AI Builders 2027: Build Useful Things with AI
Yuzheng Sun, PhD
Cornell PhD · ex Statsig—acq. by OpenAI
Yan Wang, PhD
Columbia PhD · 40 AI papers · 3K+ cites
Make what lasts.
AI can make a first version. What lasts takes judgment and architect: choosing a problem worth solving, making the result reliable, and keep improving it.
Every useful AI system must answer two questions: Is it worth building? Can it work reliably?
AI Builders brings together two forms of judgment rarely taught together.
Yuzheng, a Cornell economics PhD, was an Amazon economist, Meta data scientist, Tencent AI director, and principal at Statsig (acq. by OpenAI).
Yan, a Columbia EE PhD, has nearly 40 papers at top conferences including CVPR, NeurIPS, and KDD, 3,000+ citations, and production and open-source systems.
Learners from OpenAI, Anthropic, Google, Meta and other leading companies come to AI Builders. They do not need another tool tour. They come to sharpen the judgment tools cannot give them.
Live, you watch both judgments applied to unfamiliar problems, then scope, test, and repair your own system with us.
We take AI seriously—and teaching it just as seriously.
Bring one useful problem. Leave with a reliable system and a better way to build the next one.
Iterated over 2½ years, 13 cohorts, and 3,000+ paid learners. Rated 5.0/5. Hard-won principles. Fully rebuilt for 2027.
What you’ll learn
Bring a useful problem. Leave with an AI system you can use, explain, and keep improving—and a repeatable way to build the next one.
Find the overlap between a real need and what AI can do well.
Turn an ambiguous goal into clear outcomes, constraints, and tests.
Keep the decisions that need your judgment. Give AI the rest.
Understand how context, memory, tools, and models work together.
Separate model errors from failures in data, code, tools, or orchestration.
Know when a simpler workflow is better than another agent or framework.
Go beyond chat: use APIs, connect tools and data, and put AI to work on real tasks.
Practice with five guided projects before choosing your own capstone.
Turn your capstone into a system you can use, explain, and keep improving.
Define success before you ask AI to do the work.
Use examples, logs, tests, and lightweight evaluation to make failures visible.
Fix the prompt, context, data, tools, or workflow—whichever is actually broken.
True progress is not learning new tools or new knowledge. True progress is a new way of thinking and doing things, permenantly.
Learn the method. Build new habits. Update your mindset. That's where true transformation comes from.
Who this course is for
You already build with AI and want to turn results that work once into systems you can explain, diagnose, and trust.
You solve real problems at work and want AI to carry more of the work—without losing control of the result.
You have hard-won expertise and want to turn it into an AI product or system that other people can actually use.
Prerequisites
Bring a useful problem and a willingness to build
No professional coding experience is required. Basic Python helps with the capstone. What matters is a willingness to build, inspect, and revise.
What's included
Live sessions
Learn directly from Yuzheng Sun, PhD & Yan Wang, PhD in a real-time, interactive format.
Two intensive live workshops on the core AI Builders method
The library gives you the material. Live, you see the judgment before the answer. Across two interactive, two-hour workshops, Yuzheng and Yan work through unfamiliar problems, explain what they use and reject, and show what would change their minds. Bring your questions, project scope, and failures to the optional 90-minute office hour.
Five guided builds, then a capstone of your own
The guided builds let you practice before choosing your own problem. Between sessions, scope a useful problem, build a first version, and bring failures back for diagnosis. A focused capstone may take 5 hours; a more ambitious system may take up to 30. We give feedback as it develops. Plan at least 3 hours a week for project work.
Lifetime access to AI Builders 2027
Your live enrollment also includes lifetime access to the complete self-paced course and all future updates. The 90 lessons shown in the syllabus form a long-term reference library, not a two-week viewing assignment. Revisit the tutorials and roughly 20-hour library whenever a new project calls for them.
