🚨🚨Notes:

- For questions or bulk requests, reach out to: problemfirst.ai@gmail.com

- October will be our final cohort. We don’t plan to run this course in 2026, it’s simply gotten too hectic. As much as we love teaching AI, we love building enterrprise AI systems a bit more. So if you’ve been considering it, this is your last chance to join.

- This course is an independent offering and is not affiliated with, endorsed by, or related to the instructors' current or past employers.

⛳ The only prerequisite: you should have coded at least once in your life. The course includes low-code assignments, and even folks who hadn’t touched code in over 15 years have found it approachable and rewarding. That said, a basic understanding of coding really helps you get the most out of it — and of course, there’s AI to assist you along the way. The course is built for everyone, whether you’re a Product Manager, Architect, Director, C-suite leader, or someone seriously exploring agentic AI.

Agentic AI or AI systems capable of operating with some degree of autonomy, is transforming how we interact with technology. In the coming years, most software systems will integrate AI agents to enhance their capabilities. This shift will drive a growing demand for professionals who can move beyond surface-level understanding and apply AI effectively to solve real business challenges while navigating practical constraints.

This course focuses on practical AI agent development, covering key agentic design and usage paradigms. Instead of just explaining what these techniques are, we focus on when and how to use them, so you're equipped to make informed, business-driven AI decisions.

What You'll Learn

All core content is pre-recorded so students can focus on two-way interaction. Lectures are watched asynchronously, and we host four office hours each week for questions and brainstorming

Week 1 (Let’s get you to understand what problem-first means)

Decode why agentic AI breaks traditional software assumptions

Frame hallucinations, latency, and prompt brittleness through the determinism spectrum

Open vs. closed models: tradeoffs across compliance, latency, and cost

Problem-first, evaluation-driven design using early datasets and proxy metrics

Deconstruct a production-grade use case and redesign it across progressive system versions

Week 2 (Context engineering is the core part of agents, do it smarter with right evals)

Break down the evolution from zero-shot prompts to self-optimizing models

Master context engineering: Decomposition, meta-prompts, algorithmic optimization

Analyze when to use prompting-only systems based on task, cost, and latency

Compare model-level strategies: reasoning vs. regular, and when each makes sense

Add guardrails and evaluation layers using LLM judges, semantic scoring, and offline tests

Week 3 (RAG is not dead, it’s in fact the basis of self-improving agents)

Address statelessness via dynamic retrieval and memory-backed context injection

Build robust RAG pipelines with advanced chunking, embedding selection, and retrieval methods

Explore GraphRAG, Agentic RAG and multimodal RAG and other advanced methods and learn tradeoffs

Architect episodic, semantic, procedural, and working memory layers for self-reflective agent behavior

Week 4 (MCP from an enterprise lens and multi-agents + Fine-Tuning)

Understand planning autonomy in agents and how dynamic tool use and multi-turn reasoning go beyond static workflows

Compare agent levels and their control dimensions: action, planning, evolution, and physical autonomy

Explore MCP (Model Context Protocol) and A2A as emerging agent-tool communication standards

Investigate critical security challenges in MCP and A2A. Understand how guardrails, tool signing, audit trails improve reliability

Analyze coordination patterns in multi-agent systems, including shared memory governance, state sync, AI collusion risks, evaluation, logging, and observability

Explore fine-tuning levers (SFT, RLHF, PEFT etc.), compare with RAG, and determine when to shift from context injection to model adaptation

Week 5–6 (Put it all together in a capstone)

Work in groups of 6

Take a business problem and design/implement a solution

Demo to 4000+ public attendees including leaders, VCs, and hiring managers

Homeworks: You'll supplement your learning by building an agentic search system (Perplexity like) in 3 iterations with the final iteration using agentic RAG, MCP and multi-agents. You can choose between low-code/code routes to complete assignments.

----

❌ Who This Course Is Not For

For Those Who Have Already Deployed Gen AI in Enterprise: This course is designed as an applied foundations course for enterprise AI with only basic Python as a prerequisite and no ML background required. If you’re already familiar with deploying AI systems, you won’t gain much from the core content. However, if you're looking to network and refine best practices, you're welcome to join.

Those Seeking Heavy Theoretical Knowledge: This course emphasizes applied learning and practical problem-solving, not deep dives into theoretical topics like transformer architecture, pre/post-training optimization, inference techniques, or alignment.

Those Who Have Never Coded Before: While we provide low-code options, this course assumes you have some coding experience. It’s not suitable for those who have never written or worked with even basic code.

Individuals Expecting Deep AI Research Focus: While we’ll cover cutting-edge techniques, this course is centered on applying AI to business problems, not research-heavy exploration.

Scaling and Ops Enthusiasts: This course does not focus heavily on scaling or operational aspects (i.e., LLMOps). Deployment will be covered at a high level, but not in-depth.