## Core Concept
**27% of AI-assisted work wouldn't have been done at all without AI.** Productivity gains from AI come not primarily from doing existing work faster, but from expanding the total volume of work that gets done — creating output that was previously below the threshold of economic viability.
## The Reframing
The dominant productivity narrative is **speed**: AI makes you 2x faster at the same tasks. The data reveals a different story — **expansion**: AI enables work that was never economically justified before.
**Traditional view**: Same work, done faster → time savings
**Actual pattern**: Same work done faster + net-new work that wasn't viable before → output expansion
The 27% additionality figure means that for every 4 tasks completed with AI, approximately 1 is genuinely new — work that would have been deprioritized, deemed too expensive, or simply never considered.
## What Gets Expanded
| Category | Examples |
|----------|----------|
| **Below-threshold tasks** | Documentation nobody wrote, tests nobody had time for, code cleanup that was always deferred |
| **Exploration work** | Prototypes, proof-of-concepts, alternative approaches evaluated in parallel |
| **Quality improvements** | More thorough testing, better error handling, comprehensive edge case coverage |
| **Scope expansion** | Features that were cut from roadmaps due to cost, small improvements that were never worth a sprint |
| **Maintenance debt** | Refactoring, dependency updates, security patches that accumulated because they weren't urgent |
## Strategic Implications
- **Job impact reframing**: AI doesn't just threaten existing jobs — it creates demand for work that didn't exist. The employment question is whether new work creation exceeds old work automation
- **Value measurement**: Measuring AI productivity purely by speed misses the larger impact. Organizations should track *what gets done that wasn't done before*
- **Individual positioning**: The competitive advantage isn't being faster at assigned tasks — it's identifying and executing high-value work that was previously below the viability threshold
- **Economic dynamics**: Output expansion at 27% represents significant GDP growth potential — not redistribution of existing output but genuine creation of new value
## Connection to Existing Concepts
- **[[AI Scutwork Thesis]]** — Paul Graham's "scutwork" framing focuses on automating existing drudgery; output expansion goes further by creating *new* work categories
- **[[AI Distribution Amplification Thesis]]** — Output expansion is the mechanism through which AI "pushes out tails" — driven people produce more because previously unviable work becomes doable
- **[[Elite Small Team Leverage]]** — Small teams with AI don't just match large team speed — they expand total output beyond what large teams attempted
- **[[Day-One Build Capability]]** — Onboarding compression enables net-new productivity from day one, contributing to the 27% additionality
- **[[Engineer as System Orchestrator]]** — Orchestrators identify and direct output expansion; implementers just speed up existing work
- **[[Joy Theft Paradox]]** — Some expanded output may be work engineers don't find meaningful, contributing to satisfaction erosion even as productivity rises
## Source
- **[[2026 Agent Coding Trends Report]]** — Anthropic, "2026 Agentic Coding Trends Report" (Trend 6: Productivity through output volume, not just speed)
## Metadata
**Created**: 2026-02-09
**Primary Domains**: Future of Work, AI Productivity, Economic Impact
**Related Topics**: [[Future of Work]], [[AI-Assisted Development]], [[Durable Competitive Advantage in the Age of AI]]