Originally published on Towards AI.

A practical developer-first comparison of LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, DSPy, and more after real experimentation.

Six months ago, I decided to evaluate AI agent frameworks seriously. Not because I needed to — I had a working system — but because the space was moving so fast that I felt like I was missing something. The tools available now are genuinely different from what existed a year prior, and the conversations I was seeing online felt reductive. People would declare one framework “the winner,” then pivot three weeks later. I wanted to understand what was actually happening beneath the hype.

The author reports results from experimenting with ten AI agent frameworks (LangGraph, CrewAI, AutoGen, Semantic Kernel, OpenAI’s Agents SDK, PydanticAI, Haystack Agents, LlamaIndex Workflows, Atomic Agents, and DSPy) and argues that the ecosystem is fragmented rather than converging: each framework makes different bets about control vs. abstraction, orchestration style, tool-calling behavior, and state/memory management. Key pain points include messy tool calling (stop/retry/error semantics), underestimated work around state and memory (often requiring wrapper logic), and orchestration complexity that scales poorly beyond a few agents. They highlight which frameworks do well for structured outputs (PydanticAI, DSPy) and where debugging/observability and documentation quality vary dramatically, with many frameworks not built with cross-framework lifecycle visibility in mind. The piece also stresses practical selection criteria—framework fit to the specific problem, integration and dependency footprint, local model and provider support, and production readiness/stability—not just feature checklists. The conclusion recommends choosing based on immediate constraints (and starting with simpler function-calling loops when orchestration isn’t truly needed) and expecting the market to keep evolving.

Read the full blog for free on Medium.

Published via Towards AI

Towards AI Academy

We Build Enterprise-Grade AI. We'll Teach You to Master It Too.

15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.

Start free — no commitment:

→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day

→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages

Our courses:

→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.

→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.

→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.

Note: Article content contains the views of the contributing authors and not Towards AI.

---