🦄 Making great presentations more accessible.
This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.
Overview
📖 AWS re:Invent 2025 - Transforming FinServ AI DLC with Kiro Agentic IDE (DEV203)
In this video, AWS AI Hero Vivek Velso demonstrates transforming FinServ AI development lifecycle with Kiro through spec-driven workflows. He addresses three major challenges: technical debt consuming 20% of IT budgets, compliance burdens in regulated industries, and security vulnerabilities. The presentation showcases Kiro's key features including global agent steering for enforcing standards, agent hooks for automating policy-as-code using CFN Guard, MCP servers like diagrams for rapid architecture visualization, and Kiro CLI with custom agents for optimized context management. Practical examples include building a loan approval microservice with CDK, automatically updating outdated dependencies, and implementing shift-left compliance checks within the IDE.
; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.
Main Part
Introduction to Kiro and the Spec-Driven Workflow for FinServ AI Development
Good evening, everyone. My name is Vivek Velso. I'm an AWS AI Hero. You might be wondering what an AI Hero is. It's a new category that AWS just created by renaming the Machine Learning Heroes category to AI Heroes. I have been a former AWS community builder and AWS ambassador with 27 years of experience in the industry. I'm excited to talk about transforming FinServ AI DLC with Kiro. By show of hands, how many of you have used Kiro? Excellent. Even if you have not used Kiro, there are 99 sessions on Kiro at this re:Invent, so do sign up for those breakout sessions on Kiro.
Without further ado, we are going to talk about the challenges in FinServ development cycles and how the spec-driven workflow with Kiro and AI DLC addresses them. I'll share practical examples with best practices, and it's all available in my GitHub repo that I will share at the end, so you can feel free to clone and experiment with it.
Have you all seen the explosion of technical debt that happened today? Excellent. That was using AWS Transform, but as you know, technical debt is the invisible drag on enterprise innovation. More than 20% of the IT budget is spent on technical debt issues. The developers I worked with spend more than an hour every day fixing Lambda runtime issues, end of life support, and container-based image validation. It's an ongoing issue.
The second issue when working with regulated industries is the compliance burden. There are heavy penalties and fines if the application is not compliant with regulatory requirements. Even if I have to make a single line of code change, I have to run the code through the whole pipeline, which sometimes takes 5 minutes. That's not a good developer experience. How do we shift left instead of taking a reactive approach?
The third issue is the piling up of incidents and unpatched open-source dependency vulnerabilities that lead to huge security debt in the organization. Now let me present the hero spec-driven workflow. To summarize, the spec-driven workflow consists of Kiro specs, which is a collection of Kiro requirements, design architecture, and tasks. It's a declarative way of producing code. Kiro compiles the specs into code. Even though these LLMs are not deterministic, you can enforce that through agent steering, which can be used for achieving compliance.
Then we'll talk about agent hooks. Agent hooks take care of all the manual mundane tasks, like updating documentation. Nobody likes to update documentation every time you change endpoints. With agent hooks, you can automate all that. I'll also show you how to implement policy as code using agent hooks. Then there is an MCP server capability. I'll show you how I can use the diagrams MCP server to speed up the developer experience. MCP servers also let you connect to your organization data like Confluent, SharePoint, and Jira, and get that output to the LLM so it can produce more accurate data and more accurate application code.
Agent Steering and Agent Hooks: Eliminating Technical Debt and Automating Compliance
Finally, there is the Kiro CLI, which you might have used as QCLI, but it's been rebranded as Kiro CLI. I'll show you how sub-agents can be used to improve productivity and context engineering. Let's start with agent steering. I built a simple loan approval microservice with an API Gateway with a Lambda backend. The Lambda has the business logic and saves the results to DynamoDB, and there is CloudWatch for logging. I implemented this using CDK, which is an Infrastructure as Code tool written in TypeScript. Python runs in the Lambda, and Kiro built all this in less than an hour.
What happens when you use state-of-the-art models like Claude 4.5? Because of its training data, the CDK libraries are still at version 2.100 instead of 2.170, which is huge technical debt. Even the TypeScript libraries are way old. Claude doesn't know that Python 3.9 was end of life in October 2025.
So if I push this code to higher environments, I'm going to get a huge chain of emails. How should I fix this autonomously using Kiro? The answer is global agent steering.
Kiro now supports global agent steering. You define one steering file on your workstation, and it will apply that steering to all the projects in your workstation locally. A steering file is simply a set of rules in plain English, so you can instruct Kiro to use all these instructions from the steering, like using stable versions and not hardcoding any secrets. All your organization guidelines and standards can be used in this steering file.
One important tip is to use words like "must" and "critical." Cloud models are especially tuned to those keywords. So use those keywords, and then cloud respects those instructions from the steering files. If I have those steering files, I can just ask with a prompt to apply that steering file to my code. Then within one minute, Kiro will update all the packages and make it tech debt-free.
