May 15, 2026

Agentic Architecture: Why Files Aren't Always Enough

1 hour 24 minutes

What are the limitations of using a file-based agent workflow? Why do massive context windows tend to collapse? This week on the show, Mikiko Bazeley from MongoDB joins us to discuss agentic architecture and context engineering.

Mikiko is an applied AI engineer. She helps developers and organizations build AI and ML applications using MongoDB. We dig into the debate of files versus a database. What are some of the limitations of building an agent with just a folder of files?

We explore the surprising limitations of massive context windows and strategies for fixing them. Mikiko also shares advice and resources to help you get up to speed on building your own agent skills. Our conversation touches on multiple topics in the current development landscape.

This episode is sponsored by SerpApi.

Video Course Spotlight: Building Type-Safe LLM Agents With Pydantic AI

Build type-safe LLM agents in Python with Pydantic AI using structured outputs, function calling, and dependency injection.

Topics:

00:00:00 – Introduction

00:02:31 – Catching up with MongoDB

00:07:02 – Are the files all you need?

00:15:14 – What is a workflow agent?

00:24:43 – Sponsor: SerpApi

00:25:45 – Model vs harness

00:29:57 – Context rot and tool loadouts

00:41:07 – Sharing state and coordination of agents

00:47:27 – Video Course Spotlight

00:49:16 – What do dataflows look like

01:00:38 – The human-in-the-loop & coding agents

01:10:30 – Resources to explore

01:17:49 – What are you excited about in the world of Python?

01:18:38 – What do you want to learn next?

01:22:54 – Thanks and goodbye

Show Links:

The “files are all you need” debate misses what’s actually happening in agent memory architecture - The New Stack

MongoDB: The World’s Leading Modern Data Platform

Karpathy shares ‘LLM Knowledge Base’ architecture that bypasses RAG with an evolving markdown library maintained by AI - VentureBeat

Files Are All You Need: Context, Search, Skills Guide | LlamaIndex

Converged Datastore For Agentic AI - MongoDB

Why Developers Need Vector Search - The New Stack

Why Multi-Agent Systems Need Memory Engineering – O’Reilly

The New Skill in AI is Not Prompting, It’s Context Engineering - Phil Schmid

How Long Contexts Fail - dbreunig.com

How to Fix Your Context - dbreunig.com

AI Agents Need Memory Control Over More Context - arxiv.org

AINews - Is Harness Engineering real? - Latent.Space

The Model vs. the Harness: Which Actually Matters More?

Embeddings and Vector Databases With ChromaDB – Real Python

12-factor-agents: What are the principles we can use to build LLM-powered software that is actually good enough to put in the hands of production customers?

The Twelve-Factor App

MongoDB Courses and Trainings - MongoDB University

mongodb-mcp-server: A Model Context Protocol server to connect to MongoDB databases and MongoDB Atlas Clusters.

What is the MongoDB MCP Server? - MongoDB Docs

mongo-python-driver: PyMongo - the Official MongoDB Python driver

agent-skills: Use the official MongoDB Skills with your favorite coding agent to build faster.

Reachy Mini - Open-Source Desktop Humanoid Robot

👩🏻‍💻 Mikiko B. - LinkedIn

Building AI Products From Scratch - Mikiko Bazeley - Substack

Level up your Python skills with our expert-led courses:

Getting Started With Claude Code

Building Type-Safe LLM Agents With Pydantic AI

Using Pydantic to Simplify Python Data Validation

Support the podcast & join our community of Pythonistas

...more

View all episodes

By Real Python

4.7

139139 ratings