Skip to content

AI Agent Blog

Tutorials, framework comparisons, and production guides for building AI agents with Python. Covering Pydantic AI, LangChain, LangGraph, CrewAI, and more.

· 5 min read · Kacper Włodarczyk · Article

Python Document Parsing and OCR for AI Agents: Inside LiteparseToolset

How LiteparseToolset gives any pydantic-deepagents agent unified PDF text extraction and per-page OCR through one tool API. Tesseract local or HTTP OCR server. Page-level citations across both formats.

pydantic-deepagents ocr ai-agents python document-parsing
· 5 min read · Kacper Włodarczyk · Article

24 Claude Code Skills to Fix Your AI Stack: Introducing production-stack-skills and content-skills

Two Claude Code skill packs shipped together. 10 skills audit AI-generated code with a 0-100 score. 14 skills kill AI slop with brand-first generation. Same install, dual-CLI, MIT.

claude-code ai-agents python open-source developer-tools
· 5 min read · Kacper Włodarczyk · Article

/improve: How pydantic-deep Agents Learn from Their Own Session History

How /improve extracts UserFactInsight and AgentLearningInsight from raw tool traces to update MEMORY.md, and why raw traces beat summaries for synthesis.

pydantic-ai python ai-agents open-source developer-tools
· 5 min read · Kacper Włodarczyk · Article

Full Observability in AI Agents: What We Added to the pydantic-deep TUI

pydantic-deep v0.3.5 adds per-turn token tracking, live thinking display, cumulative cost header, and side panel with subagent status to the terminal UI.

pydantic-ai python ai-agents open-source developer-tools
· 5 min read · Kacper Włodarczyk · Tutorial

Browser Automation + /improve: AI Agents That Browse the Web and Fix Themselves

BrowserCapability adds 9 Playwright tools to pydantic-deep agents. /improve extracts session insights to MEMORY.md using raw tool traces, and why raw traces outperform summaries for agent learning.

pydantic-ai python ai-agents playwright browser-automation
· 5 min read · Kacper Włodarczyk · Open Source

We Ran /improve on 50 Sessions, Here's What It Found

We analyzed 50 real agent sessions with pydantic-deep's /improve pipeline. 180+ insights extracted, 7 types, automatic MEMORY.md updates. Here are the results.

self-improving-agents ai-agents pydantic-ai python agent-evaluation
· 5 min read · Kacper Włodarczyk · Tutorial

AI Agents in a Sandbox: Docker Execution + CI/CD Integration with pydantic-deep

How pydantic-deep's --sandbox docker flag runs agent file operations inside a container, plus the headless runner for CI/CD pipelines. Named workspaces, Harbor adapter, and workspace management.

pydantic-ai python ai-agents docker devops
· 5 min read · Kacper Włodarczyk · Open Source

Inside /improve: How We Extract 7 Insight Types From Agent Sessions

Deep dive into the 4-stage pipeline behind pydantic-deep's /improve command — how it chunks sessions, extracts 7 typed insight categories, and synthesizes proposed changes with confidence scoring.

self-improving-agents ai-agents pydantic-ai python session-analysis
· 5 min read · Kacper Włodarczyk · Tutorial

curl | bash for AI Agents: One-Command Install for pydantic-deep

The standard Python AI tool install is 7 steps and 15 minutes. pydantic-deep now installs with a single curl command — uv detection, self-update, startup notifications included.

pydantic-ai python open-source developer-experience cli
· 5 min read · Kacper Włodarczyk · Article

Context Window Blindness: Why Your AI Agent Doesn't Know It's Running Out of Space

AI agents have no intrinsic awareness of context usage. LimitWarnerCapability injects URGENT/CRITICAL warnings at 70% and 85% — before the 90% auto-compression cliff.

pydantic-ai python ai-agents open-source llm
· 5 min read · Kacper Włodarczyk · Open Source

What If Your AI Agent Could Learn From Its Mistakes?

