Software Engineering Fundamentals
Seventy years of abstraction in two slides, then the argument for why building with agents has no ceiling.
Class recording
The lecture recording posts here after the class is taught.
Class summary
History first, compressed to eight years — 1952, 1957, 1959, 1972, 1983, 1995, 2012, 2025 — and each one is asked the same two questions: what did it hide, and who could suddenly program that could not before. That ladder then becomes a literal pyramid, where width is what you can still express and height is how easy it is to write. Every rung up trades expressiveness for ease. Agents arrive off the side of the pyramid: they emit ordinary code, so they inherit no ceiling. That is the argument for building with agents rather than with no-code tools, and the deck cashes it in immediately — the next slide is a screenshot of this deck proving it built itself.
The second half is the stakes. Eight public SaaS companies indexed to their own January 2025 close show a market that has already repriced per-seat software. MIT Project NANDA’s GenAI Divide separates the 90% of employees using chatbots from the 5% of custom, workflow-embedded pilots that actually move P&L in six months. Six real forward-deployed-engineer job listings close the argument in salary terms — and the Skill Map slides then walk what the role requires: how an LLM works and where it leaks, how facts get in and why one retriever is not enough, agent loops and the connector tax, ML foundations, and evaluation-driven development.
The class closes on the engagement kickoff. A real business — The Gold Trader (thegoldtraderstore.com), a jewelry company whose intake-to-invoice loop runs through QuickBooks and takes too long — presents its problem in class, and the deck draws it as four boxes: take it in, create the invoice, send it, confirm it. The class splits into self-selected teams of three (sign-ups open on the Connect.AI dashboard the morning after class), each team carrying one trainer, one technical scoper and one builder. The following Tuesday every team pitches its solution for five minutes, the room votes, and the winning team does the follow-up call and builds the real solution. A three-question multiple-choice pop quiz — one per part of the deck — recaps the argument, the Skill Map and the engagement, and the session closes on a thank-you.