Ex-DeepMind/OpenAI researcher: agents ace tasks but can't do jobs—yet
philhchen · x · 2026-10-30
An author who trained frontier models at DeepMind and OpenAI and now runs a company almost entirely on agents explains why AI won't take your job yet: agents are great at tasks, bad at jobs. Execution got cheap, but everything around it still needs human orchestration—his company keeps hiring humans working longer hours.
Key gaps he identifies:
- Agents work for one person, not the company: today's frontier agent products act as an 'AI chief of staff' for individuals; personal productivity rises but the company doesn't necessarily move faster.
- Can't align multiple stakeholders: real work balances conflicting goals across teams, customers, and management.
- Can't learn implicit feedback: humans learn from a manager rewriting drafts, copying others, slower replies, and being consulted less—signals agents don't receive.
- Can't follow live conversations: the conversation moves on while the agent is 'looking it up.'
- Can't learn unrecorded physical work.
Each gap, he argues, is a new frontier for research and the reason most knowledge work isn't automated.
More from AGI Musings
- Crisp, clear writing becomes a huge moat when AI does all the writing — Saboo_Shubham_ · 2026-10-30
- Dan Wang: London is already a post-AGI city — 'not functional, but fun' — herbiebradley · 2026-10-30
- Philosopher argues AI could usher in a golden age of mathematics—if we redefine mathematical progress — Dr_Atoosa · 2026-10-30
- A 4-person team ships 77,000 changes/month with AI agents — how devs stay relevant — yoobinray · 2026-10-30
- Ben Horowitz: continual learning is the next big tech revolution and value accrues to its builders — brianryhuang · 2026-10-30
- Anthropic API Key Leaked; Agents Detected and Rotated It Within 15 Minutes — tylerjdunn · 2026-10-30