AI Agents Complete 16% of Freelance Work at Professional Quality — More Than 6x Growth in 8 Months
According to the latest Remote Labor Index data reported by The Decoder, the share of paid freelance projects that AI agents can complete at professional quality has surged from 2.5% to 16% in just eight months. This benchmark measures how autonomously AI agents can complete real paid engagements, meaning the top-level automation rate has jumped roughly sixfold.
What this figure suggests is the reality that AI agents are rapidly transitioning from "demo-stage technology" to "tools ready for real-world work." In particular, in highly routine task areas, there appear to be situations where AI can already deliver output at quality on par with human freelancers.
That said, 16% represents only about one-sixth of the total, and the remaining 84% of projects still require human skills. Which categories of work are more susceptible to automation—and where humans maintain their edge—will be a key lens for anticipating structural changes in the labor market going forward.
The Reality of Corporate AI Investment: Usage Throttled Over High Costs, Failed Projects Cost 2.4% of Revenue
While AI capabilities continue to advance, the companies adopting them are beginning to face the realities of cost and ROI.
According to reporting by 404media, multiple companies have started restricting (throttling) employees' AI usage because costs are "too high." AI tool usage has reportedly grown beyond expectations, with monthly subscriptions and API fees straining budgets in case after case.
A CIO Dive article corroborates these concerns. A survey found that U.S. companies lose approximately 2.4% of their revenue to failed AI projects. The findings bring into sharp relief how inadequate AI governance and strategy-less adoption are generating wasteful spending.
Furthermore, according to Bloomberg, German software giant SAP has announced plans to restrict hiring and travel in order to fund "significant" investment in AI. The judgment is that internal cost cutting is essential to winning the AI race—a sign that even at major tech companies, the burden of AI investment is starting to weigh on their finances.
What emerges from these three reports is a turning point: the overheated expectation that "AI is a magic wand that solves everything" is cooling, giving way to an era of realistic cost-benefit trade-offs. Successful AI adoption will demand clearer objectives and continuous measurement of results more than ever before.
India's Supreme Court Quashes Ruling Based on AI-Generated Fabricated Case Law — Warns It Is "Catastrophic"
India's Supreme Court has quashed a ruling that was based on non-existent case law generated by AI (a hallucination). According to The Hindu, the judge described AI-generated fabricated case law as the "methyl isocyanate of law" and warned that it is "catastrophic."
Methyl isocyanate was the substance responsible for the 1984 Bhopal chemical disaster, making this metaphor an extraordinarily stark expression of the danger AI hallucinations pose to the legal system.
This incident arose because lawyers and legal professionals used AI tool outputs in court documents without verification. AI can generate plausible-sounding text, but neglecting fact-checking—especially verifying that cited cases actually exist—can lead to situations that shake confidence in the judicial system itself.
With similar problems being reported in courtrooms around the world, this decision by India's Supreme Court could set an important precedent for how AI-generated information is handled. For professionals who use AI tools in their work, "how to verify generated information" is a challenge that can no longer be avoided.
Nvidia Expands Funding for AI Startups, Set to Ease Big Tech's Chip Dominance
According to The Decoder, Nvidia is aggressively expanding its funding of AI startups, beginning to play a role akin to a "central bank" in the compute market.
Nvidia's intent is clear. Today, most of the purchasing power for AI compute is concentrated in Big Tech companies such as Microsoft, Google, and Amazon. While these companies buy Nvidia's chips in massive volumes, they are also shifting toward their own in-house chips, creating customer concentration risk for Nvidia.
By providing funding to startups and equipping them with compute purchasing power, Nvidia is working to diversify its customer base. As a result, the situation in which access to AI infrastructure is skewed toward a handful of giant corporations could be alleviated.
However, the structure in which Nvidia itself selects and backs specific startups carries the risk of creating new dependencies. How this affects the competitive landscape will likely depend on the terms and transparency of that support.
Z.ai Releases "ZCode" — Taking On Cursor, Claude Code, and GitHub Copilot
According to VentureBeat (via Reddit's LocalLLaMA), Z.ai has released the AI coding tool "ZCode," entering a market dominated by Cursor, Claude Code, and GitHub Copilot.
AI coding is one of the most fiercely competitive areas of 2025–2026, with new players continually entering the field. As for ZCode's differentiators and actual coding quality, we will have to wait for user feedback.
This move is yet another sign that AI coding tools have taken root as a standalone product category rather than a mere "add-on feature for developers."
Google Adds TikTok-Style Short Videos to NotebookLM
According to The Decoder, Google has added a short-form video summary feature—using a format popularized on TikTok—to its learning and research tool NotebookLM.
NotebookLM has previously offered summaries in text and audio podcast formats; with the addition of this video-shorts format, users will be able to grasp the content of documents intuitively through short videos.
As the format of information consumption shifts heavily toward short-form video, learning and research tools are increasingly adopting the format as well. How accurately academically or technically complex content can be conveyed in short videos, however, will depend on future feature improvements.
---