Nov 9 to 15, 2026
Week 6 Radar
Your agent heard you. It just didn't listen.
Alex Garrett-Smith · Nov 13
Direct AI. Don't get led by it.
AI solved the easy half of your job. Unlearn trains the other half, so you build better software and grow as an engineer.
Pick up where you left off4-day streak
2:14 / 5:48Up nextLock In Technical Decisions Before AI Makes Them for YouDesign a Feature AI Can Execute Without Guessing · Lesson 2 of 7
Resume3 min left
Product EngineerStep 3 of 9 · Plan before it builds
Spec a Product
OpenSpecDesign a Feature
Build Features
Review AI CodeThis week on your RadarWeek of Nov 9
Comparison · 3 min readRequirement in the Prompt vs Requirement in a Test
TLDR · 3:04What Are Cached Tokens?
TLDR · 3:57Where Your Agent Saves the Commands You ApproveWorkshopsAll workshops
LiveSpec-Driven Development: From Idea to Production with AIAG
Generative UI: Building Apps That Compose ThemselvesAlex Garrett-Smith
AG
Code Review at Agent ScaleAlex Garrett-Smith
ToolkitAll tools
Spec Generator SkillTurn a vague idea into a detailed product spec
$
npx skills add unlearndev/skillsCopySavedWaiting to go live
Spec Auditor MCP serverMCP server
Unlearn plugin for Claude CodePlugin
Docs WriterAgent
From the team that trained 2,000,000+ developers at
Courses
Every course opens with a free lesson.
- Workflow · 33mSpec a Product Before AI Builds ItFree
- Workflow · 24mDesign a Feature AI Can Execute Without GuessingPlus
- Deep dive · 56mSpec-Driven Development with OpenSpecFree
- Workflow · 1h 38mBuild Features with an AI AgentPlus
Full course · 2h 22mTurbo Mode: Optimizing Productivity with AI Tools and AgentsPlus- Workflow · 1h 34mReviewing AI-Written CodePlus
Radar
What changed this week, and what's worth your time. Free for everyone.
Comparison · 3 min read
Requirement in the Prompt vs Requirement in a Test
An agent can talk itself out of a change that's only in the prompt, so I'd add a test that fails until it's done, and keep one or two it can't see.
TLDR · 3:04
What Are Cached Tokens?
Prompt caching reuses matching prompt tokens from recent requests at a lower cache-read price, so you can reduce input costs on repeated requests.
TLDR · 3:57
Where Your Agent Saves the Commands You Approve
Claude Code saves approved commands in an allow-list file, so check it for secrets and avoid approving commands with passwords.
Coming soon
This Week in AI Dev
A weekly episode with Alex Garrett-Smith on new releases, new tools, and what they mean for how you work.
The Unlearn Podcast
Long conversations with engineers who changed how they work with AI.
Roles
Which engineer are you becoming? A role is a route through the courses in the order you use them, and finishing it earns the role certificate.
- Live
Product Engineer
Own what gets built and how, end to end, with agents doing the typing.
For decades the code was the bottleneck, so the work got split. Someone wrote the ticket and someone else typed it out. Agents made the typing nearly free, and the gap between the two is where things break now. A product engineer closes it, turning a product goal into a spec an agent can't misread, directing the build, and owning what ships.
9 steps 78 lessons The Product Engineer certificate
- Coming soon
AI Engineer
Calling a model takes one line. Making it reliable inside a real product, with retrieval, evals and an interface that holds up, takes an engineer.
- Coming soon
Architect
Agents follow whatever structure they find. Architects design systems where the right change is easy and the wrong one is hard.
- Coming soon
AI Automation Engineer
Most automation doesn't need a model. You know when a script or a hook will do, put agents only where the work needs judgment, and build pipelines that run on their own.
- Coming soon
Engineering Lead
Tools don't change teams, habits do. You take a team from skeptical to effective without letting quality slip.
- Coming soon
Design Engineer
When the code writes itself, how it feels becomes the product. Design engineers shape the experience and ship it, with no handoff in between.
How it works
0:44 / 2:47
Lesson 3 of 10Shaping the Specification
In this lesson, you'll take your rough specification and turn it into something far more solid by removing assumptions, adding missing details.
Synced to the video; select a line to jump there.
Questions about this lesson show up here.