Build self-improving AI agents that extract 7 types of insights from past sessions and automatically update their own context files. Open-source Python framework.

self-improving-agents ai-agents pydantic-ai python machine-learning
· 5 min read · Kacper Włodarczyk · Open Source

StuckLoopDetection: How We Stopped an Agent Burning $12 on 47 Identical Calls

Most agent loops are mechanical, not model failures. pydantic-deep v0.3.8 catches 3 loop patterns — repeated calls, A-B-A-B alternating, no-op — before they waste your API budget.

pydantic-ai python ai-agents open-source llm
· 5 min read · Kacper Włodarczyk · Open Source

pydantic-deep v0.3.4–v0.3.8: What We Built in 2 Weeks of Silence

TUI redesign, browser automation, Docker sandbox, curl installer, StuckLoopDetection, LimitWarnerCapability — full changelog overview of 5 releases in 2 weeks.

pydantic-ai python ai-agents open-source llm
· 5 min read · Kacper Włodarczyk · Tutorial

SubAgentSpec: Configure Multi-Agent Systems in YAML, Run Them in Python

Declare multi-agent systems in YAML, get type-safe Python agents. Token tracking, custom agent factories, structured output serialization. pip install pydantic-ai-subagents.

pydantic-ai python multi-agent ai-agents configuration
· 5 min read · Kacper Włodarczyk · Tutorial

.claude/ Directory: AI-Agent Config in Every Generated Project

Every project from our open-source template ships with a .claude/ directory — 8 rule files, 3 slash commands, and a permissions config. The AI coding agent knows your conventions from the first prompt.

claude-code python developer-tools ai-agents fastapi
· 5 min read · Kacper Włodarczyk · Tutorial

10 AI Agent Guardrails in Python — Zero External Dependencies

Add PII detection, prompt injection defense, cost tracking, and 7 more guardrails to your Pydantic AI agents. Zero external dependencies. pip install pydantic-ai-shields.

ai-agents python security pydantic-ai open-source
· 5 min read · Kacper Włodarczyk · Tutorial

5 AI Frameworks, 1 Template: Architecture Deep Dive

Switch between Pydantic AI, LangChain, LangGraph, CrewAI, and DeepAgents with one config change. Architecture deep dive with code from all 5 implementations.

ai-frameworks python software-architecture fastapi ai-agents
· 5 min read · Kacper Włodarczyk · Open Source

Pydantic Deep Agents 0.3.3: ACP, Thinking, Lifecycle Hooks, and Opinionated Defaults

Run your AI agent inside Zed with ACP. Subagents are now full deep agents. Thinking and Anthropic caching on by default. The biggest pydantic-deep release yet.

pydantic-ai pydantic-deep acp ai-agents python open-source
· 5 min read · Kacper Włodarczyk · Tutorial

File Upload to LLM Vision: Images, PDFs, and DOCX in Your AI Chat

Upload images, PDFs, and DOCX to your AI chat. Backend parses documents, images go through LLM Vision. Full implementation with FastAPI and Pydantic AI.

file-upload llm-vision python fastapi ai-agents
· 5 min read · Kacper Włodarczyk · Open Source

Pydantic AI Capabilities, Hooks & Agent Specs -- What Changed and How Our Libraries Migrated

Pydantic AI shipped capabilities, hooks, and agent specs. We migrated 5 open-source libraries. Here's what changed, before/after code, and lessons learned.

pydantic-ai capabilities hooks agent-specs python ai-agents
· 5 min read · Kacper Włodarczyk · Tutorial

Full RAG Pipeline: 4 Vector Stores, Hybrid Search, and Reranking in One Template

One template, every RAG decision already made — from vector store to reranking strategy. 4 vector stores, 4 embedding providers, hybrid search, and reranking.

rag vector-database python fastapi ai-agents
· 5 min read · Kacper Włodarczyk · News

Why Reading PDFs is Hard

LlamaIndex explains why PDF parsing remains fundamentally difficult: PDFs store drawing instructions for glyph shapes at coordinates rather than semantic text.

llm pdf-parsing rag llamaindex document-processing
· 5 min read · Kacper Włodarczyk · News

Practical Guide to Evaluating and Testing Agent Skills

Learn how to systematically test and improve agent skills using deterministic checks and a real-world Gemini API example.

ai-agents evals testing agent-skills gemini
· 5 min read · Kacper Włodarczyk · News

