The Signal — March 15, 2026

THE SIGNAL

Future Shock Daily — March 15, 2026

Anthropic quietly made one of the biggest developer-facing moves of the month. Meta's bad week got worse. And China is funding a future where one person plus an AI agent equals a company.

Anthropic Opens the Full 1M Context Window to Everyone

Anthropic made its 1 million token context window generally available for both Claude Opus 4.6 and Claude Sonnet 4.6 on Thursday, and dropped the long-context surcharge entirely. Standard API pricing now applies across the full window.

That second part is the real news. Extended context was already available in beta, but the premium pricing made it impractical for many production use cases. Removing the surcharge means developers can now feed Claude up to 1 million tokens — roughly 700,000 words, or about 10 average-length novels — at the same per-token rate as a short prompt. Media limits also expanded to 600 images or PDF pages per request.

For developers building retrieval-augmented generation systems, coding assistants, or document analysis tools, the cost math just changed. Long-context windows had been priced as a premium feature by every major lab. Anthropic is betting that making it standard drives adoption faster than the revenue hit from dropping the surcharge.

Consider the competitive backdrop. Google's Gemini 2.5 offers a 1M context window, and OpenAI has been expanding its own limits. But context length without affordable pricing is a demo feature, not a production tool. Anthropic just turned it into a production tool.

Sources: Anthropic Blog · Simon Willison · The Decoder

Reuters: Meta Considering Layoffs Affecting Up to 20% of Workforce

The Meta story got bigger over the weekend.

We noted Friday that Investor's Business Daily reported Meta might cut 20% or more of its workforce to fund its AI infrastructure spending. Late Thursday, Reuters published a detailed exclusive — triple-sourced by reporters Katie Paul, Jeff Horwitz, and Deepa Seetharaman — reporting that Meta is planning layoffs affecting up to 20% of its roughly 79,000 employees. That's approximately 15,800 positions.

Meta spokesperson Andy Stone called it "speculative reporting about theoretical approaches," which is the kind of non-denial that typically means the story is directionally correct even if the final numbers shift. The Verge, Business Insider, TechCrunch, Forbes, and Fox Business all independently confirmed elements of the Reuters report.

The context makes the layoffs harder to dismiss as routine belt-tightening. This is a company that has committed $135 billion to AI infrastructure, just delayed its flagship Avocado model after it failed internal benchmarks, and is reportedly weighing a deal to license Google's Gemini as a stopgap. Cutting 20% of your workforce while your AI strategy is underperforming creates a compounding narrative problem: the money is flowing, the results aren't showing, and the people are leaving.

This would be Meta's third mass layoff cycle in four years. The employees who survived the 2022-23 cuts were told the AI future would justify the pain. Now that future is delayed and the cuts are back.

Sources: Reuters · The Verge · Business Insider · TechCrunch

China Is Subsidizing One-Person AI Companies

While American tech companies are cutting tens of thousands of jobs, China is doing something different with AI: paying people to start solo companies powered by agents.

Multiple Chinese cities and provinces have rolled out subsidy programs specifically targeting "one-person companies" — businesses where a single founder uses AI agents to handle functions that would traditionally require a full team. The subsidies range from direct grants to office space, computing credits, and fast-tracked business registration. Wired and The Decoder both covered the policy push, which sits inside China's broader 15th Five-Year Plan goal of integrating AI into 90% of the economy by 2030.

The concept inverts the standard narrative about AI and labor. Instead of "AI replaces workers at existing companies," the Chinese model asks: what if AI lets one person do the work of ten, and you build an economy around that? It treats agent capability as a force multiplier for entrepreneurship rather than a headcount reduction tool.

Whether this works at scale is an open question. Solo founders still face the same market, regulatory, and operational challenges that kill most startups. Agents can handle tasks, but judgment, relationships, and strategy still rest with the human. Still, as a policy experiment, it's the most direct government bet on AI-augmented solo entrepreneurship we've seen from any major economy.

Sources: Wired · The Decoder

On the Editor's Desk

Saturday's pipeline delivered 35 events, and the significance scoring was too generous again — two items came in rated sig-5 that our fact-checking process downgraded to sig-3. Mirendil, a startup from ex-Anthropic researchers Behnam Neyshabur and Harsh Mehta, got the full treatment: The Information broke it, a16z and Kleiner Perkins are reportedly co-leading at a ~$1B valuation. Interesting, but a well-funded startup launch isn't field-defining until there's a product. We're holding it.

Elon Musk admitted that xAI "was not built right first time around" and announced a full restructuring. Three independent sources confirmed it. We considered including it but decided the Anthropic pricing move and the Meta layoffs tell a more coherent story about the competitive landscape this week: one lab is making its technology cheaper and more accessible while another is spending $135 billion and falling behind.

Garry Tan's gstack (an open-source Claude Code system) hit 12,000 GitHub stars. Cool developer tool, but a YC president releasing a side project doesn't move the needle for this publication.

We killed 12 items total — three duplicates, five non-news opinion pieces, and four off-topic articles including NPR tax advice that somehow ended up in our AI pipeline.