Umur Ozkul umuro

Why AI Agents Keep Failing: What 2025 Research Really Shows

Despite all the hype and billions of dollars invested, new research from 2025 shows that AI agents fail way more often than companies want to admit. Simple tasks? They do okay. But when things get complicated, these systems crash and burn - with failure rates jumping from 5% to over 80% depending on what you ask them to do.

Bottom Line: AI agents work fine for easy stuff, but they're nowhere near ready for the big promises companies are making. Real businesses are struggling to make them work, and the whole AI investment boom might be heading for a major reality check.

The Massive Failure Problem: Numbers Don't Lie

Scientists at top universities put AI agents through serious tests in 2025, and the results were pretty shocking. They tested six of the "best" AI systems - including the ones companies brag about the most - on thousands of different problems.

Preperation

Installation

Install Elixir: http://elixir-lang.org/install.html
Install Phoenix http://www.phoenixframework.org/docs/installation

sudo apt-get install inotify-tools #ubuntu
sudo apt-get install postgresql #Ubuntu

mix local.hex #install hex
mix archive.install https://github.com/phoenixframework/archives/raw/master/phoenix_new.ez #phoenix archive

Polymorphic Associations reversed

It's pretty easy to do polymorphic associations in Rails: A Picture can belong to either a BlogPost or an Article. But what if you need the relationship the other way around? A Picture, a Text and a Video can belong to an Article, and that article can find all media by calling @article.media

This example shows how to create an ArticleElement join model that handles the polymorphic relationship. To add fields that are common to all polymorphic models, add fields to the join model.

	customDecoder decoder toResult =
	Json.andThen
	(\a ->
	case toResult a of
	Ok b -> Json.succeed b
	Err err -> Json.fail err
	)
	decoder

	# Creates a demo blog in a new hobo project
	# Usage: rake rails:template LOCATION=template_demo_blog_hobo.rb
	# Or: rake rails:template LOCATION=https://raw.github.com/gist/3405055/9aaac7c1c9e9571b32c4359fa9fff77c89de82a9/template_demo_blog_hobo.rb

	say ("Installing gems so that we can scaffold...")
	run 'bundle install'
	generate 'hobo:setup_wizard' if yes? "Wanna run hobo:setup_wizard?"

	generate 'hobo:resource', 'Post title:string text:text'

	# kate: syntax ruby;
	# -- mode: ruby --
	# vi: set ft=ruby :

	require 'rubygems'
	require 'thor'

	class Gemset < Thor
	include Thor::Actions

	# kate: syntax ruby;
	# -- mode: ruby --
	# vi: set ft=ruby :

	# Add those comments to colorize non-rb files automatically. It happens that not all ruby files have the rb extension.

	# kate: syntax ruby;
	# -- mode: ruby --
	# vi: set ft=ruby :

	# IRB improvements: Command history, coloring, more shell like commands cd, pwd, dir, cat
	# Use powerful shell commands
	# dir
	# dir '*/.rb'
	# cd 'app'
	# pwd