Product Management on the AI Exponential

Cat Wu, Head of Product for Claude Code at Anthropic, describes how exponentially improving AI models are fundamentally reshaping product management.

ai-agents product-management anthropic claude-code
· 5 min read · Kacper Włodarczyk · News

Why SWE-bench Verified No Longer Measures Frontier Coding

OpenAI argues that SWE-bench Verified is no longer reliable due to flawed test cases and training data contamination across all frontier models.

benchmarks coding-agents openai swe-bench evals
· 5 min read · Kacper Włodarczyk · Open Source

From create-react-app to create-ai-app: The New Default for AI Applications

In 2016, create-react-app standardized how we build frontends. In 2026, AI applications need the same moment — and it's here.

ai-agents fastapi pydantic-ai python template
· 5 min read · Kacper Włodarczyk · News

How We Built LangChain's GTM Agent

LangChain details how they built an internal GTM agent that automates outbound sales, increasing lead conversion by 250% and saving 40 hours per rep per month.

ai-agents langchain sales-automation deep-agents
· 5 min read · Kacper Włodarczyk · News

Introducing GPT-5.4

OpenAI releases GPT-5.4, their most capable and efficient frontier model for professional work, with state-of-the-art coding, computer use, tool search, and 1M-token context.

openai llm gpt-5 computer-use benchmarks
· 5 min read · Kacper Włodarczyk · News

Developer Guide: Nano Banana 2 with the Gemini Interactions API

Learn how to use the Gemini Interactions API to build a personalized Japan travel brochure with Nano Banana 2 in four progressive steps.

google gemini image-generation api developer-guide
· 5 min read · Kacper Włodarczyk · News

Gemini Embedding 2: First Natively Multimodal Embedding Model

Google releases Gemini Embedding 2, the first natively multimodal embedding model mapping text, images, videos, audio, and documents into a unified space.

google gemini embeddings multimodal rag
· 5 min read · Kacper Włodarczyk · News

GPT-5.2 Derives a New Result in Theoretical Physics

A new preprint shows GPT-5.2 proposing a novel formula for gluon amplitudes, later formally proved and verified by OpenAI and academic collaborators.

openai llm science physics ai-research
· 5 min read · Kacper Włodarczyk · News

Harness Engineering: Leveraging Codex in an Agent-First World

OpenAI describes building and shipping a product with zero manually-written code -- every line generated by Codex agents, resulting in a million lines and 1,500 merged PRs.

openai codex coding-agents agent-harness engineering
· 5 min read · Kacper Włodarczyk · Tutorial

AGENTS.md: Making Your Codebase AI-Agent Friendly (Copilot, Cursor, Codex, Claude Code)

Every AI coding tool reads your repo differently. Here's how AGENTS.md — the emerging tool-agnostic standard — gives them the context they need.

ai-agents developer-tools python fastapi open-source
· 5 min read · Kacper Włodarczyk · News

From Model to Agent: Equipping the Responses API with a Computer Environment

OpenAI describes how they built an agent runtime using the Responses API, shell tool, and hosted containers to run secure, scalable agents.

ai-agents openai agent-runtime coding-agents
· 5 min read · Kacper Włodarczyk · News

Gemini 3 Deep Think: Advancing Science, Research and Engineering

Google announces a major upgrade to Gemini 3 Deep Think with record benchmarks including 84.6% on ARC-AGI-2 and gold-medal Olympiad performance.

google gemini benchmarks reasoning science
· 5 min read · Kacper Włodarczyk · News

Gemini Deep Think: Redefining the Future of Scientific Research

Google DeepMind presents how Gemini Deep Think mode is solving professional research-level problems across mathematics, physics, and computer science.

google gemini ai-research mathematics science
· 5 min read · Kacper Włodarczyk · News

ByteDance Deer-Flow: Open-Source SuperAgent Harness

DeerFlow is ByteDance's open-source super agent harness built on LangGraph that evolved from a deep research framework into a general-purpose agent runtime.

ai-agents open-source agent-harness langchain bytedance
· 5 min read · Kacper Włodarczyk · News

Can We Close the Loop in 2026?

