Deploying an application to AWS has always involved a predictable sequence of friction. Research which services fit your stack. Compare pricing. Write the infrastructure-as-code. Configure the deployment pipeline. It’s not difficult work for experienced teams, but it’s time-consuming — and it repeats every time the architecture changes.
AWS wants your coding agent to handle that sequence for you. On February 17, AWS announced Agent Plugins for AWS, an open-source repository of plugins that give AI coding agents the domain knowledge to architect, deploy, and operate on AWS. The first plugin, deploy-on-aws, works with Claude Code and Cursor today. Type “deploy to AWS” and the agent walks through the entire process — from codebase analysis to live infrastructure.
What Agent Plugins Actually Are
The concept is straightforward, but the packaging is thoughtful. An agent plugin bundles different types of expertise artifacts together.
Agent skills are structured workflows — encoded domain expertise that guides the agent through deployment step by step, rather than having to figure it out from scratch.
MCP servers connect the agent to live external data at runtime. The deploy-on-aws plugin uses three: AWS Knowledge (documentation and architecture guidance), AWS Pricing (real-time cost data), and AWS IaC (CDK and CloudFormation best practices).
Hooks add automation and guardrails — validating changes, enforcing standards, or kicking off workflows automatically.
References provide documentation and configuration defaults without bloating the prompt context.
This replaces the common workaround of pasting long AWS guidance into prompts. Teams encode deployment guidance once as a versioned, reusable plugin that agents invoke when relevant.
The Five-Step Workflow
The deploy-on-aws plugin follows a structured sequence that mirrors how an experienced cloud architect would approach deployment.
Analyze: The agent scans your codebase — including framework, database, and environment dependencies, as well as expected traffic.
Recommend: Based on the analysis, select AWS services with rationale — App Runner for the backend, RDS PostgreSQL for the database, and CloudFront and S3 for static assets.
Estimate: Real-time pricing from the AWS Pricing MCP server shows projected monthly costs before you commit.
Generate: After review, the agent produces CDK or CloudFormation code, a Dockerfile, migration scripts, environment configuration, and a CI/CD workflow.
Deploy: You review, adjust, and confirm. The agent provisions resources, builds containers, creates the database, and configures the CDN.
AWS claims the process takes under 10 minutes, even though it typically requires hours of research and IaC authoring.
“The friction between writing code and running it in production has always been a constraint on how fast software actually moves. AWS encoding deployment expertise directly into agent plugins removes a category of work that has historically required specialized knowledge, repeated research, and manual configuration. That workflow compression matters because deployment friction has never just been inefficiency; it has been a governor on how quickly teams can test, iterate, and ship,” according to Mitch Ashley, VP and practice lead, software lifecycle engineering, The Futurum Group.
“When that governor comes off, the pace of production changes structurally. Teams that previously measured deployment cycles in hours begin measuring them in minutes. The bottleneck shifts from ‘how do we get this to AWS’ to ‘what do we build next.’ That is a different kind of pressure on product and engineering leadership.”
Why This Matters for DevOps Teams
Three things stand out about this approach.
First, the expertise is versioned and standardized. When deployment guidance lives in a plugin rather than in individual engineers’ heads, it becomes consistent across teams. Every developer gets the same architecture recommendations, cost methodology, and IaC patterns.
Second, the plugin model is extensible. AWS says more plugins are coming—presumably covering operations, monitoring, security reviews, and cost optimization. The architecture supports packaging domain expertise as plugins.
Third, the human stays in the loop. The agent recommends, estimates, and generates — but the developer reviews and confirms before anything gets provisioned. AWS is explicit: plugins are accelerators, not replacements for judgment. Their best practices include running security scans on generated IaC and following least-privilege for credentials.
The Broader Pattern
AWS isn’t alone in moving this direction. Google’s Conductor extension encodes development workflows as persistent Markdown files that guide Gemini CLI agents. Cline CLI 2.0 embeds agents directly into CI/CD pipelines. The industry pattern is consistent: cloud providers and tool vendors are packaging domain expertise in formats that AI agents can consume and act on.
For DevOps teams, the implication is that infrastructure-as-code is becoming infrastructure-as-conversation. The IaC templates still get generated — CDK and CloudFormation aren’t going anywhere — but the process of choosing services, estimating costs, and writing the initial templates is shifting from human research to agent-guided recommendation.
That’s a meaningful change in the deployment workflow. And because Agent Plugins for AWS is open source, teams can inspect, modify, and contribute to the expertise that guides their agents — a transparency model that matters when the output is production infrastructure.
