Naveen ChatlapalliSolution Architecture · AI Agents
← All episodes

The field notes

The Daily AI Drop - Evening Edition, October 9, 2026

Thirteen afternoon videos: LangChain on skills, evaluations, and managed agents; OpenAI Dots go mobile; Gemini runs remote robots and live translation; Iris-3B paints pixels locally; plus quantum chemistry tooling.

Thirteen afternoon videos: LangChain on skills, evaluations, and managed agents; OpenAI Dots go mobile; Gemini runs remote robots and live translation; Iris-3B paints pixels locally; plus quantum chemistry tooling.

Listen: https://muse.ai/podcasts/media/1180242951829162/4c015424-5891-46ff-a366-0f9e546ae170/ep-c22daffd-af4c-42ec-b4b3-fb29461970d9.mp3

Videos covered

Iris-3B: General Vision Learner and Pixel-Space Generation Locally

Fahd installs and tests Iris-3B locally, a three-billion-parameter vision model that paints every pixel directly in pixel space instead of working through compressed tokens. It continues his week's through-line of small, open models you can run privately on your own hardware with no API bill. Worth watching if you care about owning your vision stack end to end.

This week's repos: PS5 game on a PC! Turn agents into employees. AI video maker.

Andrew Warner and Adam break down ten trending GitHub repositories, from running PlayStation 5 games on a PC to turning individual coding agents into coordinated employees, generating video with code, and building custom 3D models with AI. The pattern matters more than any single repo: open source is where agents stop being chatbots and become inspectable infrastructure.

15 years of tech making in 60 years

A short marking fifteen years of Priyanka's visual explainers, built on a simple belief: the future of tech should not belong only to the people who already speak its language. She started because brilliant ideas kept getting lost behind jargon and crowded slides. Fifteen years in, that clarity-first approach is why this channel stays in the lineup.

What is a Skill?

LangChain defines a skill as a folder of instructions, scripts, and reference docs for a specific task. Your agent loads only each skill's name and description at startup, then opens the full skill only when a task calls for it. That keeps the context window small, so you spend fewer tokens and your agent runs faster. If your agent gets slower and dumber the longer it runs, this is the fix.

Three Questions Every Agent Engineer Should Be Able To Answer

Jake Broekhuizen argues owning your intelligence comes down to three questions: where is your agent failing, how can you reliably make it fail again, and what needs to happen so it stops failing. When frontline LLM judges get too expensive at scale, fine-tune a smaller judge to mine production traces. This is the unglamorous work that separates a demo from a product.

Managed Deep Agents | Deploying mission critical agents to production

Managed Deep Agents, now in public beta, bundles the Deep Agents harness with managed LangSmith infrastructure. Define an agent as a folder of instructions, skills, and tools, test it locally, and deploy with the mda CLI. LangChain uses its own internal go-to-market agent as the example — eating their own cooking on something mission critical is the real beta test.

Master OpenAI's Dot Agent in 8 Minutes (Complete Guide)

A complete walkthrough of OpenAI's proactive personal Dot agent after Dev Day: how to set one up, connect calendar and email, and move from casual chatting to handing it real recurring work. The shift is from prompt-and-reply to an agent that watches for what needs doing.

Set up your dot in the ChatGPT mobile app

You can now create a Dot from the ChatGPT mobile app, edit the name, customize the avatar, and set your Dot to be the first conversation you see when you open the app. That placement matters — making the agent the front door of the app is how a feature becomes a habit.

Building the Codex community

Pauline Narvas and VB talk through growing the Codex Ambassador program, favorite moments from local meetups and Build Week, and what it means to build with heart. Tools spread through people showing each other what they built, and the community layer is doing real distribution work for Codex.

Order in any language with Gemini 3.5 Live Translate

Gemini 3.5 Live Translate and Gemini 3.5 Transcribe work together so an English speaker can place a sushi order with a Japanese-speaking host, translating the conversation in real time. Live translation has been promised for years; seeing it hold a real transaction together, turn by turn, is the version that changes travel and small business.

Operating real hardware with Gemini AI

When your friend has a robot arm in another state, you find a way to use it. Matías and David used Gemini to bridge 1,700 miles, building custom 3D web controllers that operate physical hardware in real time. Gemini leaving the chat window and touching the physical world, with the web as the control plane.

Building low-latency remote robotics with Gemini

The full build behind the shorts: working 1,700 miles apart, Matías and David used Google Antigravity, Google AI Studio, and Gemini to construct a remote robotics framework powered by custom 3D web controllers, with low-latency hardware controls operating physical machinery across thousands of miles.

QDK Tutorial: Build Chemistry applications with the Quantum Development Kit for Chemistry

Microsoft's Quantum Development Kit for Chemistry is an open-source library designed to bridge domain researchers and quantum algorithm developers. David Williams-Young walks through the library's purpose, workflow integration, and its role in advancing practical quantum applications — a reminder that the developer story is not only language models.

Original episode notes