Kiro is a VS Code extension, just like Windsurf and Cursor. With Kiro, you can install all the linting extensions like Pylint if you're using Python, or ESLint for TypeScript. With these extensions, you can get quick analysis of your code for any errors and see if you're not following any best practices. Kiro will show you right in the IDE with these linting extensions. But remember, you have to enable these extensions. They are not enabled by default.
Next up is agent hooks, which I use for automation. I'm going to show you how I use policy as code using Kiro for shift left. AWS has a wonderful open-source tool called CFN Guard. It's a domain-specific language that can read a set of rules from every framework. For PCI there is a custom set of rules. For New York DFS there is a set of rules. HIPAA has a set of rules, so everything is available in the GitHub repository that's shared here. You can customize these rules for your organization.
CFN Guard runs those DSL rules on CloudFormation templates. It can also run on Kubernetes configuration files and Terraform plans. So it's not just CloudFormation, but I'll show you how to run it on CloudFormation. To run it with the CLI on those rules and apply it on the configuration locally without pushing it to the cloud, and then once all the rules pass, you deploy. That's the two-step process. But I don't want to do manual execution.
So how do I do that using agent hooks? I just tell Kiro to create an agent hook for me. How do I trigger it? Just when the infrastructure file is changed, on the file save action, it will trigger these hooks. I just showed the hook how it runs and pointed to the fractional credits and showed me what rules passed within the IDE in Kiro.
Now there is another way to run CloudFormation Guard using AWS Toolkit. AWS released the CloudFormation language server that now has the capability to show you the CloudFormation stack errors, everything within the Kiro IDE. So you don't have to go to the AWS console and do context switching anymore. Everything is available within Kiro. The same guard rules can be applied within the AWS Toolkit in the settings. This is very helpful if you're using CloudFormation YAML or JSON.
The other thing I want to share is that if you're using JSON files, it's always good to use JSON when you're running the CFN Guard, because in YAML, the intrinsic functions like exclamation marks sometimes don't work very well. So the JSON format is very good for using CloudFormation Guard and with the toolkit. Whatever errors you have with the YAML configuration in the stack, it will just highlight with squiggly blue lines within the stack, so you don't have to run anything. This is very helpful if you are using YAML files.
MCP Tools and Kiro CLI: Enhancing Developer Productivity with Custom Agents
My favorite feature is the MCP tools. AWS and Kiro have made it very easy to install MCP servers from Kiro.dev. All the available MCP servers are listed on Kiro.dev, and earlier there used to be separate documentation servers and AWS API servers. Now there is one unified AWS MCP server that can be installed with a single click and enabled from Kiro.
Earlier, it used to be very difficult to share MCP files across your team because of tokens, so you had to redact those token variables and share them across teams. But now with support for environment variables in Kiro, you don't have to do that. You can share your MCP JSON files within your team, which simplifies team development. When another team member uses it, the system asks for confirmation that you have to approve for that environment variable locally on your site and then run it.
My favorite is the diagrams MCP server. Earlier, I used to take an hour to create a draw.io diagram or architecture diagram, but now with diagrams MCP it takes less than a minute to generate architecture diagrams. There are also other useful MCP servers like the Headless Chrome Dev Tools browser, which is very useful for debugging the UI. It runs Chrome Dev tools and finds errors for you, and the agent can then do the debugging. This is very useful for front-end applications.
The last one is the Kiro CLI, which is the rebranding of the QCLI. Context is the king in coding assistance. Whenever I have a large code base, after four to five prompts, I see my context fills up to 100 percent, and then I have to start from scratch again. One solution is custom agents. With custom agents, you can create dedicated agents like a reviewer agent, an architecture agent, or a compliance agent. With specialized agents, your context is heavily optimized, and you don't run out of the context window.
There is a two-step process using Kiro CLI for creating custom agents. You use the slash agent create command, and then it will prompt you for the JSON configuration for that agent. The agent configuration is just a JSON file where you have to give the name of the MCP server that you are going to use for the agent, the prompt, and the steering files. That is it. If you are working on the front end, use the front-end MCP server like the Chrome Dev Tools MCP server. You don't have to pass all the MCP servers, which would take up your context. That is a very efficient way of using custom agents.
Launching custom agents is just a single command. With delegate launch, it will launch that agent in the background, do its task, and then report back the status. Then you can continue with your next task. The second way to run these agents is using the slash agent swap command, which I haven't shown here. With that, the agent comes in the foreground and becomes the default agent, and you can work on that code review agent or the compliance agent.
I end with a famous quote from Henry Ford, who transformed the automobile industry, just like Kiro is transforming the financial services industry and any other industry for AI agentic AI coding. Here is my QR code for the GitHub repository on the top right. Feel free to scan it. It has all these examples shared in the demo. Connect with me on LinkedIn, and I am available at the back after this talk. I would love to hear about your experiences with Kiro and how you use Kiro for your projects. Remember to complete the session survey in the app, as that really helps the team. Thank you.
; This article is entirely auto-generated using Amazon Bedrock.
Top comments (0)