Fundamental CS for AI Era

Building human-centered digital products in the AI era

Zain Fathoni

Zain Fathoni

Senior Software Engineer based in Yogyakarta, Indonesia. Previously backend, manager, frontend, now fullstack.

Community builder (Ketua VibeFromCafe.id) · Agentic Engineering Practitioner.

https://zainfathoni.com

4 July 2026 | Fundamental CS for AI Era

P2 · Common Ground

AI makes code generation fast and accessible

Anyone in this room can ship working code today — developer or not.

Generating code is no longer the hard part. We all agree on that.

Development loop with AI

4 July 2026 | Fundamental CS for AI Era

P3 · Coming Problem

Understanding becomes the bottleneck

AI generates code faster than we can understand it.

The constraint moves from writing code to understanding what changed, why it works, and what it might break.

Thariq, on Fable 5: work quality is now “bottlenecked by my ability to clarify my unknowns.” — Finding Your Unknowns

Geoffrey Litt · 2 July 2026

Hot take: I think it's still important to understand the code that our agents write!

Understanding matters not just to verify, but to participate. — “Understanding is the new bottleneck”
x.com/geoffreylitt/status/2072522251300409556

4 July 2026 | Fundamental CS for AI Era

P4 · Emotional Win

Imagine calm confidence in any codebase

Opening an unfamiliar repository without dread.

Reviewing an AI-generated PR and knowing what to look at first.

That feeling is trainable — and AI itself can train it.

AI
×
CS

The win is not “skip fundamentals.” The win is “learn fundamentals faster, with feedback.”

4 July 2026 | Fundamental CS for AI Era

P5 · False Hope

“AI means I can skip fundamentals”

Tempting — but this is what it looks like from the receiving end:

Dex Horthy · 20 June 2026

If people are tokenmaxxing bugs into production with kLOC PRs that they didn't read themselves, those people shouldn't have jobs.

“Coaching juniors on SWE fundamentals is hard work and it takes time.”
x.com/dexhorthy/status/2068433796182270203

4 July 2026 | Fundamental CS for AI Era

P6 · Audacious Reality

Fundamentals are the AI multiplier

Data structures, algorithms, systems thinking, debugging, trade-offs.

They give you the taste to judge AI output and the words to steer it.

Gergely Orosz · 2 July 2026

If you don't know what good code looks like, you will have no idea if what the models generate are any good.

“This is exactly why experienced software engineers are valuable and will be valuable.” (via @mitchellh)
x.com/GergelyOrosz/status/2072831495463428580

4 July 2026 | Fundamental CS for AI Era

P7 · We Can Do This

Guard AI's work at three time horizons

Each guard is an open-source skill you can install today.

The first two guard the code. The third guards you.

⏱️ Now — E2E evidence

/pr-e2e-evidence — prove the change works today, in a real browser, before it ships.

📆 Near future — unit tests

/tdd — lock the behavior in so it can't quietly regress. (Matt Pocock, aihero.dev/skills)

🧠 Always — understanding

/teach — keep yourself able to chime in whenever future issues or plans arise.

4 July 2026 | Fundamental CS for AI Era

P8 · Call To Action

Start with the long-term guard: /teach

An open-source skill that turns understanding into a persistent learning workspace.

  1. Grab it: github.com/zainfathoni/agent-workflows
  2. Run /teach with any topic you want to learn
  3. Answer the interview, take one lesson and its quiz
  4. Return anytime — it remembers where you left off
🎯 MISSION.md — why you're learning
📚 lessons/ — short, single-win lessons
📝 quizzes — speed regulators for real understanding
🗂️ learning-records/ — evidenced progress

Quizzes as speed regulators — the same technique Geoffrey Litt proposes for understanding AI code.

4 July 2026 | Fundamental CS for AI Era

P9 · Early Benefits

You can learn anything you want

Not just CS fundamentals. The same workspace teaches you:

  • a new framework or language
  • an unfamiliar codebase, one PR at a time
  • algorithms and system design
  • even non-code topics
Instead of:

passive tutorials you abandon by chapter three.

You get:

a personal curriculum with lessons, quizzes, and memory — your own Brilliant, crafted from Papert-style micro-worlds.

Live proof: ai.zainf.dev — my own AI-inference learning lab. That journey just started, in public.

4 July 2026 | Fundamental CS for AI Era

P10 · Long Win

Engineers become better problem framers

In the AI era, human-centered builders are not the people who type the most code.

They are the people who can frame problems clearly enough for humans and machines to solve together.

Gogo (@lwastuargo) · 1 July 2026

It's actually a huge relief that the future of software engineering… is still software engineering.

“Even expensive loop engineering with a state-of-the-art model can't out-engineer bad system design.”
x.com/lwastuargo/status/2072256396633260350

4 July 2026 | Fundamental CS for AI Era

Live Demo

Let's learn something together — for real, for the next hour.

🙋 You pick the topic. You answer the quizzes.

