What will be left for us to work on?

2025-04-15 · Source: AI as Normal Technology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Economic Analysis & Policy · Depth: Advanced, extended

Summary

Arvind Narayanan's keynote at ICML 2026, titled "What will be left for us to work on?", addresses anxiety about increasing AI capabilities by presenting three core arguments. First, the "AI as Normal Technology" framework accurately describes AI's impacts unless recursive self-improvement creates a discontinuity. Second, while recursive self-improvement is serious, no single lab milestone will suddenly eliminate jobs. Third, future jobs will be radically different, requiring significant adaptation towards human/AI "co-superintelligence." The talk highlights a significant capability-reliability gap in AI agents, where accuracy has dramatically increased over 24 months from 3 frontier AI companies, but reliability only by 5-10 percentage points. It also introduces a four-part framework for technological diffusion (invention, innovation, diffusion, adaptation) and the "decide-execute-deliver sandwich" for software engineering, arguing that non-coding tasks remain bottlenecks.

Key takeaway

For AI professionals navigating increasing AI capabilities, prioritize developing skills in evaluation, judgment, and domain knowledge over purely technical building tasks. Recognize that AI is an amplifying tool, not a direct replacement, and strategically reinvest productivity gains into continuous learning and mastering tasks to avoid over-reliance and maintain human control, fostering a "co-superintelligence" future.

Key insights

AI is a transformative tool requiring human adaptation and a shift from building to evaluating systems, not a direct job replacement.

Principles

• AI is a normal technology unless recursive self-improvement creates discontinuity.
• Economic impacts of AI are gradual, driven by downstream adaptation, not just model capability.
• Human roles shift from "building" to "evaluation" as verifiable tasks are automated.

Method

The "decide-execute-deliver sandwich" framework analyzes software engineering tasks, identifying bottlenecks beyond coding in the decide and deliver layers.

In practice

• Reinvest AI-saved time into learning new, complementary skills.
• Resist black-box AI temptation; maintain human control and understanding.
• Master tasks yourself before using AI for augmentation to avoid dependence.

Topics

• AI Adaptation
• AI Evaluation
• Future of Work
• AI as Normal Technology
• Human-AI Collaboration
• Recursive Self-Improvement

Best for: Research Scientist, AI Scientist, Director of AI/ML, Policy Maker

Related on AIssential

• AI Won’t Just Replace Jobs. It Will Accelerate the Evolution of Everything — AI functions as a powerful accelerator, amplifying human capabilities and driving rapid technological and scientific evolution across all domains.
• New Future of Work: AI is driving rapid change, uneven benefits — AI is transforming work by enabling human-AI collaboration, not just accelerating tasks, demanding new skills and thoughtful integration.
• Will AI Replace Humans? The Quiet Reality Behind the Hype — AI causes job displacement, not total replacement, shifting demand from junior roles to "AI managers."
• How AI Changes Everything After 2030. — After 2030, AI will transition from a tool to an autonomous societal layer, necessitating human-AI collaboration for future progress.
• Is AI Really Coming For Your Job? (Not What You Think) — AI automates tasks, not jobs, necessitating a shift in professional skill sets.
• Microsoft’s Nadella Wants to Change AI’s Reputation — AI should augment human potential, acting as a cognitive amplifier rather than a job replacement.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI as Normal Technology.