Feifan Zhou

Guardrails for Coding Agents: Pointing Them the Right Way

Stripe

Cornell University

Tanagram CEO & Cofounder

Learn about different tools that can be used to constrain agent output, ranging from deterministic to natural-language.

Ways to keep coding agents on-track

Each tool has different trade-offs and performance implications; many of them can be used together.

When and where to use each tool

Treat guardrails as first-class in your codebase. Share ownership across teams and update them as the code evolves.

Keeping guardrails up-to-date

Code is structured text with repeating patterns. Use tools to enforce what you want and migrate from what you don’t.

From guardrails to patterns

If you’re a software engineer or data scientist, chances are you’ve built or seen a proof of concept (POC) for a generative AI app. It’s exciting at first, an impressive demo that showcases the potential of large language models. Here’s the problem: that’s where it usually stops. These POCs often fail to scale into reliable, production-ready applications. The result? Endless iteration cycles, unreliable outputs, and frustration as teams struggle in what we call “POC purgatory.” It doesn’t have to be this way. With the right principles, processes, and workflows, you can build LLM applications that work in production, without overcomplicating pipelines or chasing trends. This course focuses on the iterative development, evaluation, and debugging practices that make AI systems production-ready. You'll learn how to build with state-of-the-art tools like RAG, fine-tuning, MCP, and agents, but the emphasis is on building scalable, robust systems, ensuring you leave with the mindset and skills to adapt as the technology evolves. You’ll not only learn how to design, test, and deploy generative AI applications, but also how to iterate and improve rapidly. This is about creating AI systems that solve real problems, deliver real value, and scale with confidence. What to ExpectThis is a highly interactive course designed to get you results. Across live workshops, discussions, and hands-on projects, you’ll build your skills while solving real-world challenges. •	Live coding sessions to design and refine LLM applications.•	Hands on exercises to explore how small changes can dramatically improve reliability.•	Discussions on user-centric development and integrating feedback loops.•	Expert-led Q&amp;As to help you tackle your toughest challenges. What You’ll Learn •	How to transform POC demos into production-grade applications.•	Strategies for managing non-determinism and optimizing outputs through prompt engineering.•	How to monitor, log, and debug AI systems to ensure reliability and performance.•	Best practices for handling structured outputs and integrating function calling into your applications.•	Building workflows for iterative development and experimentation.•	Practical skills through hands-on app development, including creating a Generative AI app to query PDFs. Bonus Perks All students receive over $500 credits to build with from top companies. We'll announce additional partner credits soon. Past cohorts have typically received $1,500+ in total credits; one cohort topped $2,500 when multiple partners stacked. (Actual amounts vary.) You will receive - $500 in Modal credits for cloud compute (confirmed).- Additional partner credits: announcing soon.- Past totals: $1,500+ typical; one cohort &gt;$2,500.- Recent partners: Google Cloud &amp; Gemini, Hugging Face Pro, Braintrust, Baseten, Chroma Cloud, Pydantic Logfire, Prodigy, Learn Prompting Plus, Replicate, and Mistral. Guest lectures Learn directly from practitioners shipping AI/LLM systems in production. This cohort's guest speakers will include We'll announce guest speakers soon. Enrollment includes on-demand access to all recorded guest talks from prior cohorts (plus slides/resources). Previous guest speakers in previous cohorts include: - Jason Liu (Instructor, 567 Labs)- Shreya Shankar (UC Berkeley, ex-Google Brain)- Zach Mueller (Hugging Face)- Paige Bailey (Google DeepMind)- Ines Montani (Explosion AI, spaCy)- Alan Nichol (Rasa)- Nathan Danielsen (Carvana)- Charles Frye (Modal)- John Berryman (Arcturus, ex-Github)- Skylar Payne (AI executive for startups. Ex-Google. Ex-LinkedIn)- Vincent Warmerdarm (marimo)- Samuel Colvin (Pydantic, Logfire)- Hamel Husain (Parlance Labs)- Samuel Colvin (Pydantic)- Jeff Huber (Chroma)- Natalia Rodnova (Eon)- Thomas Wiecki (PyMC Labs)- Ravin Kumar (DeepMind) This Course Is For •	Data scientists and machine learning engineers who are sick of unreliable POCs and want to ship reliable LLM applications.•	Software engineers who want to learn how to build generative AI systems and master the LLM software development lifecycle. This Course Is Not For Those •	Looking for an introductory course in AI: some programming knowledge is required.•	Expecting ready-made solutions for AI problems without active participation.•	Unable to commit to hands-on projects, live discussions, and iterative development exercises. Why Take This Course By the end of this course, you’ll: •	Understand the first principles of generative AI development.•	Gain hands-on experience building robust, scalable applications.•	Learn how to troubleshoot, monitor, and optimize AI systems for production.•	Develop workflows to iterate and refine AI applications effectively.•	Leave with the confidence to deploy reliable AI systems at scale.

