Skip to content
AI

AI Agents Aren't Ready: The Benchmark That Deflates Automation Hype

The DELEGATE-52 benchmark tested 52 professional domains and found AI agents corrupt outputs in long task chains. Only Python cleared the reliability bar.

Francesco CampisiBy Francesco Campisi Updated 24 Jul 2026 4 min read How we work

Related Articles

Claude Embeds an Invisible Watermark in AI Text: What Actually Changes Hamza Ahmed Hamza Ahmed 13 Aug 2026 Claude Embeds an Invisible Watermark in AI Text: What Actually Changes Anthropic's Claude now embeds an invisible watermark in every text it generates. The signature lives in word choice, survives copy-paste, but paraphrasing… MARA Pledges Half Its Bitcoin for $750M Energy and AI Bet Francesco Campisi Francesco Campisi 11 Aug 2026 MARA Pledges Half Its Bitcoin for $750M Energy and AI Bet MARA pledged 18,750 BTC (53% of its reserves) as collateral for $750 million in loans to acquire a 2-gigawatt Ohio energy site for AI. The stock dropped 11%…
EU AI Act Article 50: What Crypto Exchanges and Influencers Must Do Now EU AI Act Article 50: What Crypto Exchanges and Influencers Must Do Now Giulia Ferrante 9 Aug 2026 ChatGPT Can Now Buy Things for You: MoonPay's PayBox Brings Crypto Payments Into AI Chat ChatGPT Can Now Buy Things for You: MoonPay's PayBox Brings Crypto Payments Into AI Chat Hamza Ahmed 2 Aug 2026
Consent Preferences