AWS Bedrock is Amazon’s answer to the rise of generative AI tools joining the ranks of Google S Vortex and Microsfts OpenAI. 

But Bedrock stands out from the top competitors because it has much easier accessibility and its access to a wide range of different agents, enablding users with low-code customization. 

But what exactly are AWS Bedrock agents? How can they benefit FinOps and how should you be using them to ensure your models are performing to their full potential? 

Learn all you need to know from in this comprehensive guide. 

What is AWS Bedrock?

AWS Agents for Amazon Bedrock banner featuring workflow and document icons on a dark background

Source: Amazon

 

First, let’s define some terms. 

AWS Bedrock is a fully managed serverless offering. It is an abstraction layer atop foundation models (FMs) made from familiar names like Anthropic, Meta, and Cohere. 

You’ll have access to an ever-growing list of FMs and seven providers. Here are the providers you can expect to work with: 

  • AI21 Labs
  • Amazon
  • Anthropic
  • Cohere 
  • Meta
  • Mistral AI
  • Stability AI

Each offers a level of flexibility and customization you ordinarily would have to be a data scientist to achieve. Different FMs are used for various purposes. Meta’s Llama 3 8B Instruct FM can be used for text input and text and chat output, whereas Cohere Embed Multilingual can be used for text input and embedding output. No matter your model of choice, you can always add an extra layer of customization – so long as you’re prepared to pay a little extra. 

What are Bedrock agents?

“Agent” can be a somewhat confusing term. It’s a piece of software powered by a language mode. Each agent can reason, access tools, and complete tasks, depending on the natural language orders received. 

Bedrock agents can work independently or in “agent fleets”. The latter combines different agents’ knowledge to break complex tasks into smaller, more approachable jobs. 

Agent development and deployment are relatively simple and quick to do. With Bedrock, Amazon wanted to create a tool that anyone could use to help improve cloud performance, which is why AWS designed its agents, likes CloudWatch, to collect cloud metrics to help with automated alerts. 

What can AWS Bedrock FM agents do?

Bedrock’s agents are designed to simplify workflows by automating simple tasks – and you accomplish all of that and more without advanced ML knowledge. 

Agents can help you accomplish all of the following: 

AWS vector database selection screen showing Amazon OpenSearch Serverless and Pinecone options with fields for collection ARN and vector index name.

Source: Amazon

Retrieval augmented generation

Bedrock agents will use your company data sources to generate customized responses swiftly and securely using retrieval augmented generation (RAG). This might look like Bedrock agents running your chatbot for you, automating and customizing user responses.

RAG can help a FinOps organization by retrieving cost data for AWS EC2 in previous quarters, and even going the extra mile to compare those costs to the current quarter. 

Manage and complete complicated tasks

Bedrock is designed to make complicated tasks button-click easy. For example, a user could ask their FM to break down the budget for a new QA environment and send alerts for when costs exceed a certain number.

Generate and execute code

You won’t have to worry about complex analytical queries with Bedrock agents because those can be fully automated. That means generating and executing code can be as simple as a click of a button. Ideal use cases for this include data analysis and visualization. That means visualizing and identifying underutilized resources can be made as simple as an agent generating a script and you clicking a button to find unused EC2 instances and more. 

Demonstrate chain-of-thought memory

A Bedrock agent will recall and build upon previous asks and answers, creating a seamless and fully customized user experience. 

You should be able to follow its reasoning and previous learnings with sidebar navigation, breaking down the agent’s every step. You can also retroactively adjust agent reasoning to suit your request better. 

This would look like asking your agent how much your organization spent on data transfers the previous month, and then, based on previous and current spend, what optimizations you can apply. 

How to get started with Bedrock

Bedrock sought to make agent use as approachable as possible. Here’s how you can get started: 

  1. Determine what your use case is. 
  2. Select an agent and provide it with at least one action to perform. 
  3. Using an API call, define the action you would like your chosen agent to perform. 
  4. The API will execute a pre-created lambda function. 
  5. Define a knowledge base. 
  6. Associate that knowledge base with your agent of choice. This will ground your agent with source-specific knowledge, ensuring you know where your agent’s information is coming form. 
  7. Test your agent in the Bedrock console through API calls. Adjust if needed. 
  8. Once you feel satisfied with your agent’s ability, prepare to deploy. 
  9. Create an alias and point that to the version of your agent. 
  10. Confirm your application is set up to send API calls to your agent alias. 
  11. Execute. 

Congrats! You now have a functioning Bedrock agent. 

AWS Bedrock agent code examples

Here are a few of the most common code examples so you feel comfortable sending asks to your agent. 

These coding examples are broken down into three categories: 

  • Basic: Essential must-know code. Examples include Hello Amazon Bedrock Agents.

  • Actions: How you can call individual server functions. Examples include CreateAgent, DeleteAgent, GetAgent, ListAgentKnowledgeBases.

  • Scenarios: How to accomplish specific tasks. Examples include creating and invoking agents and orchestrating generative AI applications. 

NOTE: These examples assume you’re using an AWS SDK (software development kit), so your code might look different depending on if you’re using a different method to access Bedrock. 

How to optimize your Bedrock performance

You’re ready to get started with your new AWS Bedrock agent now and deliver a customized, automated experience… there’s only one concern left to address: how much will this cost you? 

It’s no surprise that Bedrock prices are shrouded in confusion and mystery like most cloud costs. It’s still unpleasant to be unclear on what your monthly or annual bill will look when it arrives. There is one easy way to ensure you always stay within budget – or even leave some room to spare by saving up to 30% on annual spend. 

Cloud cost management tools like Umbrella are create to ensure you optimize 100% of your cloud spend. We make sure Bedrock spend doesn’t sneak up on you with our CostGPT tool (which also uses Bedrock!) We accomplish this by: 

  • Complete visibility into your entire, multicloud environment
  • Data captured down to the hour with up to a two-year retention period
  • AI-powered cloud management and forecasting recommendations
  • Customizable multicloud dashboards
  • 24/7 automated budget monitoring

 

We offer all of that – easy integration with your pre-existing cloud stack and a low learning curve. Your cloud bill can be a pleasant surprise instead of a nasty one with our cloud monitoring services. 

But, why Umbrella? Well, we’ve been working with FinOps organizations of all sizes and locations, demystifying cloud costs for the last decade. 

Want a proof of concept? Talk to us to learn how much you can save with Umbrella’s tools.