AI News · The field notes
The Daily AI Drop - Evening Edition, October 29, 2026
Agents go to work at Zendesk, Skyworks, and in your database, plus a serverless shake-up, offline speech AI, and the case that SaaS is dead.
Twelve new videos this afternoon, with one thread running through them: AI agents are going to work in real companies, and the infrastructure underneath them is quietly getting out of the way.
Listen: Listen to this episode
ChatGPT Enterprise at Zendesk: Putting Company Data to Work — OpenAI
Zendesk has about 600 applications built by different people across the company, and ChatGPT Enterprise is now available to every employee. Business users ask questions about their own company data directly, and in sales an AI agent connects product usage with support tickets to surface expansion opportunities. It matters as a preview of enterprise AI beyond the pilot phase: hundreds of small internal tools plus agents sitting on top of real company data.
ChatGPT Enterprise at Zendesk: Putting Company Data to Work — OpenAI
A second upload of the same Zendesk case study, posted later in the afternoon. Same story, same takeaway: the version of enterprise AI that sticks is not a single department pilot but company-wide access with agents working across sales, marketing, and operations data.
Stop wiring infrastructure, start building agents. — Microsoft Developer
Microsoft's Copilot harness in Foundry lets you build a prompt agent as a model, instructions, and tools, defined once and versioned — no container, no server, no pipeline. The pitch is that most agent projects rebuild the same scaffolding for no reason. It matters because starting simple and only paying the infrastructure cost when the agent earns it is the fastest path from demo to production.
Stop wiring infrastructure, start building agents — Microsoft Developer
The short version of the Foundry Copilot harness episode, pointing to the full episode above. Same message in condensed form: most of the time you do not need the containers, scaling, state, and auth scaffolding that agent projects usually start with.
How to Use Database Agents in Microsoft Fabric | Data Exposed — Microsoft Developer
Database Agents in Microsoft Fabric help database administrators detect performance issues before users complain, moving database work from a reactive model to a proactive one. It matters because databases are where agent hype meets reality: agents that monitor the database itself are the unglamorous layer that makes the flashy agents reliable.
BIGGEST AWS Lambda Update In Years — Mehul Mohan
Mehul Mohan breaks down the biggest AWS Lambda update in years. It matters because serverless is where a huge share of agent workloads run — agents spin up, do a job, and shut down — so a major Lambda change reshapes what is cheap and easy to build for anyone running agents in production.
Nemotron 3.5 Transcribe 30B-A4B: Offline Speech Model Locally — Fahd Mirza
Mirza locally installs and tests Nemotron 3.5 Transcribe, an open-weight, offline speech and audio model. Following this morning's local diffusion model for images, the local AI stack now covers voice too. It matters for privacy — transcription on your own machine without sending audio to the cloud — and for builders who want voice features without a network connection or a per-minute API bill.
Building Semiconductor Agents at Skyworks with LangGraph + LangSmith — LangChain
Skyworks Solutions is putting agents to work on end-to-end test automation, root cause analysis, and design assistance, each with its own KPIs. Root cause analysis that took about 20 days manually is down to 10 on a first pass. It matters as proof that agents pay off on long, messy, hard-to-reproduce work — and that moving from homegrown tooling to LangSmith improved deployment, streaming, and threading for the team.
The Hardest Part of Testing Agents: Write Access — LangChain
LangChain PM Ben Tannyhill explains why read-only agents like Chat LangChain are easy to branch and test, while write-access agents that touch production data are still an unsolved evaluation problem for most teams. It matters because a wrong read-only agent breaks nothing, while a wrong write-access agent can change a database, close a ticket, or move money — testing that safely is the gap between a demo and an agent you trust near production.
Two AI bots for spam calls and data privacy — Matthew Berman
Berman looks at two AI bots aimed at spam calls and data privacy — AI showing up in the least glamorous, most personal places. It matters because if agents are going to earn trust, this is where they earn it: saving you from interruptions and keeping your information closer to home.
SaaS is dead? — Greg Isenberg
Isenberg argues that software as a service is dead, or at least breaking: for 15 years SaaS meant paying engineers to build a tool and charging a subscription, and AI that builds and customizes software on demand shifts value from the tool to the outcome. It matters even if you disagree, because every subscription product now has to answer what it does that an agent cannot just do for you.
Accelerate shipping to scale with Google Antigravity — Google for Developers
Using Google Antigravity, Move Cars delivers fully tested production features in a single day for software that moves tens of millions of vehicles, with agents running autonomously while developers stay in total control. It matters as the difference between an agent demo and an agent you would trust with a production system: autonomy with a human grip on what ships.