This syllabus is tentative and subject to change as the course progresses. Deliverables are due the Sunday before the listed session, alongside the weekly update.
There is no class on Tuesday November 3, Election Day, which is a university holiday. The week 9 deliverable is still due that Sunday and the weekly update still runs, so the cadence does not break.
The weekly cadence
Sunday: the update
Five lines, from every team, every week from week 2 through week 12. No extensions.
1. Shipped this week:
2. Our number, and last week's number:
3. What a user told us (a real quote, not a summary):
4. Biggest blocker:
5. Next week's target:
About line 2. Through week 3, your number is the count of people you have talked to who have the problem. At the week 4 proposal you choose your own metric and it locks. You may change it once after that, in the week 9 pivot-or-persevere decision, and not otherwise. A metric you can redefine whenever it suits you is not a metric.
Tuesday: the session
| Minutes | Block |
|---|---|
| 0:00-0:15 | Talk. One idea, timed to where teams are that week. |
| 0:15-0:55 | The round. All ten teams, 4 minutes each at the front. Your number and its change, what you shipped, your one blocker. Then questions from the room. No slides. |
| 0:55-1:35 | Deep dives. Two teams, 20 minutes each. Real grilling from the instructor, visiting mentors, and the class. |
| 1:35-1:45 | Commitments. Every team states next week's target out loud. Logged, and checked against the round next week. |
| 1:45-1:50 | Buffer. |
Deep dives run in weeks 2 through 11, excluding week 8 and excluding week 9, which falls on Election Day, two teams per week. That is 16 slots for 10 teams, so every team goes at least once and six go twice. The schedule is drawn and published in week 2. Teams may trade slots with each other as long as someone fills the week.
Office hours are where depth happens. By appointment with the instructor, and separately with visiting mentors when they are available. Any team can book 30 minutes. Teams that are stuck should be booking every week. Teams that are moving may go a month without booking. Class time is for what everyone benefits from watching; your specific problem is for office hours.
Two mandatory 1:1s with the instructor, 20 minutes, calendar-booked, in weeks 4 and 10. Week 13 is a separate all-day Demo Day rehearsal, not one of these.
Schedule
| # | Day | Date | Talk | Due | Session |
|---|---|---|---|---|---|
| 1 | Tue | Sep 8 | Orientation. Why agent startups die: the gap between a demo and something people rely on. Idea vetoes. Deep dive schedule drawn. | Nothing. Show up. | Open session, anyone may attend. Applications close Thu Sep 10. |
| 2 | Tue | Sep 15 | How to talk to users. Watch them do the task, time it, price the time | One-line problem statement. Weekly updates begin. | First full session. The round and deep dives begin. |
| 3 | Tue | Sep 22 | Picking the wedge. One task, one persona, one workflow | 10 logged user conversations, with quotes | Talk, round, deep dives |
| 4 | Tue | Sep 29 | Evals before features. If you cannot measure whether the agent did the job, you cannot improve it | Proposal: problem, user, wedge task, what "working" means, your metric | Talk, round, deep dives. Mandatory 1:1s. |
| 5 | Tue | Oct 6 | Unit economics. Token cost, retry cost, human fallback cost, against what anyone will actually pay | Eval set of 20 real cases, current pass rate | Talk, round, deep dives |
| 6 | Tue | Oct 13 | How to launch small. Ten users who care beats a front page | Cost per successful task, measured not estimated | Talk, round, deep dives |
| 7 | Tue | Oct 20 | Distribution. You are selling into an existing workflow, not launching an app | LAUNCH. Live URL, one external user. Non-negotiable. | Talk, round, deep dives |
| 8 | Tue | Oct 27 | No talk. Mock diligence with visiting investors, full period | 5-slide data room | No deep dives. Mock diligence runs the whole session. |
| 9 | Tue | Nov 3 | No class. Election Day, university holiday. | Written pivot-or-persevere decision with evidence. Still due Sun Nov 1. | No session. Your metric may change here, once. Decisions are read in the week 10 round. |
| 10 | Tue | Nov 10 | Pricing. Per seat, per task, or per outcome, and why agents break seat pricing | 3 distribution experiments run, with numbers | Talk, round, deep dives. Mandatory 1:1s. |
| 11 | Tue | Nov 17 | Trust and the enterprise objection. Permissions, prompt injection, audit logs, what a buyer asks before letting your agent touch their data | A price quoted to a real user, and their answer | Talk, round, deep dives. Last deep dive week. |
| 12 | Tue | Nov 24 | No talk, no round. Rehearsal Demo Day. 3 minutes per team, hard clock, full-class critique | Deck v1. Last weekly update. | Full-class critique |
| 13 | Tue | Dec 1 | No class. Mandatory 1:1s all day | Deck v2, rehearsed | Calendar-booked 1:1s, no session |
| 14 | Tue | Dec 8 | Demo Day. Outside judges. 3 minutes plus 2 minutes of questions | Final submission due one week later | Demo Day |