Spec a Product Before AI Builds It
Workflow2 of 10 done
- Why Planning Beats Vibe Coding4:05
- Turn a Vague Idea Into a Structured Spec6:27
- Shaping the Specification2:47
- Give AI Technical Constraints from Day One1:17
- Tell AI What It's Not Allowed to Build2:48
5 more lessons
Keyboard shortcuts
- SpaceK
- Play or pause
- JL
- Back or forward 10 seconds
- T
- Transcript beside the video
- O
- Outline
- S
- Save this course
- F
- Fullscreen
- ShiftN
- Next lesson
- ?
- This list
That's the loop. A platform you work alongside, not a library you never finish.
Toolkit
The skills and starters from the courses, free and open source.
One command, every skill
Spec generator, code review, ship gate, triage and more, installed into any coding agent.
npx skills add unlearndev/skills- SkillSpec GeneratorTurn a vague idea into a detailed product spec
- SkillCode ReviewReview staged changes or a specific area of the codebase
- SkillShip GateScan the staged diff for secrets, debug code, and stray files and return a straight PASS or BLOCK
- Rule fileAGENTS.md StartersStarter AGENTS.md files for 13 web stacks: Laravel, Nuxt, Next.js, Rails, SvelteKit and more
- SkillTriageGroup a branch's changes into feature areas and assign risk tiers
- SkillZombiesSuggest the most relevant tests to write for a feature using the ZOMBIES heuristic
Blog
Short, opinionated posts on working with agents.
The people behind unlearn
A company that teaches developers for a living, and the engineer who leads the curriculum.
We've been teaching developers since 2017, and more than 2 million of them have learned with us. We run FrontendNation, one of the biggest online dev conferences out there, and the official Vue.js and Nuxt certifications, built hand in hand with the people who created those frameworks.
Unlearn is all of that, pointed at the part of the job AI changed.
Your instructor
Alex Garrett-Smith, Technical Education Lead at BitterBrains.
Built Codecourse into a 300,000-subscriber channel and a full learning platform, acquired in 2025. Now leads the curriculum and judgment frameworks behind Unlearn. Sixteen years teaching developers how to build, now writing the playbook for what comes next.
Free and Plus
Free gets you started. Plus gets you everything.
FreeFree to start
- Lesson 1 of every course, no account needed
- Selected courses in full, with a free account
- Role progress that follows you across devices
- The weekly Radar, every TLDR video and Versus
- Open-source agent tools, skills, starters and templates
- Every tutorial and every live workshop
PlusEverything in Free, plus
- Every lesson of every course
- Course and role certificates when you finish
- Unlearn apps like Spec Auditor
- New courses and roles as they ship, at no extra cost
Take the lead
Start free today, and upgrade to Plus when you want every course and every role.
Rather not make an account yet? The weekly Radar can come to your inbox instead. Get free content
FAQs
Do I need to pay to start?
No. Lesson 1 of every course is free without an account, and so are the Radar, the tutorials and the open-source tools. A free account adds selected courses in full and role progress, with no card. Plus unlocks every course, every role and the Unlearn apps.
Does it work with my stack?
Yes. The judgment applies to any stack and any AI tool. For the setup, the AGENTS.md starters cover 13 stacks, including Laravel, Nuxt, Next.js, Rails and SvelteKit, and the skills install into any coding agent with one command.
What's a role?
A route through the courses in the order you use them, built around the engineer you're becoming. Finish every step and you earn the role certificate. Product Engineer is live, and five more roles are on the way.
I already use AI every day. Why would I need this?
Using AI and trusting what it hands you are two different skills. In a survey of 3,080 developers, 44% rated themselves average at judging AI output. They ship faster and second-guess more.
If you already lean on AI, you'll get more out of Unlearn than someone just starting, because you're already making the calls it trains.
I've tried AI courses before. Why is this different?
Most of them teach a tool, and the tool changes next month. Unlearn teaches the calls you make around it, like what to spec, what to hand off, and what to reject in review. The tools part stays current too, because new courses and tools keep shipping.
Is it for my level?
If you're senior, you'll move through the principles fast and spend your time where it pushes you, like context engineering, spec-driven development and reviewing AI-written code. If you're earlier in your career, every course starts with the principle before the tool, so you build the judgment that usually takes years of mistakes.
I don't have much time. Will I actually use it?
Lessons run under seven minutes on average, and each one ends with something you use on your own project the same day. You don't have to finish a course to get value. One lesson on reviewing AI-written code changes how you read the next PR.
Questions about billing, teams or refunds? See pricing.