Agentic AI: build systems that reason, act, and know when to stop.

Careers: You Built Agents. Now What?

I added a Careers tab that matches the skills from this course to the jobs that use them, using public data from OpenAI ↗, BLS ↗ and Lightcast ↗. See which skills you built in this course and the careers they can open up for you. Want to know something the page doesn’t answer? Ask it at the bottom of the page, and if there is data that can answer it, I will add the analysis there.

Built your capstone in Fleet? Fleet has no public link, and here is the way around it

• The problem: Fleet’s Sharing drawer offers Private, Workspace (paid) and Specific people, and all three need a LangSmith login. Milestone II asks for a link anyone can open.
• The fix: every Fleet agent has an API endpoint. A 40-line Gradio page on your laptop forwards messages to it with your key and is shared the Week 9 way with SHARE=1. Your agent, its tools and its traces stay in Fleet. Tested on one of my own agents on Nov 2: it costs no money, and the gradio.live link needs no account.
• Where: the file, the three values to copy from Fleet, the exact commands for Mac and Windows, and a troubleshooting table are in the Week 10 summary, in the box after the five Week 9 steps. Switching to another platform remains fine too.

Week 11 is agentic commerce plus a trace lab: create a free LangSmith API key before class

• The lab: a purchasing agent on your laptop shops a teaching web store built for this course, decides on its own whether to buy, and LangSmith records every step. You read the record and find out why it bought, or did not.
• Five minutes at home: sign in at smith.langchain.com (your Week 2 account), create an API key, create an Observability agent named shop-agent, and run pip install -U langsmith. The steps, the add-on download and the store link are on 03 · Trace it.
• Also this week: how Meta’s Muse pays with a one-time card, and whether stores let agents in at all (01 · Agents that buy). Group Assignment 3 is due. No quiz.

Week 10 wrap-up: GA3 due next week, Milestones II and III released, and the next three sessions are project time

• Group Assignment 3 is due next week in your own section. Join a GA3 group on iCollege first (Assessments → Groups → the GA3 category); until you do, the assignment is invisible to you.
• Milestones II and III are released. II (due Week 12): a public link anyone can chat with, an eval harness with pass^k, a governance note. III (due Week 13): the complete system, the security review, honest limitations.
• Weeks 11–13: a short lecture, then project time in class with me going desk to desk. Week 11’s topic is agentic commerce. Bring your laptop with your capstone running.
• Any tool, framework or platform is fine for the final project; the required artifacts are the same on every path.

Everything in one place: Week 10 summary →

Global Analytics Summit 2026: Organizing AI Agents, on behavior, incentives and design in human-AI teams

Two half-days of keynotes and panels from the Center for Analytics and Transformative Technologies at UT Austin’s McCombs School of Business, on the questions of our Weeks 9 to 11: how a company organizes its agents, how it governs the data they touch, and what changes in a team when some of the members are agents. Free, online over Zoom, registration required; the agenda lists academics from UT Austin and Santa Clara alongside practitioners from TransUnion, ConocoPhillips, Precisely, Salesforce, Franklin Templeton and DigiKey.

• Most useful hour for this course: the pre-conference primer “Organizing Agents in the Enterprise” (Kumar Muthuraman, UT Austin), Thursday 10:00–11:00 AM Central, which is 11:00 AM–12:00 noon in Atlanta; later that afternoon an academic panel on managing agentic systems at scale with Peter Stone (UT Austin computer science).
• Friday morning: an industry keynote and two panels, “Architecting the agentic enterprise” and “Agentic data governance” (behavioral economics, incentive design, information stewardship), 8:45 AM–12:15 PM Central, which is 9:45 AM–1:15 PM in Atlanta.
• Timing note: Thursday’s main program runs 12:45–5:15 PM Central (1:45–6:15 PM Atlanta) and overlaps class for the Thursday section; Friday has no class. The organizers send the program to all registrants afterwards.

Register: get.mccombs.utexas.edu/catt2026 ↗ (free, Zoom webinar, confirmation by email)  ·  organizer page: CATT Analytics Summit ↗

Google Cloud Security Summit (Americas): securing and governing AI agents

A 90-minute online session from Google Cloud with Mandiant and Wiz, built around exactly the questions our security week asks: how to securely build and govern AI agents, how to safeguard applications from the ground up, and how to respond to incidents in minutes with agentic security tooling. It lands the week after Week 10 (Security, Trust & Governance), so you will recognize the vocabulary: least privilege, prompt injection, identity, audit trails.

• Who should go: anyone interested in security as a career direction, and anyone whose capstone has a real tool boundary (that is all of you) and therefore a security review coming in Milestone III. Hearing how Mandiant and Wiz frame agent security in production is good preparation for writing yours.
• Online (Google Cloud OnAir); registration needs a Google account and no fee is listed. Times: 9:00 AM Pacific = 12:00 noon Atlanta.

Register: cloudonair.withgoogle.com › Security Summit AMER ↗  ·  event details: Google Cloud Security Community listing ↗

AI Systems Playground: learn how the model under your agent actually runs

This course treats the model as a component you call. AI Systems Playground ↗ opens that component up: short browser lessons where you operate the system instead of reading about it. The first lesson follows one sentence through a model as it picks its next word (no background needed). The KV-cache lesson (about 15 minutes) lets you set the model size, GPU count, context length, concurrent users and precision, and watch memory fill up until an 8-GPU server runs out, which is the physical reason long contexts cost what they cost (Week 3) and why the local model on your laptop is 0.5B and not 70B (Week 5).

Each lesson ends with an interview question and a reading list. Free, in the browser, nothing to install.  ·  source on GitHub ↗

Claude Startups is expanding: a free year of Claude Team, $1,000 in API credits, and office hours with Anthropic’s Applied AI team

Anthropic widened its Claude Startups program on Oct 6. Approved companies get one year of Claude Team free (up to five Premium seats; for organizations new to Team), a one-time $1,000 in Claude API credits with higher rate limits, biweekly 45-minute office hours with Anthropic’s Applied AI team, and the new Claude Startup Stack: up to $45K in offers on the tools a startup runs on (design, data, engineering, sales), provided by partner companies rather than by Anthropic.

Who qualifies: read this before you apply: startups founded in the last five years or funded in the last two. No VC required: bootstrapped and pre-seed count. But it is a program for companies, not class projects: you need a Claude Console account, a company email on your own website domain, and a short description of what you are building. If your capstone team already is a company, or is about to become one, it is worth the 10-minute application; most decisions come back within minutes, the rest in 2–3 business days. If not, the course path stays what it has been: free tiers and your local Ollama model.

Apply: platform.claude.com/offers/startups-application ↗  ·  program details: claude.com/programs/startups ↗  ·  the announcement ↗