Free tools Windows power users keep installed
One-click scans. No signup required.
For Hemapriya Kanagala, the hardest part of building a first AI agent with AWS AgentCore was not getting a model to respond. It was understanding how the surrounding services, permissions, backend operations, deployment, and monitoring fit together. Her project—a fictional customer-support agent—shows why an agent is a system, not just a model with a prompt.
What the project was meant to do
Kanagala describes building the agent as part of Udacity’s Future AWS Agent Engineer Nanodegree Program, supported through the AWS AI & ML Scholarship. The fictional support agent was intended to handle requests such as “Where is my order?” and “What is the return policy?” It was also expected to process refunds, remember customer information between sessions, calculate loyalty discounts, and browse live websites.
As an Amazon Associate I earn from qualifying purchases.
In her account, Amazon Nova 2 Lite through Amazon Bedrock handled understanding requests and choosing tools, while Strands provided the agent framework. That model-and-framework pairing was only one part of the application: each capability depended on other services and on the connections between them.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallHow the parts fit together
Kanagala’s architecture assigned different jobs to different components. They were complementary rather than competing ways to solve the same problem.
#1 Best Overall
| Component | Role in the described project |
|---|---|
| Amazon Bedrock with Amazon Nova 2 Lite | Understood the request and selected a tool, according to Kanagala’s description. |
| Strands | Provided the agent framework. |
| AgentCore Runtime | Hosted the deployed agent. |
| AgentCore Gateway | Connected the agent to tools and backend resources. |
| API Gateway and Lambda | Exposed order operations through an API and performed backend work. |
| Bedrock Knowledge Base | Supplied application-specific product and policy information. |
| AgentCore Memory | Provided retrieved customer context across sessions. |
| Code Interpreter | Handled the loyalty calculation. |
| Browser Tool | Interacted with live webpages. |
| CloudWatch | Supported runtime monitoring; Kanagala says she also created a CPU-usage alarm. |
A useful way to reason about this layout is to follow the work the agent needs to do: retrieve product or policy information from a Knowledge Base; retrieve relevant customer context from Memory; invoke a backend action through Gateway and the API/Lambda path; compute a value with Code Interpreter; or access a live page with Browser. AWS describes AgentCore as a modular platform that can be used with different frameworks and foundation models, and describes Gateway as a connectivity layer whose targets can include Lambda functions and REST API services. Gateway tool schemas and authorization configuration are part of making those connections usable and controlled.
Why permissions became part of building a feature
The clearest failure in Kanagala’s account involved the Browser Tool. She had configured the capability, but the runtime could not start a browser session because it did not have a required permission. After she added the permission, she reports that the test worked.
The lesson is practical: configuring a tool is not the same as granting the running agent access to use it. When a service is present but an operation fails, check both the integration and the runtime’s permissions to the relevant resource. In this case, the capability’s configuration alone did not establish that the runtime could start a session.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Why a successful tool call matters more than a confident response
Refunds exposed a different boundary: the difference between the agent deciding to act and the backend actually completing the action. A customer-facing response should come only after the refund operation succeeds and its result returns to the agent. If the backend rejects or fails the request, the agent must not tell the customer the refund was processed.
As Kanagala puts it, “a model saying that something happened and a system actually performing that action are two different things.” For an action-taking agent, the meaningful chain is intent, tool invocation, backend outcome, returned result, and then the response. A plausible sentence is not evidence that the underlying operation happened.
Calculations still need ordinary software checks
Kanagala reports that an early loyalty calculation was wrong because points and dollar values were treated incorrectly. She corrected the calculation. The episode is a reminder that adding an AI layer does not eliminate familiar software risks: formulas can be wrong, assumptions can be inconsistent, edge cases can be missed, and configuration or code can contain bugs.
Rank #3
For calculations that affect a customer’s balance or benefit, define the units and conversion rules explicitly, then test representative cases. The model’s ability to invoke a calculation tool does not validate the formula used by that tool.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Deployment was not the same as validation
Kanagala says she tested six capabilities separately and that all six eventually worked. These are the tests she reports for this project, not an independent assessment or evidence of production readiness.
- Track an order.
- Process a refund.
- Answer a question using the Knowledge Base.
- Retrieve information across sessions.
- Calculate a loyalty discount.
- Browse a live website.
Testing these paths individually makes failures easier to locate: an order lookup depends on the backend path, while a policy answer depends on retrieval from the Knowledge Base. A successful deployment alone does not show that each intended capability works end to end.
AWS’s current AgentCore developer guide describes several ways to work with the platform, including the CLI, Python SDK, MCP server, AWS SDK, console, and AWS CLI. AWS notes that the CLI and Python SDK do not cover every operation supplied by the AWS SDK; when other services such as Lambda are used alongside the AgentCore SDK, AWS SDK integration may be needed. The console includes an agent sandbox for testing, and the CLI supports creating, configuring, deploying, and managing agents. These interfaces can change, so consult AWS’s live documentation for current procedures rather than relying on a fixed sequence of console labels.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Monitoring needs to match the risks
Kanagala says she used CloudWatch to monitor the AgentCore runtime and create a CPU-usage alarm. She also notes that a production version would need attention to failed requests, errors, latency, tool failures, resource usage, service health, and cost. Those are operational concerns she identifies, not measured outcomes from the project.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →The monitoring targets follow from the architecture: a response can fail because the runtime, a tool, a backend operation, or a connected service did not behave as expected. Monitoring only whether the agent process is running would not reveal every failure that matters to a customer.
Best Value
The learning approach that made the system manageable
Kanagala says that seeing many service names at once made it difficult to understand how they fit together. What helped was asking what each component did, what it connected to, and what should happen if it failed—rather than trying to master the entire architecture before beginning.
Her question for the next step was: “What is the next thing I need to understand?” For a first agent build, that means choosing one capability, tracing its path through the system, and checking its expected result before adding the next moving part. The project’s hardest work, in her telling, was learning those connections and verifying them—not simply prompting the model.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