4 July 2026 | Fundamental CS for AI Era

How the next hour works

  1. Pick (10 min) — you shout topics, we vote on one
  2. Mission (15 min) — /teach interviews us; you answer, I type
  3. Learn (20 min) — we take the first generated lesson together
  4. Quiz (15 min) — you answer; wrong answers steer the next lesson
Ground rule:

I will not edit the AI's questions or lessons. What you see is what the skill does.

Case study, if the room is shy:

Big-O notation — from my earlier slides at zainf.dev/big-o, with the workspace already live at big-o.zainf.dev.

4 July 2026 | Fundamental CS for AI Era

What we'll watch it build

A learning workspace, growing live on screen:

  • MISSION.md — the room's shared goal
  • RESOURCES.md — sources we trust
  • lessons/*.html — one win per lesson
  • learning-records/*.md — proof of what we learned

Understanding matters not just to verify, but to participate.

Geoffrey Litt — this hour is that idea, practiced.

Latest demo lessons: SEO · UTUNE AI Lessons

4 July 2026 | Fundamental CS for AI Era

References

Fundamentals are not the past.

They are how we steer the future.

4 July 2026 | Fundamental CS for AI Era

QR code for zainf.dev/fundamental-cs-ai

4 July 2026 | Fundamental CS for AI Era

P1 — Who & What / Clarity. Fundamental Computer Science for AI Era. Frame with the product-building tagline.

P2 — Common Ground: AI makes code generation fast and accessible. Everyone here has felt this. Establish agreement before introducing tension.

P3 — Coming Problem: understanding becomes the bottleneck. Litt's tweet (333.8K views) + his article. Key idea: understand to participate, not merely verify. Second voice: Thariq's "A Field Guide to Fable: Finding Your Unknowns" (651K views) — the map (prompts, skills, context) is not the territory (codebase, real constraints); quality is bottlenecked by your ability to clarify your unknowns. x.com/trq212/status/2073100352921215386

P4 — Emotional Win: calm confidence in any codebase. Paint the feeling before the argument. This is what fundamentals buy you emotionally. "Trainable" is Thariq's exact conclusion: "reducing your unknowns is the skill of agentic coding — a skill you can improve at, by working with Claude." Confidence = few unknowns, and unknowns shrink with practice.

P5 — False Hope: "AI means I can skip fundamentals." Dex's rant against lazy engineers. Skipping fundamentals taxes your seniors and your future self.

P6 — Audacious Reality: fundamentals are the AI multiplier. Gergely (via Mitchell Hashimoto): knowing what good looks like is the judging skill AI can't replace. Same argument from Thariq: "the best agentic coders are good but have relatively few unknowns" — they know what they want in detail, deeply in-sync with both the codebase and the model's behaviors. Fundamentals are what convert unknown unknowns into known knowns.

P7 — We Can Do This: three guards, three time horizons. E2E = works NOW. TDD = won't regress NEAR. Understanding = you can ALWAYS participate. /pr-e2e-evidence and /teach: github.com/zainfathoni/agent-workflows. /tdd: Matt Pocock's aihero.dev/skills (/teach also started as Matt's — credit him out loud).

P8 — Call To Action: the first doable step is using the /teach skill. Open-sourced at github.com/zainfathoni/agent-workflows. Concrete, tonight-sized steps.

P9 — Early Benefits: the audience can learn anything they want. The immediate reward of the CTA: a tutor that adapts and remembers, for any topic. "Your own Brilliant" via micro-worlds — Litt's technique (inspired by Seymour Papert): x.com/geoffreylitt/status/2072522314240127393. Show ai.zainf.dev briefly if time allows: lessons, learning records, experiments — the /teach workspace shape, applied to AI inference.

P10 — Long Win: engineers become better problem framers. Gogo's lessons from building Anna: the highest-leverage work is the right system design. Litt: "The point was always to augment, not just automate." Tie back to human-centered digital products.

Section break. ~15 minutes of slides done; now ~60 minutes of hands-on demo with audience participation. Energy shift: close the laptop lid metaphorically, open the terminal.

Demo agenda. Timebox each round out loud. If the room is quiet, use the Big-O case study: the workspace is pre-seeded at ~/Code/GitHub/zainfathoni/big-o (live at big-o.zainf.dev) — MISSION.md, RESOURCES.md, and lesson 1 with its quiz are ready; run the quiz, then build lesson 2 live from the wrong answers (best/worst/average case, then Θ vs O vs Ω). Bonus live moment: RESOURCES.md records an erratum in my own 2020 deck (slide 34, Big-Ω). Round 2 (Mission) is where audience participation locks in — their answers shape MISSION.md live.

Anatomy slide — narrate the file tree as it grows. Connect back to Litt: explainer docs, quizzes as speed regulators, micro-worlds ("inspired by the visionary educator Seymour Papert" — x.com/geoffreylitt/status/2072522314240127393). The demo IS the argument of the talk. This hour is also Thariq's playbook practiced: interviews and quizzes to surface unknown unknowns.

Reference slide. Keep citations visible without turning the talk into a literature review.