employees instant context on every inquiry, helping them respond faster and more meaningfully. AI doesn't replace the human connection here, it strengthens it. Documentation on Medium: https://lnkd.in/eCAK9PpF GitHub Repository: https://lnkd.in/eGVgYjYc

I recently took some time to explore AWS Lambda Durable Functions and it genuinely changes how I think about serverless workflows. For a long time, if you wanted to build multi-step, stateful workflows in AWS, your go-to was Step Functions. But Durable Functions introduce a different model: Write your workflow as code inside Lambda itself. The biggest shift is state management. Durable Functions are regular Lambda functions enhanced with stateful execution capabilities! Durable functions automatically checkpoint progress, suspend execution for up to one year during long-running tasks, and recover from failures. Instead of managing state externally, or defining workflows in JSON for your Step Functions, durable Functions let you: Resume exactly where you left off which is a big deal for real-world systems snd Pause execution for events without paying for idle compute From what I tested, this fits really well for: • Approval workflows (human intervention) • Async event-driven pipelines • Multi-step business logic (orders, onboarding, etc.) • AI/ML workflows that pause between stages Basically, anywhere your workflow feels more like application logic than infrastructure orchestration. This isn’t a replacement for Step Functions, It’s a shift in abstraction. AWS is basically saying: “You can now treat serverless workflows as code, not just configuration.” #AWS #Serverless #Lambda #StepFunctions #CloudEngineering #DevOps #Observability

🏗️ Moving an AI Image App to AWS: The Infrastructure Setup Getting an app to run is one thing. Making it secure, automated, and repeatable is where the real work happens. Here's the infrastructure breakdown for my containerised AI SaaS deployment on AWS: 🌐 Network isolation Custom VPC with a clear split, the app lives in private subnets with zero direct internet access. The Application Load Balancer is the only public entry point. The containers are never reachable from outside. 📦 Container optimisation Multi-stage Docker builds strip everything except what the app needs to run. → Final image is 90% smaller. Faster deploys, lower ECR storage costs. ⚙️ Infrastructure as Code Every resource defined in Terraform - VPC, ECS, ALB, IAM, ECR, Auto scaling. Remote state in S3 with DynamoDB locking so CI/CD pipelines never step on each other. Tear it down, rebuild it, same result every time. 🔑 Secrets management AWS SSM Parameter Store holds everything encrypted, injected directly into containers at startup. Application code just sees environment variables. 🚀 Automated CI/CD GitHub Actions handles the full lifecycle. A simple code push triggers a build, an ECR push, and an ECS Fargate update with zero downtime. The AI app is the engine. Building the infrastructure around it was the most rewarding part of this project. Check out the code and the full walkthrough here: 🔗 GitHub: https://lnkd.in/dGZv3h4c 🔗 Blog: https://lnkd.in/dYw6CV8h #AWS #Terraform #DevOps #ECSFargate #Docker #CloudComputing #GitHubActions #IaC

🚀 AWS just made building resilient, long-running applications even easier — introducing the Lambda Durable Functions Kiro Power! If you've been working with Lambda durable functions, you know the complexity involved: replay model best practices, step and wait operations, concurrent execution patterns, error handling with retry strategies and compensating transactions... it's a lot to keep in mind. Now, with the new Kiro Power for Lambda Durable Functions, an AI agent dynamically loads all that expertise directly into your local development environment as you code. No more context switching between docs and your IDE. 🧠 What the AI agent brings to your workflow: Replay model best practices Step and wait operations Concurrent execution with map and parallel patterns Error handling: retry strategies & compensating transactions Testing patterns Deployment with CloudFormation, AWS CDK, and AWS SAM 💡 Real-world use cases it accelerates: - Order processing pipelines - AI agent orchestration with human-in-the-loop approvals - Payment coordination workflows The power is available today with one-click installation from the Kiro IDE and the Kiro powers page — and the source is open on GitHub. If you're building multi-step, long-running applications or AI workflows on AWS, this is a must-try. The gap between "idea" and "working durable function" just got a lot smaller. 👉 Get started: https://lnkd.in/dBqtMy7S #AWS #Lambda #DurableFunctions #Kiro #ServerlessComputing #AIAgents #CloudDevelopment #AWSLambda #GenerativeAI #DevTools

“Cloud is no longer just infrastructure , it’s becoming the backbone of intelligent systems.” Attended the “Best of AWS re:Invent” session recently, and it truly shifted the way I think about building backend systems. What stood out was the scale of innovation coming out of AWS re:Invent and how cloud today is about designing intelligent, scalable, and resilient applications and not just managing infrastructure. 🚀 Key takeaways I’m carrying forward: • Serverless and event-driven architectures simplify scaling while reducing operational overhead. • Modern applications rely on distributed systems and managed services to stay efficient. • AI/ML is becoming an integral part of everyday applications. • Strong architecture patterns balance performance, cost, and reliability. What I found most valuable is how everything connects as an ecosystem,it really feels like an ocean of possibilities with endless ways to optimize and build smarter systems. Curious to hear-what’s one AWS concept that changed how you build applications? #AWS #AWSReinvent #CloudComputing #SoftwareEngineering #BackendDevelopment #Java #FullStackDevelopment #LearningJourney #TechGrowth #CloudArchitecture

Being able to run shell commands directly from an Amazon Bedrock AgentCore Runtime session is exactly the kind of quality-of-life upgrade AI agents needed. The new InvokeAgentRuntimeCommand API lets you execute commands in the same container and environment as your agent, stream the output over HTTP/2, and get the exit code back - without bolting on your own command runner. This makes it way easier to mix LLM reasoning with deterministic tasks like tests, dependency installs, or git operations, and it’s already available across fourteen AWS Regions. Credit to AWS for shipping a very practical upgrade here - worth a read if you’re building agentic workflows on Bedrock. #AmazonBedrock #AgentCore #AIAgents #AWS #DevTools

I’ve put together a new repository called agentcore-cloudwatch-agent: https://lnkd.in/eZHCJJWp, aimed at simplifying log analysis in CloudWatch. It’s a small step towards making log monitoring more efficient and accessible. If you’re interested, feel free to check it out and share your thoughts. If you want to hear more about don't forget to follow the AI AWS Community we're going to launch the next week: https://lnkd.in/ewU-PyfB #Cloudwatch #AWS #DevOps #LogAnalysis #OpenSource #Societech #EPAM #AIAWSUGColombia

For workflow-based processing on AWS we now have 2 good choices of services - Lambda Durable Functions and Step Functions. AWS has just announced integrations for more AWS services with Step Functions including Bedrock AgentCore and S3 Vectors which sound really interesting to me. Being able to invoke agent runtimes with AgentCore including built-in retries, running agents in parallel via Map states, and manage provisioning as workflow steps removes a lot of glue code. The S3 Vectors and Lambda durable execution support round it out nicely. There are now over 1,100 new API actions you can orchestrate directly from your workflows using the GUI builder or code they offer. Useful for anyone building AI agent pipelines or distributed apps on AWS. https://lnkd.in/exENRnd9

🚀 🚀 Automated the one-time AWS account infrastructure setup. Created Terraform Script to deploy (NLB, ALBs, VPC Links, API Gateway, custom domains, domain name associations, ACM certificates, and Route53 records) all provisioned through a single CI/CD pipeline with GitHub Actions. What used to be hours of manual console work across multiple AWS services is now a repeatable, version-controlled git push. Infrastructure as Code isn't optional anymore. It's how you eliminate Data and Infrastructure drift, reduce human error, and onboard new accounts in minutes instead of days. #Terraform #AWS #InfrastructureAsCode #Automation 🔥 🔥