This week on the channel +1.1M views this week

AWS builds a research agent with a detective persona

Tejas Kashinath takes AgentCore from terminal commands to a deployed assistant in GenerAIly Speaking.

Channel:

Amazon Web Services Story of November 4, 2026Trari Newsroom

Why this story? The channel's weekly summary: what went up this week, what is drawing views now, and where the channel stands among similar channels.

GenerAIly Speaking: Let's build a research agent from scratch

Things we can learn from the video

Press to jump

2:08Managed harnesses take a system prompt and model, while custom agent code gives developers a different implementation path.
3:22Tejas creates a Python agent project with agentcore create and selects deployment and runtime options.
5:43agentcore dev opens the local browser inspector, where Tejas tests a basic greeting.
7:11agentcore deploy handles the cloud deployment workflow, including dependencies and CDK synthesis.
8:05The deployed ResearchDetective appears in the inspector, with traces and stored memory strategies available to examine.

Can a research assistant have a detective's personality? On Amazon Web Services, the practical demonstration ends with just that: ResearchDetective answering in a gritty persona, after a walkthrough of the commands used to create, test and deploy an agent.

"GenerAIly Speaking: Let's build a research agent from scratch" was the most-watched of the channel's 15 uploads this week, reaching about 45,000 views. Bharathi opens by promising a live build with nothing prerecorded. Tejas Kashinath handles the terminal, with Jay Busch and Madhu Samhitha Vangara joining the session. This is AWS's own product demonstration, using its AgentCore tools.

The useful thread is the progression from local experiment to cloud deployment. Tejas creates the project, opens a browser-based inspector to test it, then deploys it and examines traces and memory. The detective voice gives an otherwise technical workflow a small flourish of character.

That episode and "GenerAIly Speaking: Let's train some humanoids" also carry the channel's strongest daily viewing momentum. The latter shifts from research to a managed training-and-evaluation pipeline, including validation checks and approval controls. Elsewhere, Brian Simpson-Adkins and Rafael Lopes introduce an operations series built around both AI for operations and operations for AI.

Subscribers were unchanged over the week, while the channel collected roughly 1.1 million views. Daily viewing is currently well below its usual level, despite the two GenerAIly Speaking uploads finding substantial audiences. For viewers catching up, those episodes offer distinctly different sides of agent development: building an assistant and organizing a training workflow.

Published on Trari