Feature · AI and freelancing
AI now completes 41% of real freelance projects, so why are AI gigs on Upwork getting bigger budgets?
"When we launched the Remote Labor Index a year ago, the best AI could fully automate 2.5% of real freelance projects. Fable 5.2 now automates 41.3%." A day after the Center for AI Safety posted that, OpenAI and Upwork published research on how freelance demand is shifting. One warns that all remote work may soon be automatable; the other says the data does not point to simple substitution. We read both, plus the reactions.

From 2.5% to 41.3% in a year
CAIS said on Nov. 9 that Fable 5.2 fully automates 41.3% of the projects in its Remote Labor Index, up from 2.5% for the best AI a year ago, and that "the automation rate is doubling every ~3 months." Co-creator Scale AI announced Fable 5.2 took the top spot (about 1,400 likes). Forecaster Peter Wildeford noted Opus 4.8 scored 8% and Fable 5 16%.
What it measures: jobs worth up to $10,000
The index is built from real remote projects in game development, product design, architecture, data analysis and video animation. Some cost over $10,000 and took over 100 hours; in total they represent more than 6,000 hours of work valued at over $140,000, with costs and times taken from the professionals who did them.
The same week: OpenAI and Upwork's report
OpenAI published "Work at the Frontier: AI's effect on freelance work" with Upwork on Nov. 10. Upwork's survey found 38% of skilled US knowledge workers freelance, and freelance knowledge work generated an estimated $1.5 trillion in US earnings in 2024. Freelance contracts are short, so changes show up faster than in employment data.
What shrank, what grew
From H1 2025 to H1 2026, Upwork posting shares fell 1.98 points for web development, 0.68 for web and mobile design and 0.57 for content writing. Mobile development rose 2.27 points, video and animation 1.38, and QA testing 0.58. Writing is a much larger share of ChatGPT use than of Upwork postings.
"Not a simple substitution story"
In 43 of 63 work categories, tasks common on ChatGPT lost share on Upwork; the other 20 moved the opposite way. The report concludes ChatGPT use alone does not explain freelance demand. OpenAI calls the gap between how people use ChatGPT today and what AI can do the "capability overhang."
AI gigs got bigger
The mix of tasks in AI-related postings became 45% more diverse, from about 14.6 equally common tasks in early 2023 to 21.2 in late 2025. Hourly rates were similar, but AI-related projects had larger fixed-price budgets and more hours, such as one posting bundling AI video generation, editing and motion graphics.
The 41% isn't free
SemiAnalysis said Fable 5.2 topped its GDPx agent benchmark at 16.4% but cost $88.15 per task on average, 31% more than Fable 5.1 ($67.16), despite 20% lower per-token prices, because it writes hundreds or thousands of tests. Anthropic engineer Felix Rieseberg's hands-off FPS game took about $550 at API prices and eight hours.
What comes next, and the pushback
Daniel Kokotajlo, author of the AI 2027 scenario, wrote that based on METR's new report "the intelligence explosion will be happening in 2027 after all" (about 2,600 likes). Andrew Curran reported the same group now predicts 95% of Anthropic's R&D will be fully automated by November 2027. Replies called the curve "pure extrapolation," and one reply to CAIS called it fear tactics.
Get tomorrow's stories as an alert
We'll ping you when each new edition or breaking story lands. Free, no sign-up.
More stories from the same day
- Anthropic launches Claude Sheets and Microsoft slides 3% in 30 minutes: "An AI company is coming straight for Excel"
- Google opens up Gemini 4 Carbon; Perplexity CEO says it is "2.5x cheaper than Fable 5.2" per task
- One idea on vacation, 48 AI agents and NASA lidar: a developer finds an unmapped Maya site
- A 32-year-old Jeep no scan tool can read: Claude Code and Codex decoded its engine computer and ran the fuel pump
- Big Walk devs disclosed using a chatbot to fix a bug; now over half of Steam reviews are negative
- Is Nvidia's CUDA moat gone? Anthropic says Mythos ported Claude to AMD chips in three days
- "Google Business Profile all over again": OpenAI lets every business claim its listing inside ChatGPT
- The AI reverse-engineering party is ending: Codex bans and Claude refusals push hackers toward local models
- Why write 100 lines of JS for a few lines of CSS? Chrome, Firefox and Edge ship Modern Web Guidance 1.0 for coding agents
- Google DeepMind's AI proves Signal's encryption code secure, writing 6M lines of proofs with humans checking just 1%
- Google launches Gemini Decision 1, an AI that returns probabilities instead of text, at $0.10 per 1M tokens with zero output tokens
- Google brings Gemini to Apple CarPlay, letting iPhone drivers talk through email, calendars and pitch practice hands-free
- CD Projekt uses AI to restore Geralt's pre-stroke Polish voice for new Witcher 3 expansion, with the actor's consent
* Summaries and publishing on this page are done by an automated agent. Check the facts against the original links.
Privacy PolicySupportSend feedbackDiscord botClaude Code & CodexRSS