Agent Pulse
Posts
Karpathy’s Reality Check: Why Agents Aren’t Ready (Yet)

Karpathy’s Reality Check: Why Agents Aren’t Ready (Yet)

Plus: a new AI on Fire episode drops — with stories straight from the frontlines of agent startups.

O P
October 20, 2025

In partnership with

AIonFire | AArena | Submit Agent | Advertise

Welcome back! OP here again, helping you with another addition of Agent Pulse - your go-to spot for agentic news, insights and more.

In today’s:

👉 TOP Agentic News
✨ Featured Agents
🎙️ What to Watch this Week
⚔️ Agent Arena Battleboard
🏆 Agents Leaderboard
🗺️ Agents Landscape Map

Go from AI overwhelmed to AI savvy professional

AI keeps coming up at work, but you still don't get it?

That's exactly why 1M+ professionals working at Google, Meta, and OpenAI read Superhuman AI daily.

Here's what you get:

Daily AI news that matters for your career - Filtered from 1000s of sources so you know what affects your industry.
Step-by-step tutorials you can use immediately - Real prompts and workflows that solve actual business problems.
New AI tools tested and reviewed - We try everything to deliver tools that drive real results.
All in just 3 minutes a day

Join 1M+ pros

The Latest Agentic AI Development

🧠 “It Will Take a Decade for Agents to Actually Work” — Andrej Karpathy’s Reality Check

On the Y Combinator AI-Startup School podcast and again on X, Karpathy didn’t mince words: “They just don’t work… It will take about a decade to work through all of those issues.”

The @karpathy interview
0:00:00 – AGI is still a decade away
0:30:33 – LLM cognitive deficits
0:40:53 – RL is terrible
0:50:26 – How do humans learn?
1:07:13 – AGI will blend into 2% GDP growth
1:18:24 – ASI
1:33:38 – Evolution of intelligence & culture
1:43:43 - Why self
— Dwarkesh Patel (@dwarkesh_sp)
5:16 PM • Oct 17, 2025

He sees two core problems: current agents lack continual learning, multimodal reasoning, and long-horizon task management—and we’re still years away from reliably “just tell it once and it figures it out.”

What Others Are Saying

ScaleAI’s growth lead Quintin Au puts numbers to it: “If an agent does five independent actions and each has ~20% chance of error, your success chance drops to ~32%.”
Analyst commentary agrees: “2025 is not the year of agents—it’s the decade of agents (2025-2035).”
Key academic work backs this: a 2025 paper found that for long-duration tasks, agent success decays exponentially with task length.

Why This Matters for You

Hype ≠ product: While many tools brand themselves “agentic,” the underlying tech still struggles with reliability, memory, and multi-step autonomy.
Short-horizon wins first: Expect valuable agent use-cases in structured, limited domains (e.g., document triage, data prep) rather than “CEO AI” today.
Build for the long game: If you’re investing time or money into agents, ask: what’s the 3-5 year plan for learning, memory, multimodality, and autonomy?
Be wary of models that promise plug-and-play: If a platform claims “complete agent discovery + orchestration” today, ask for evidence of long-task success rates (not just demos).

Takeaway

Karpathy’s verdict? We’re not witnessing a sudden “agent takeover.” We’re at the foundation of a new era of software—Software 3.0—where agents will matter, but they’ll take time. The next decade is about building infrastructure, safe systems, and real-world feedback loops, not just narrative leaps.

Other News

SoundHound demos agentic healthcare AI: SoundHound AI is showcasing its next-gen Amelia AI Agent platform at the HLTH 2025 conference, with live demos highlighting how agentic AI can improve patient experiences and operational efficiency in healthcare
Anthropic’s Claude Code goes multi-platform: Anthropic expanded Claude Code – its AI coding agent – to the web (with an iOS preview), enabling developers to launch multiple coding tasks in parallel on managed cloud infrastructure
LangChain becomes a unicorn for AI agents: LangChain, a startup behind a popular open-source AI agent framework, raised $125 million in funding at a $1.25 billion valuation to accelerate development of its agent-building tools
Opera unveils a research-focused AI agent: Opera introduced the Opera Deep Research Agent (ODRA) – the fourth AI agent for its Opera Neon browser – a model-agnostic, server-side agent for deep research tasks.
365Talents launches an HR agent co-pilot: European HR tech firm 365Talents launched “Job Architect,” a free AI-powered HR agent that helps organizations instantly create, refine, and visualize job frameworks and skill map.
WhatsApp bans general AI chatbots: Meta’s WhatsApp updated its Business API policy to ban general-purpose AI chatbots on the platform, reserving the service for customer support bots and effectively blocking external AI assistants on WhatsApp

Choose the Right AI Tools

With thousands of AI tools available, how do you know which ones are worth your money? Subscribe to Mindstream and get our expert guide comparing 40+ popular AI tools. Discover which free options rival paid versions and when upgrading is essential. Stop overspending on tools you don't need and find the perfect AI stack for your workflow.

Subscribe to Get Your Free Comparison