Two capabilities separate effective AI agents from poor ones: self-awareness and closing the loop -- verifying work against external signals before responding.

ai-agents self-verification agent-design coding-agents
· 5 min read · Kacper Włodarczyk · News

How Coding Agents Are Reshaping Engineering, Product and Design

Harrison Chase argues that coding agents are fundamentally reshaping EPD roles, shifting the bottleneck from building to reviewing.

ai-agents coding-agents product-management engineering
· 5 min read · Kacper Włodarczyk · News

The Anatomy of an Agent Harness

A systematic derivation of core agent harness components -- filesystems, sandboxes, memory, context management, planning loops -- by working backwards from model needs.

ai-agents agent-harness context-management coding-agents
· 5 min read · Kacper Włodarczyk · News

Autonomous Context Compression

LangChain introduces autonomous context compression in the Deep Agents SDK -- a tool that lets models trigger their own context window compaction at opportune moments.

ai-agents langchain context-management deep-agents
· 5 min read · Kacper Włodarczyk · News

How Autoresearch Will Change Small Language Models Adoption

Autoresearch lets an AI agent run hundreds of model training experiments overnight. Learn how it works, early results from Karpathy and Shopify, and how to apply it.

llm machine-learning autoresearch slm training
· 5 min read · Kacper Włodarczyk · Tutorial

Zero to Production AI Agent in 30 Minutes with 5 AI Frameworks

Step-by-step walkthrough: web configurator, pick a preset, choose your AI framework, configure 75+ options, docker-compose up — working production AI app.

ai-agents fastapi full-stack python pydantic-ai
· 5 min read · Kacper Włodarczyk · Open Source

Same Chat App, 4 AI Frameworks: Pydantic AI vs LangChain vs LangGraph vs CrewAI

I built the same chat app 4 times with 4 different AI frameworks. Same backend, same frontend, same DB. Here's what the code looks like.

pydantic-ai langchain langgraph crewai comparison full-stack-template
· 5 min read · Kacper Wlodarczyk · Article

Build an AI PR Reviewer with 3 Parallel Subagents in Python

Security, style, and performance checks in 30 seconds — using pydantic-deepagents to run 3 specialized subagents in parallel against your git diff.

pydantic-ai ai-agents python code-review open-source
· 5 min read · Wojciech Achtelik · Article

Tau-Bench: Testing AI Agents Where It Actually Matters - Customer Service

Sierra's Tau-Bench tests AI agents in realistic customer service with simulated users, domain policies, and real databases.

ai-agents benchmarks tau-bench llm-evaluation customer-service
· 5 min read · Kacper Włodarczyk · Article

PyAI Conf SF Recap: Where Python Meets Production AI Agents

Same-day recap from PyAI Conf — talks by Samuel Colvin, Guido van Rossum, hallway conversations, and what it means for production AI agents.

ai-agents conference pydantic-ai python open-source multi-agent
· 5 min read · Wojciech Achtelik · Article

BrowseComp: The Benchmark That Tests What AI Agents Can Actually Find

OpenAI's BrowseComp flips benchmarks — questions easy to verify but brutally hard to solve. Here's why it matters for AI agents.

ai-agents benchmarks browsecomp deep-research llm-evaluation
· 5 min read · Kacper Włodarczyk · Article

GPT-5.4 Is Here — What It Means for AI Agent Developers

OpenAI released GPT-5.4 with native computer use, 1M context, and tool search. Here's what matters for teams building AI agents.

gpt-5.4 openai ai-agents llm news
· 5 min read · Kacper Włodarczyk · Tutorial

We Built a Web Configurator for AI Agent Apps — 75+ Options, Download as ZIP

Web configurator for AI agent apps: 246 templates, 5 AI frameworks, client-side Nunjucks rendering. 75+ options, download as ZIP — zero server needed.

pydantic-ai fastapi ai-agents python template
· 5 min read · Kacper Włodarczyk · Open Source

Sub-90ms Cloud Code Execution: How Daytona Replaced Docker in Our AI Agent Stack

Docker cold starts cost 2-5 seconds per sandbox. Daytona does it in under 90ms — here's how we integrated it into pydantic-ai-backend.