One year of Stay Superlinear membership
Continue with Q&A from instructors and peers, project feedback, member programs, and curated tools in our paid learning community. Your enrollment also includes ongoing free access to our private AI Builders network of 20,000+ learners and practitioners.
Certificate of completion
Share your certificate and project work with your employer or on LinkedIn.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Course syllabus
3 live sessions • 90 lessons • 5 projects
Week 1
Oct
17
Live Sessions - Week 1
Automate Your Work with GenAI
GenAI Internals and Best Practices
Lightweight Projects: Bridging GUIs and APIs
Week 2
Oct
24
[Optional] Office Hour
Oct
25
Live Session - Week 2
Effective Management of GenAI
Become Future Proof
Free resource
The MCP Myth: Do You Really Need a Unified Agentic Protocol?
Why was MCP created in the first place?
Understand the hidden business and technical motivations behind the push for a unified protocol in the LLM ecosystem.
How can you strategically leverage MCP?
Gain practical insights into making informed choices about MCP adoption based on real-world trade-offs.
When should you actually avoid MCP?
Discover scenarios where adopting MCP could slow you down rather than accelerate your development.
Schedule
Live sessions
5-6 hrs
Two required 2-hour workshops plus one optional 90-minute office hour.
Sat, Oct 17
11:00 PM—1:00 AM (UTC)
Sat, Oct 24
11:00 PM—12:30 AM (UTC)
Sun, Oct 25
11:00 PM—1:00 AM (UTC)
Guided project work
6 hrs
Plan at least 3 hours a week to build along, test the ideas, and bring real questions to the cohort.
Capstone project
5-30 hrs
Scope varies. Complete the capstone, and you’ll leave with an AI system built around a useful problem of your own. We provide feedback as you build.
Testimonials
There are many courses teaching specific tricks. This course helps you build the right mindset and teach yourself more effectively.
Shuyang
Member of Technical Staff · OpenAI | Former Sr. AI Manager · UberThe tools, patterns, and lightweight project workflows are things I can use right away in my day-to-day work as a software engineer.
Ye
Staff Engineer · SeeqYan is very articulate in teaching and very skillful in practical applications of genAI. Yuzheng is very good at challenging your default thinking and offers good insights on genAI as a technology itself.
Nana
Principal Data Scientist · DIRECTVIf you’re new to coding, this course is a great entry point. If you’re more experienced, you’ll gain a deeper understanding of how AI fits into real product workflows.
Peter Zhang
Student · UCLInstead of collecting tips and theory, we spent our time building real workflows and shipping small, production-grade projects.
Janet 🇮🇪
Technology Advisor · Connectd
Built over two and a half years. Rebuilt for how AI works now.
An AI training session at DoorDash in Seattle.
What we learn in the field comes back to the course.
AI Builders has been refined through two and a half years and 13 rounds of teaching. We also run AI training for teams at Tencent, Meituan, Xiaohongshu, Pinterest, and DoorDash.
Real projects and business constraints quickly reveal what holds up beyond a demo. Every new question, failure, and example improves what we teach and how we teach it.
What does a useful AI system look like?
There is no single capstone template. A useful project begins with a real constraint.
AI job radar: scraping 150 listings took 10 minutes; analyzing them took an hour. The learner changed a planned daily “strategic radar” into a weekly job-search system.
AI agent debugger: describing a GUI failure to an agent led to guesswork. The learner built a way for the agent to inspect function-activity records and see what actually ran.
A hearing app for a grandmother: a simple family need became 42 iterations as real use exposed what the first version missed.
These are three of 55 AI Architect course projects. Different starting points, same discipline: choose a useful problem, build, observe what fails, and keep improving.
Browse the public project collection: https://www.superlinear.academy/c/share-your-projects/ai-architect-summary
Frequently asked questions
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedTeam discount
Learn with your teammates
Save 20%+ when 2 or more teammates enroll in the same cohort.
Save 20%+ with a teamPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort$1,999
USD