The workflow to become a strong software engineer in the AI era.

Not a language. Not a stack. A new developer workflow: how you code with agents, how you understand the systems they hand you, how you keep learning while you build.

Scaffold is free and open source. · Taught at UC Berkeley.

What makes a software engineer strong now

A strong software engineer plans with their agent first, working out any disagreement before it builds, then reviews what it executed against that plan. They keep growing as the agent does, sharper with everything they ship, not weaker.

Anthropic ran a randomized trial with 52 professional developers, most with four or more years of experience, half coding with an AI assistant, half without. Those using AI scored 17 percentage points lower on a quiz about the code they'd just written: 2 grade points in the paper's terms, roughly two letter grades.

The ones who asked it to explain, and asked why, held their scores. The ones who only asked for code clustered at the bottom.

Anthropic, “How AI Impacts Skill Formation” (Shen & Tamkin, 2026). arXiv:2601.20245.

That habit breaks the day you're tired, rushed, or sure you already know the answer.

A plan-first workflow turns judgment into something you do every time. The build waits on your yes, not the agent's.

The workflow: Scaffold

Scaffold is a plan-first developer workflow: a free skill that builds real software and grows your judgment at equal weight. Like plan mode, except you call the change first: Scaffold answers with its own plan as a diff against yours, argues where it disagrees, then builds; anything you want to write, you write. Every commit reports what you've shown you understand, in a dated log, one page per concept. The seven building blocks are a philosophy for designing systems, one you can use at any experience level.

Free to try today.

What this workflow can do

Wherever you're starting from, here's what changes once Scaffold is your workflow.

Where you're starting Before After
Junior Developer You hand the task to the AI to keep up with the velocity the industry expects of you, and your own understanding falls behind what you ship. You set your plan first, spar with the AI over it, still execute, and build your understanding while you ship, instead of losing it.
Vibe Coder You ship fast with an agent, and when it breaks at 2 am, you can't solve it. You can say why it works, and see where it breaks, before it ever wakes you at 2 am.
Career Changer You can code a little, and you want to learn how real systems get built with AI. There has been no clear path in. You learn how real systems are built, with AI and a plan-first workflow, a faster ramp than anything before it.

Junior Developer

Before

You hand the task to the AI to keep up with the velocity the industry expects of you, and your own understanding falls behind what you ship.

After

You set your plan first, spar with the AI over it, still execute, and build your understanding while you ship, instead of losing it.

Vibe Coder

Before

You ship fast with an agent, and when it breaks at 2 am, you can't solve it.

After

You can say why it works, and see where it breaks, before it ever wakes you at 2 am.

Career Changer

Before

You can code a little, and you want to learn how real systems get built with AI. There has been no clear path in.

After

You learn how real systems are built, with AI and a plan-first workflow, a faster ramp than anything before it.

The knowledge: the seven building blocks

Service
Worker
Key-Value Store
File Store
Queue
Relational Database
Vector Database

Service · Worker · Key-Value Store · File Store · Queue · Relational Database · Vector Database. Scaffold's loop asks you to predict a plan and argue with the agent's diff, and you can only push back on a plan you have a mental model for. These seven pieces, plus three outside forces, are that mental model: every system you'll touch, from a todo app to Instagram, is some arrangement of them.

When Scaffold writes up your plan or your architecture, these are the words it uses. Once you can name them, you can read any plan and spot what's missing.

Free to learn today.

No account required.
See it applied

The seven blocks, applied

The same seven building blocks design fifteen real applications. Each one is broken down block by block, free to watch on video.

See all 15 case studies →

The guided path: the courses

The courses teach this way of working, in order: Course 0, then Courses I-IV. One ramp, not two separate products.

Start here

Course 0

Course 0 teaches the plan-first workflow: commanding an AI agent the way a senior engineer would. The plan before the code, the brief, the review that catches what it missed.

8 lessons, 2 labs, 3 challenges.

$99

Start Course 0, $99

Courses I-IV

The seven building blocks, applied to real systems. Four courses, each 8-11 video lessons, 2 labs, feedback on your answers, and a 3-part design challenge.

$299, one bundle

Explore Courses I-IV

Want the workflow first? Scaffold is free.

Taught at UC Berkeley

Built from ten years of teaching engineers how real systems get designed. The same workflow, now available to everyone.

Kay Ashaolu
Kay Ashaolu

Continuing Lecturer at UC Berkeley School of Information. 15+ years as a software engineer and engineering manager, including AncestryDNA and Morgan Stanley. Now teaching engineers how to direct AI and still get stronger doing it.

Learn more →

What students say

“The way Kay integrates AI tools into his teaching gave me the confidence I once had in software development.”

Student

“The way Kay teaches software development has helped me use tools like Claude Code more effectively.”

Student

“I find your teaching very valuable and needed for me to get back into the job market.”

Course Student

Frequently Asked Questions

Do I need programming experience?

This is built for people who already write code: if you can read a function and ship a change, you're in. The framework isn't about syntax or any one language. It's about how you direct an agent, how you judge what it built, and how you keep learning while you do it.

What's free vs paid?

The free option above costs nothing: the 7 building-block articles, the full video series, the building blocks game, Design with Blocks, and Scaffold.

Course 0: $99. Command an AI agent on real work.

Courses I-IV: $299, one bundle. Build the judgment: 8-11 video lessons, interactive Python labs, feedback on your answers, and a 3-part design challenge, in every one of the four courses. Already own Course I? The same $299 also unlocks Courses II-IV and Course 0.

How does this help me work with AI?

AI is good at writing code. It can't tell you whether the design is right. That call is yours. When you understand the building blocks, you can tell an agent exactly what to build: Service for the API, Relational Database for users, Key-Value Store for sessions, instead of “build me an app” and hoping.

How long does one of your courses take?

Course 0: 8 lessons, 2 labs, 3 challenges. Courses I-IV: a 10-week experience at 1-2 hours a week. Both are self-paced, and everything stays yours for good.

What is Scaffold?

A free skill. Before it writes a line, it asks what you think should change, and why, then shows its own plan as a diff against yours. It runs the workflow this page teaches, free, on your own code, and you can try it today without buying anything. Install it here.

What are the 7 building blocks?

Service, Worker, Key-Value Store, File Store, Queue, Relational Database, Vector Database. The vocabulary for reading any system, and for judging what an agent built. Read the 7 free articles, no account required.

What do the courses give me that the free material doesn't?

Course 0 teaches the workflow on real work: labs and challenges, not just reading about it. Courses I-IV teach the seven building blocks applied to real systems: labs, feedback on your answers, and a design challenge in every course. Both are the guided, ordered path, if that's the way you learn best.

Is AI making developers worse?

Not exactly. In DX's study of 500+ engineering organizations, more than half of the code written last quarter was AI-authored. DX's own research says: “While AI is making it easier for developers to understand what's in front of them, the data suggests that they have less trust in the code they are actually pushing into production” (DX, “The State of AI Impact in Engineering: Q2 2026”). In Anthropic's trial, not everyone who used AI scored lower: the developers who held their scores were the ones who used it to understand the code they were writing, not just to produce it. That's the habit this workflow teaches.