Course Launch Completion and Scale Your Intelligence

Wednesday focused on questions about the AI Product

• We walked through the AI Product Student Guide in depth.
• We discussed stakeholder paths, shared stakeholders, scope, evidence, and grading.
• The guide and milestone packages remain the authoritative detailed instructions.
• Course material stated or explained in lecture—even when it does not appear on a slide—is fair game for quizzes and exams.

Oral quizzes are closed-book during each attempt

• During an attempt: use no notes, AI, websites, classmates, or other sources.
• Between attempts: use AI, notes, course materials, and discussion to learn before retaking.
• Never share or receive specific quiz or exam questions, prompts, topics that appeared, or other assessment content.
• General discussion of course topics is welcome.

Using prohibited help—or sharing or receiving assessment content—will result in an F in the course under the syllabus policy.

We will now finish the syllabus and schedule walkthrough

The syllabus is required reading. Quiz questions may assess syllabus content.

Open-to-all study groups can earn up to 3 bonus points

• Earn 0.3 percentage points per qualifying week; ten weeks reaches the maximum.
• A group has at least two students, remains open to the class, and meets about 30–45 minutes.
• Submit one Gradescope group list, a meeting photo or screenshot, and an audio or video recording.
• Meetings are entirely in person or entirely online—not hybrid—and use one lead contact.
• Automated checks support manual review; ordinary warnings retain credit, but an unusable recording must be resubmitted.

AI scales the quality of your judgment

• Weak understanding and poor discernment can scale confusion, drift, or harm.
• Strong understanding and careful judgment can scale useful work.
• The human remains central: choose values, define what is good, shape the environment, and make consequential tradeoffs.

Chat gathers and synthesizes information; agents change the work

• Chat makes repeatable processes difficult because the work remains inside a conversation.
• Use chat for bounded information tasks—not sustained project work.
• Move consequential, multi-step work into an agentic environment.

Cheap code changes the engineering bottleneck

• Coding agents can produce and revise implementation extremely quickly.
• In a large system, more output creates more interactions, decisions, and possible failures.
• Requirements, architecture, evidence, deployment, maintenance, and accountability remain.

Scale creates a reasoning-horizon problem

• A large system exceeds what any one human or agent can actively reason over.
• Larger context windows and retrieval move the boundary; they do not remove finite reasoning.
• Without useful external representations, each new task reconstructs system knowledge from code and history.

MAGE answers the scaling problem with an engineering method

MAGE = Model-Based Agentic Software Engineering, a method developed by James C. Davis for engineering large software systems with coding agents.

1: Modeling makes intent explicit at useful scales

“A model is a purposeful reduction of a system. It preserves the information one engineering question needs and suppresses the detail that question does not.” — James C. Davis, MAGE, p. 38

• Explicit: Make engineering knowledge and intent explicit.
• Compact: Suppress details that do not matter at that level or for that question.
• Understandable: Make the representation usable by both people and machines.

1: Modeling uses different models for different questions

2: Alignment gives modeled obligations authority

Models can state what should hold, but representation alone does not make the work follow the model. Alignment gives selected obligations consequence.

3: Governance conversion lets judgment accumulate as engineering capital

MAGE builds a governed engineering environment

Effective CEOs already scale judgment through people and culture

• Focus on consequential choices: strategy, values, and tradeoffs.
• Delegate logistics, reporting, coordination, and routine decisions.
• Shape culture and feedback so others can act well without constant oversight.

Agents make this leverage accessible—but not automatically good

• Agents make delegation far cheaper and more widely available.
• Give them explicit goals and boundaries; require evidence and escalation.
• Start with bounded work, preserve lessons, and expand incrementally.

Offload work that does not require your attention so you can invest it where your judgment creates the most value.