Skip to main content

The 2026 AWS application development stack

Build new applications on AWS
without picking the wrong service

AWS ships a new service every other Tuesday and quietly retires three the same year. This is the opinionated 2026 shortlist we actually use on FactualMinds engagements — grouped by where it fits in the application lifecycle, with the deprecated services we tell clients to avoid.

Modern AWS application development services overview

What changed since 2020

Most "best AWS services for application development" pages on the internet are five years old. They still recommend Honeycode (discontinued February 2024), CodeWhisperer (renamed to Amazon Q Developer in April 2024), Cloud9 and CodeCommit and CodeStar (all closed to new customers July 2024), App Mesh (deprecating September 2026), and Graviton2 (superseded by Graviton4). Following those guides today produces a stack that is already three retirements behind.

The 2026 stack, by lifecycle phase

Five categories cover most application work. Each card lists the services we actually reach for, with links to deeper FactualMinds writing on each.

Code & assist

Amazon Q Developer for inline completions, agents, and Java/.NET code transformation. AWS CodeArtifact for private package registries. Amazon CodeCatalyst for unified projects, dev environments, and pipelines. See our Q Developer service →

Infrastructure as code

AWS CDK (TypeScript / Python / Go / Java / .NET) for typed infrastructure with real abstractions. AWS SAM for Lambda-first stacks. AWS Infrastructure Composer (the renamed Application Composer) when a designer needs to draft architecture visually. CDK vs. Composer deep-dive →

Build, deploy & host

AWS App Runner for containerized web apps with zero infra — auto-scales to zero, supports VPC connectivity, bills only while handling requests. AWS Amplify Gen 2 for TypeScript fullstack with typed end-to-end data. ECS or EKS on Fargate for serious container workloads. CodePipeline + CodeBuild + CodeDeploy when you need first-party CI/CD. CodePipeline patterns →

Run, scale & integrate

Lambda + SnapStart for sub-second Java/Python/.NET cold starts. Graviton4 (M8g, C8g, R8g) for 40% better price-performance. Step Functions Distributed Map and JSONata for orchestrating millions of items. EventBridge Pipes for filter/enrich/route. AppSync Events for serverless WebSockets. VPC Lattice for service-to-service connectivity. Bedrock AgentCore for production agents. Step Functions → · EventBridge →

Observe & secure

CloudWatch Application Signals for SLO-aware OpenTelemetry tracing without instrumenting yourself. AWS X-Ray for cross-service traces. Amazon CodeGuru Profiler for CPU and memory hotspots; Amazon Q Developer code reviews for the security and quality findings that used to live in CodeGuru Reviewer. Amazon Inspector for continuous CVE scanning of EC2, Lambda, and ECR. IAM Access Analyzer with unused-access findings. Security services →

What to use when

Six of the most common decisions teams hit on day one of a new build. Save them an hour.

You need a typed fullstack with auth, data, and frontend hosting in one place

Use AWS Amplify Gen 2. The TypeScript-first developer experience, sandbox environments per branch, and direct integration with Cognito, AppSync, and DynamoDB get you from zero to production faster than wiring services yourself.

You have a Dockerfile and want it on the internet by lunch

Use AWS App Runner. No clusters, no load balancers, no task definitions. It auto-scales to zero, supports VPC connectivity, and bills per request. The right answer for 90% of internal tools and most early-stage SaaS APIs.

You are running workloads that need pod-level control or service mesh-style routing

Use Amazon EKS on Fargate with VPC Lattice for service connectivity. ECS vs EKS comparison. Skip App Mesh — it is being deprecated in September 2026.

You are building event-driven systems

EventBridge Pipes handles filter / enrich / transform / route in one resource. Step Functions Distributed Map handles parallel iteration over millions of records. SQS handles work queues. Use SNS only when you genuinely need fan-out to multiple subscriber types.

You are building agents or RAG features

Use Bedrock AgentCore with Knowledge Bases. AgentCore handles memory, tool routing, and observability so you stop writing those yourself. Pair it with S3 Vectors when you need cheap vector storage. Modern AWS AI services →

You want first-party CI/CD without leaving AWS

Use CodeCatalyst for greenfield projects (unified source, builds, environments, issues). Use CodePipeline + CodeBuild + CodeDeploy when you already have GitHub or GitLab and need orchestration into AWS deploy targets.

Services we no longer recommend in 2026

Each of these is still searchable, still indexed by competitor blog posts, and still shows up in old reference architectures. None of them belong in a new build. Here is what to use instead.