pydantic-ai ai-agents python docker daytona
· 5 min read · Kacper Włodarczyk · Open Source

We're Building an Open-Source Claude Code Alternative in Python

Meet pydantic-deepagents — 5 modular packages, 30+ features, and a terminal UX that makes AI coding agents feel like a native dev tool.

pydantic-ai ai-agents python cli open-source
· 5 min read · Kacper Włodarczyk · Article

Pydantic AI vs LangChain for Production AI Agents (2026)

A practical comparison of Pydantic AI and LangChain for building production AI agents. Type safety, streaming, dependency injection, and real-world trade-offs.

pydantic-ai langchain ai-agents python comparison
· 5 min read · Kacper Włodarczyk · Tutorial

How to Build a Full-Stack AI App with FastAPI and Next.js

Step-by-step guide to building a production AI application with FastAPI backend and Next.js frontend. From zero to deployed in 30 minutes.

fastapi nextjs tutorial python ai-agents
· 5 min read · Kacper Włodarczyk · Open Source

Hashline Edit Format: How 2-Character Hashes Fixed AI File Editing

Hashline replaces error-prone str_replace with 2-character content hashes. Benchmark: +5 to +64pp accuracy across 16 models.

ai-agents file-editing hashline pydantic-ai python llm-tools
· 5 min read · Kacper Włodarczyk · Open Source

Predictive AI: Give Your Agent a Docker Lab to Run Models

How to build an AI agent that runs sklearn predictions inside an isolated Docker sandbox. Environment-as-a-tool pattern with Pydantic AI.

ai-agents docker sandbox pydantic-ai python machine-learning
· 5 min read · Kacper Włodarczyk · Open Source

AI Agent Selective Memory: Stop Storing Everything, Start Remembering What Matters

Every memory system stores garbage. Here's how we solved it with file-based persistent memory, context window limits, and a save-immediately philosophy.

ai-agents memory memv pydantic-ai python
· 5 min read · Kacper Włodarczyk · Open Source

Real-Time AI Agent Streaming with WebSockets and FastAPI

How to implement real-time streaming for AI agents using WebSockets in FastAPI. Token-by-token output, error handling, and production patterns.

websockets fastapi streaming python ai-agents
· 5 min read · Kacper Włodarczyk · Open Source

Observability for AI Agents Is Broken. Here's What We Built Instead.

Standard dashboards fail for LLM traces. We built a natural-language-to-SQL assistant for debugging AI agents with Pydantic Logfire.

ai-agents observability logfire pydantic-ai python monitoring
· 5 min read · Kacper Włodarczyk · Tutorial

Ship a Production AI App in 5 Minutes: FastAPI + Next.js + 20 Integrations

One CLI command, one production-ready stack: JWT auth, WebSocket streaming, Pydantic AI, PostgreSQL, Redis, Docker, and 20+ integrations.

fastapi nextjs pydantic-ai python ai-template docker
· 5 min read · Kacper Włodarczyk · Open Source

Task Planning for AI Agents: Dependencies, Events, and Hierarchical Todos

Why agents need structured planning, how subtasks and cycle detection work, and a PostgreSQL backend for multi-tenant production deployments.

ai-agents task-planning pydantic-ai python todo-management
· 5 min read · Kacper Włodarczyk · Open Source

AI Agent Memory: Why Your Agent Forgets After 50 Messages (And the Fix)

SummarizationProcessor vs SlidingWindowProcessor - two strategies for keeping your agent's memory alive when the context window fills up.

ai-agents memory-management pydantic-ai python context-window
· 5 min read · Kacper Włodarczyk · Open Source

Pydantic AI vs LangChain vs LangGraph vs CrewAI: Which Framework to Choose?

A comprehensive comparison of Python AI agent frameworks for 2026. Pydantic AI, LangChain, LangGraph, CrewAI, and DeepAgents — when to use each one.

pydantic-ai langchain langgraph crewai deepagents comparison

Ready to ship your AI app?

Pick your frameworks, generate a production-ready project, and deploy. 75+ options, one command, zero config debt.

Configure & Download