What Anthropic's hiring data reveals about AI

Anthropic is hiring fewer juniors as AI takes on more work

At Anthropic, AI agents now log six times as many working hours as employees.

That's a striking number.

In a new internal study, Anthropic found that junior technical hires fell from 27% to 9% of technical hiring since early 2025, even as the company continued expanding.

The work is changing too.

Engineers are producing more code, while employees increasingly spend their days reviewing Claude's output, managing agents, and deciding what needs human attention.

I think the hiring numbers matter most here. Anthropic can retrain existing employees and move them into new roles, but that doesn't solve the problem for graduates who never get hired.

There's another interesting detail: Claude sessions spend roughly 80% of their time waiting for human input. More automation is creating more work around supervision.

If other companies follow Anthropic's approach, entry-level opportunities could shrink before the broader workforce sees major job losses.

Where will people gain experience if companies increasingly automate the work that used to train beginners?

ServiceNow wants every AI agent to prove its permissions

An AI agent could soon need its own verified credentials before accessing company systems.

ServiceNow has announced AI Passport, a system that checks what AI agents are authorized to do while they're running, including agents built by other companies.

It can restrict an agent's actions or shut it down when it exceeds its permissions. The company is working with NVIDIA on enforcement technology, with availability expected in Q4 2026.

This matters more than it sounds.

Companies are giving AI agents access to customer records, internal tools, and business workflows, often without a reliable way to verify every action.

I think ServiceNow is addressing a problem that will become harder as businesses deploy hundreds or thousands of agents. One customer, LTM, already has 1,500 AI agents, each assigned an employee ID and manager.

The harder question is whether these controls will work consistently across different vendors and systems.

Will businesses trust AI agents with more responsibility once they can control exactly what those agents are allowed to do?

Meta's new AI model runs locally, if your Mac can handle it

Meta has released an AI model that can run directly on a Mac, provided it has at least 1 terabyte of RAM.

That's a serious hardware requirement.

Called Muse Spark Open, the 816-billion-parameter model supports coding, video editing, and AI agents, with a 1-million-token context window.

Meta demonstrated it fixing bugs in a Mastodon client and reviewing footage in DaVinci Resolve. The model is also available through API platforms.

I think the local execution is the interesting part. Companies could potentially run capable AI models on their own machines, giving them more control over sensitive data and how the models are used.

But requiring 1TB of RAM puts this well beyond what most Mac owners have sitting on their desks.

For now, this looks more relevant to developers and organizations with expensive hardware than everyday users. If hardware requirements fall, local AI could become a more practical alternative to cloud services.

How much will companies pay to keep powerful AI running on their own hardware?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.