Master the patterns and processes that underpin successful GenAI app development

Course syllabus

Mondays and Wednesdays

To ensure hands-on practical time, there will be project work to complete throughout the course.

Weekly projects

Course schedule

Data scientists, machine learning engineers who are sick &amp; tired of seeing and building prototypes &amp; want to ship reliable LLM applications

Software engineers who want to learn how to build Generative AI systems and learn the LLM software development lifecycle.

Who is this course for

What you’ll get out of this course

You’ve seen impressive generative AI demos, but how do you take them to production? In this course, you’ll learn the exact workflows, tools, and engineering practices needed to turn concepts into reliable, scalable applications that deliver real value.

Move beyond POCs and build production-ready AI systems

Struggling with inconsistent outputs? Through interactive exercises, you’ll explore proven techniques for optimizing prompts and ensuring your AI delivers actionable results in production settings.

Master prompt engineering for reliable results

When your AI application breaks, you’ll know exactly how to fix it. We’ll walk you through real-world debugging sessions, monitoring techniques, and tools to ensure your systems stay reliable at scale.

 Learn to debug and monitor AI systems like a pro

Learn everything you need to build robust and reliable agentic systems, from incorporating tool-calling, memory and retrieval to augmented LLM workflows and beyond.

Build State-Of-The-Art AI Agents

Stop wasting time in endless iteration cycles. Through guided projects, you’ll discover how to set up workflows that refine your applications faster and with less frustration, and how to use "Evals" and LLM Judges to guide your software development lifecycle.

Accelerate your development process with iterative workflows, best-in-class "Evals", and LLMs-as-Jduge

Tired of theoretical knowledge? You’ll build a fully functional app that uses text and image models to query PDFs, giving you practical experience solving complex real-world problems.

Develop hands-on expertise by building GenAI and LLM apps

Whether you’re looking to lead AI projects, impress stakeholders, or level up your technical expertise, this course gives you the tools to stand out as an AI/ML practitioner.

Advance your career with practical, proven techniques

Leave the course with a project you can add to your portfolio, demonstrate to your boss, or use to showcase your ability to deploy reliable AI systems.

Showcase Tangible Results

Build AI-powered software reliably & from first principles. Learn the GenAI software development lifecycle: agents, evals, iteration & more

Cohort 4

Hugo Bowne-Anderson

Host of Vanishing Gradients podcast; ex-Outerbounds; expert in DS/ML.

Stefan Krawczyk

Agentforce; Ex-Stitch Fix; 13+ Years building / productionizing data, ML, Agents

Cohort 3

Cohort 5

Build production-grade agents that get you hired, funded, and leveled up. Join 200+ builders. Don’t just learn AI. Ship it.🚢

Founder, AI Educator to 145k Linkedin 

Meri Nova

Technical founder, 145k LInkedin community

Senior AI Engineer, Investor & Advisor

Hai Nghiem

Investor | Builder

V3 (Cohort 4)

In just 5 weeks, you'll go from beginner to full-stack AI developer, mastering the latest AI coding tools and building most in-demand AI Engineering skills.Trusted by 200+ engineers, PMs, designers, and founders building real agentic apps from zero experience. How are we different?... 1. Learn the tech stack that is actually used in the industry. (Claude Code, Next.js, AI SDK, Python) 2. Build alongside senior full-stack AI engineer evert day. (Mon-Fri)3. Get real-time feedback from our AI project grader. (Improve consistently with feedback from instructors, TAs and our in-house AI helper)🧠 Learn how to think like a real AI engineer, without spending 6 months learning by yourself.

Join the most hands-on AI Engineering program in the field.

Master the fundamentals of Agent Engineering

Gain confidence in software engineering with LLMs.  

