OpenAI's Agent: Ten Years to Stop Forgetting Everything
OpenAI cofounder Andrej Karpathy recently confessed on the Dwarkesh Podcast that AI agents, hailed as 2025's holy grail of autonomous coding and task-solving, are basically glorified goldfish with a 10-second memoryâscratch that, a decade to fix their act. He slammed agents for lacking 'continual learning,' multimodal smarts, and computer savvy, calling their cognitive state 'not working.' Karpathy, a famously fast talker stuck explaining himself again on X (Twitter if youâre behind on acronyms), decried AI's delusional future where humans are relics and AI spews endless 'slop.' Meanwhile, ScaleAI's Quintin Au pinpoints a 20% error rate per AI action, meaning five-step tasks are as reliable as your drunk uncle's GPS directions. Ten years to functional autonomy, or a decade of AI just shrugging and forgetting? Either way, humans apparently still get to feel superiorâfor now.
Share the Story
AI agents in 2025 are like toddlers who forget what you told them five seconds ago â except the toddler is supposed to write code and take your job. Patience, humanity: these digital goldfish need a decade to stop drowning in their own forgetfulness.
Karpathy says AI agents wonât truly work for about 10 years. Meanwhile, humans are still struggling to remember their passwords and why they walked into the kitchen.
2025: The Year of The Agent, or The Year AI Forgets What It Just Did? Depends if you like your virtual assistants glorified amnesiacs.
Source: Businessinsider | Published: 10/19/2025 | Author: Lakshmi Varanasi
More Articles in Technology
Bezosâs Fire Phone: 400 Cameras, Zero Friends, Dead on Arrival
Pope Leo XIV Tells AI to âGet Moral or Go to Confessionâ
Halo Infiniteâs Last Update: Maintenance Mode or Death Spiral?
Zuck's $600B Hot-Mic Bid: Betting Big on AI and Water Bills
Polestar 4: The $58K EV That Said âWho Needs Rear Glass?â
