Week 3 Self-Paced Module: How Engineering Teams Work with AI
CSIC30216 · Fall 2026
Overview
Today's lab asked you to write your team's AI usage guidelines and an evaluation plan under time pressure — which means what you produced is exactly that: a first draft. This module gives you the real-world evidence to calibrate against: how actual engineering teams write guidelines like these, run evaluations, and handle the AI-specific risks of a shipped system. By the end, you should be able to reread your own team's draft and tell whether it would survive contact with a new teammate joining in Week 10 — or with a customer who took its output at face value.
Why This Matters
DORA's 2025 State of AI-assisted Software Development report — this week's required reading — found that AI functions as an amplifier: it strengthens high-performing teams' existing practices and magnifies the dysfunctions of struggling ones. Teams with a “clear and communicated AI stance” see that amplification lean positive, with AI's benefits on individual effectiveness and organizational performance amplified, and friction reduced — compared to teams that leave AI use to individual discretion. This is the empirical case for the exercise today's lab put you through, and the report's two named case studies — Adidas and Booking.com — show what that looks like at the scale of 1,000+ and 3,000+ developers respectively.
Format and Timing
- Framing video (posted on E3) — link coming soon; check back before starting the module.
- Required reading — selected sections of the DORA 2025 report (~20 minutes). Linked below.
- Multiple choice questions (單一選擇題) — four questions on E3, about 10 minutes. Submit before the end of the week to mark the module complete.
- Reflection prompt — not submitted, but discuss with your team before CP1 (~20 minutes).
- Total estimated time: ~45–60 minutes
What You'll Learn
- Evaluation approaches in depth — what structured observation, pre/post surveys, and brief interviews actually look like when run within a semester project, and why you need at least one of them with people outside your team.
- How real teams adopt AI tooling — DORA's finding that a clear, communicated AI stance amplifies AI's benefits and reduces friction, and why the specific form of a team's norm matters less than the fact of having one at all.
- What the case studies show — Adidas saw 20–30% productivity gains from AI in loosely coupled architecture, but little benefit in teams tightly coupled to legacy systems; Booking.com's adoption stayed uneven until they invested in training developers to give AI more explicit instructions and context, after which merge requests rose by up to 30%. Neither story is really about the AI tool itself — one is about architecture, the other about training.
- Disclosure and liability — the Air Canada chatbot case, where a court held the company responsible for incorrect information its own system gave a customer. “The AI said it” is not a defence, and documenting your design decisions is your primary protection.
Required Materials
- Framing video: posted on E3 under Week 3 Self-paced Modules.
- Reading: DORA, “State of AI-assisted Software Development” (2025) — also available via E3 for download. Read the Executive summary (pp. 3–7), the “AI adoption and use” chapter (pp. 23–32), and the “Clear and communicated AI stance” section of the DORA AI Capabilities Model chapter (pp. 50–53). This is a public report — no account or download form required.
- Multiple choice questions: posted on E3 under Week 3 Self-paced Modules Questions.
Optional Materials
If you want to go deeper on any of these topics (not required, but helpful if you're curious):
- Spotify Engineering, “AI Changed How Spotify Builds: What We Learned (and Fixed) About Quality at Higher Velocity,” 2026 — Spotify's own account of adopting AI in software engineering, published only days before this module.
- Travis Media, “AI Code Is Cheap. Open Source Maintainers Pay the Bill” — how open source maintainers are struggling with AI-generated contributions.
- open-source-ai-contribution-policies (GitHub repository) — a running list of how different open source projects have written their own policies on AI-generated contributions.
(Optional) Reflection Prompt
Open your team's draft guidelines and evaluation plan from today's lab. Read them as if you are a new engineer joining the team in Week 10: could you follow them without asking anyone a question? Then, after finishing the readings, ask which of your guidelines is most likely to fail under pressure. You don't need to submit an answer, but discuss it with your team before CP1 — teams that do this consistently submit stronger checkpoints.