Build a unique and impressive portfolio

See real-world Agentic Use cases

Build life-long connections

Developers and engineers ready to transition into high-paying AI engineering roles and stay competitive in the job market.

Senior leaders, executives, and PMs who need AI fluency to make smart decisions and effectively lead technical AI projects.

Entrepreneurs and founders ready to build successful AI products using the latest AI tools and proven frameworks from YCombinator startups.

Design, deploy, and scale enterprise-grade AI agents with a hands-on, problem-driven approach for engineers and tech leaders.

Founder @ JUTEQ Inc.

Rakesh Gohel

AI leader with 25 years in Agentic AI, Automation, and Intelligent Workflows

Build, collaborate, and demo advanced multi-agent AI systems together.

Cohort 1

Understanding and developing an Enterprise Mindset for Building Agentic AI that scales.

Cohort 2

AI agents are transforming enterprises by automating workflows, reasoning over data, and solving complex problems together. Many teams still struggle to move from demos to production-ready systems.This course helps you build, test, and deploy Enterprise AI Agents and multi-agent RAG pipelines using LangChain, CrewAI, and LangGraph. Work in guided Python notebooks, follow weekly milestones, and get live feedback through code review clinics.🏆 Why It’s Unique:<ul><li>Learn enterprise-grade design for scale and reliability.</li><li>Master context engineering to improve agent reasoning.</li><li>Use ready-to-run notebooks for every lesson.</li><li>Build multi-agent RAG systems ready for deployment.</li><li>Join weekly reviews and guest sessions with industry experts.</li></ul>- What You’ll Build: Scalable, memory-sharing agents and modular RAG pipelines with observability and retries.

Design and deploy advanced AI agents that are ready for production

Enterprise-Ready Design Principles

Context Engineering for AI Agents

Build and Deploy Multi-Agent RAG Systems

Comfort with writing, debugging, and testing Python code. You will clone repositories, install packages, and build modules from scratch.

Python Proficiency

Familiarity with how large language models work and common API patterns (e.g., sending prompts, handling responses).

Basic Concepts of Language Models

Understanding of core software engineering principles such as modularity, version control (Git), unit testing, and debugging.

Software Engineering Principles

Technical Architects and Data Scientists who want to integrate LLMs into agent workflows using memory, tools, and task planning logic.

Product Managers and AI Solution Strategists translating business needs into AI agent workflows, metrics, and scalable solutions.

Tech Leads and Backend Developers managing multiple agents and services looking to scale, monitor, and orchestrate workflows.

Ship production features 2x faster by using coding agents across research, planning, implementation, testing, and review workflows

Creator of Stanford's AI Coding Course

Amazon

Stanford University

Y Combinator

Mihail Eric

Created Stanford's first AI coding class. Former YC founder and Amazon AI lead. 

Today many developers use AI, but few are truly productive with it. I created Stanford's first AI software development class, and after building a YC-backed coding company and leading AI at Amazon, I've seen how top engineers integrate AI into production workflows. My techniques have been used to train 200+ Stanford engineers and industry professionals.My goal is simple: make you dramatically more productive writing software with AI than without it.In this course, you'll learn practical, end-to-end workflows for using AI in real-world development, not toy examples. We'll cover:<ul><li>Building production features with AI agents using the research → plan → implement → test workflow</li><li>Configuring an AI native dev environment for your specific tech stack (IDE, code review, tool integrations, and beyond)</li><li>Setting up review and CI processes that catch AI errors, hallucinations, and slop before production</li><li>Enabling multiple agents to work together on the same codebase without conflict, accelerating software delivery and throughput</li></ul>If you're ready to write better code, ship faster, and stay in control of your stack, let's get started.

Increase developer productivity with AI-first, production-ready workflows

Configure optimal AI-first developer environments to improve your coding output

Learn how to navigate the ecosystem of cutting-edge AI developer tools 

Exercise effective strategies for prompting and building with coding agents

Reduce hallucinations, software errors, and AI slop all while shipping faster

Become a manager of coding agents so you can keep many agents productive 

Develop a deep intuition for how AI coding platforms work under-the-hood

Engineers who want to increase their productivity with AI-generated code while making it actually production grade.

Engineering managers who want to ensure their teams aren't being left behind when it comes to the new way of developing software.

Those confused about how to develop with AI the right way (how to maintain context and manage multiple agents without compromising code)