Don't pick this Why Use this instead
Amazon Honeycode Discontinued February 2024. App Runner + Amplify Gen 2 for internal tools, or Bedrock-driven workflows.
AWS CodeCommit Closed to new customers July 2024. GitHub or GitLab as source of truth, with CodePipeline or CodeCatalyst.
AWS CodeStar Deprecated July 2024. Amazon CodeCatalyst (the supported successor).
Amazon CodeWhisperer Renamed Amazon Q Developer in April 2024. Amazon Q Developer (same product, materially expanded capabilities).
AWS Cloud9 Closed to new customers July 2024. Q Developer in JetBrains or VS Code, or CodeCatalyst Dev Environments.
AWS App Mesh End-of-support September 30, 2026. VPC Lattice for cross-VPC; ECS Service Connect for in-cluster.
EC2 Graviton2 (M6g, C6g, R6g) Two generations behind; comparable list price. Graviton4 (M8g, C8g, R8g) — up to 40% better price-performance.

Already running one of these in production? See our application modernization service — we migrate off deprecated services without big-bang rewrites.

Frequently asked questions

What CTOs and engineering leaders ask us before kicking off a new AWS build.

Which AWS services do you actually need to start a new app in 2026?
Far fewer than the 200+ AWS publishes. For a typical greenfield SaaS in 2026, the short list is: an account baseline (Control Tower, IAM Identity Center), a runtime (App Runner or Lambda or ECS Fargate), a database (RDS for Postgres or DynamoDB), an API tier (API Gateway HTTP APIs or AppSync), an observability layer (CloudWatch Application Signals + X-Ray), and a CI/CD path (CodeCatalyst or CodePipeline). Everything else is added when the workload demands it.
Is CodeCommit deprecated? What should we use instead?
CodeCommit closed to new customers in July 2024. Existing customers can keep using it, but new accounts cannot enable it. We recommend GitHub or GitLab as the source of truth, with CodePipeline or CodeCatalyst orchestrating builds and deployments into AWS. CodeStar and Cloud9 closed at the same time and have the same answer: use Q Developer in JetBrains/VS Code, and use CodeCatalyst Dev Environments when you need cloud dev boxes.
Should we use Amazon Q Developer or GitHub Copilot?
Both work. Q Developer has a sharper edge for AWS-specific workflows: it knows your account, can /transform Java applications between versions, and integrates directly with the AWS console. Copilot is more language-agnostic and tends to win on raw IDE polish. Most teams we work with use both — Copilot for general code and Q for AWS-shaped tasks. We have a longer comparison in our <a href="/blog/amazon-q-vs-github-copilot-2026/" class="text-primary font-medium hover:underline">2026 head-to-head post</a>.
App Runner vs. Amplify Gen 2 vs. ECS Fargate — how do we pick?
App Runner: you have a Dockerfile and want a public URL with no orchestration. Amplify Gen 2: you are a TypeScript team building a fullstack app and want auth, data, and hosting wired together. ECS Fargate: you have multiple services with custom networking, sidecars, or per-task IAM roles. EKS Fargate is the same conversation as ECS Fargate but for teams that already speak Kubernetes or need pod-level isolation that ECS does not give you.
Is AWS App Mesh dead? What replaces it?
AWS announced App Mesh end-of-support for September 30, 2026. New service-to-service connectivity belongs on VPC Lattice (cross-VPC, cross-account, identity-aware) or ECS Service Connect (in-cluster, simpler model). We migrate App Mesh customers off proactively rather than waiting for the deadline.
Do we still need Application Composer if we already use the CDK?
Mostly no. Infrastructure Composer (the new name for Application Composer) is most useful as a visualization layer for existing CloudFormation/SAM templates and as a starting point for non-CDK teams. If your team is already productive in the CDK with proper testing and CI gates, Composer adds little. Where it shines: onboarding a new engineer, sketching an architecture in a meeting, or generating starter SAM templates that you'll then iterate on.
How do these services map to a typical FactualMinds engagement?
A new application engagement usually starts with a 1-week architecture sprint that picks the runtime (App Runner, Lambda, or ECS Fargate), the data tier (RDS Postgres or DynamoDB), the API contract (REST via API Gateway or GraphQL via AppSync), and the CI/CD path. Implementation runs in 2-week iterations with a working deploy at the end of each. We use the CDK for everything we own, GitHub Actions or CodePipeline for CI, and CloudWatch Application Signals for SLO-aware observability from day one. <a href="/services/">Browse our services</a> or <a href="/contact-us/">book an architecture sprint</a>.

Pick the right AWS services for your next build

Book a free 30-minute architecture review. We'll map your build to the 2026 stack, flag any deprecated services in your reference architecture, and send you a one-page recommendation.

Book an architecture review
Browse our AWS services