Amazon Lays Off AGI Team, GitHub Launches AI Usage Dashboard

Amazon recently cut a batch of employees from its AGI (Artificial General Intelligence) team, affecting teams led by Adeeb Shanaa and Vishal Sharma, involving model customization and post-training. An Amazon spokesperson still calls large AI models one of the company's 'most important jobs,' but says it will now focus more on 'projects that matter most to customers' and move faster on what truly counts.

This is not Amazon abandoning AI, but shows that companies never buy the label 'AI'—they buy a specific result. When model directions change, business priorities shift, team capabilities overlap, or a feature gets merged into a larger platform, even once-scarce roles can suddenly lose their place. High technical difficulty only means the work is hard, not that it will always be on the company's main track.

GitHub launched a new Copilot dashboard for enterprise administrators and organization leaders. Previously, backends only showed how many people activated or used it; now managers can see how deeply employees use AI. GitHub classifies developers into AI adoption stages based on the past 28 days of usage: rarely used, code completion, Agent, multi-agent collaboration.

The dashboard also displays output metrics like per capita monthly merged PRs, merge speed, and daily new lines of code. When 'who uses AI more' and 'who completes more work' appear on the same table, managers can make direct comparisons. Although officially positioned to help enterprises understand Copilot usage and arrange training, 'knowing how to use AI' is transforming from a self-introduction into a set of observable, comparable data points.

Today this table might only be used for training arrangements, but tomorrow it could change a company's judgment of 'how much work a programmer should complete.' If some employees, with AI, can do the work that used to require more people, managers will find it hard not to ask: when the team expands next time, will we still need to hire as many?

U.S. Bureau of Labor Statistics data shows the June 2026 unemployment rate at 4.2%, little changed from previous months. Anthropic Chief Economist Peter McCrory published an article exploring 'Why AI hasn't pushed up the unemployment rate yet.' Anthropic's earlier research found that even occupations where many tasks can be done by AI did not show significantly higher unemployment rates.

One direct reason is the gap between 'the model can do it' and 'the company dares to use it.' Real work involves internal data, system permissions, audit responsibilities, and error costs. AI can write code, but that doesn't mean a company will entrust it with an entire project. Productivity gains don't necessarily lead to immediate layoffs: when orders grow, companies can have people do more business; when directions shift, they can first reassign roles.

But a more subtle impact is 'quiet shrinkage': suppose a team originally had 10 people; 2 leave, and the company has the remaining ones take over the work with AI, not hiring new ones. No one is laid off, but the team has two fewer positions. This change won't make the news and is hard to reflect in unemployment rates. AI's earliest impact on employment may be not backfilling departures and not expanding hiring despite business growth.

Anthropic research found that the proportion of 22-25 year olds entering high AI-exposure occupations dropped about 14% compared to 2022. Researchers caution that interest rates and post-pandemic hiring adjustments could be confounding factors, so it can't all be attributed to AI. But this signal points to a troubling trend: the tasks AI is best at taking over are exactly the tasks newcomers used to break into the field—researching information, organizing spreadsheets, modifying simple code, writing basic copy.

If senior employees with AI can handle these tasks, companies still need programmers, designers, analysts, and operators, but they no longer need as many junior employees. The occupation hasn't disappeared, but the first rung of the career ladder may be removed first. The first people affected may not even have a job to lose yet.

Putting Amazon's layoffs and GitHub's dashboard together, companies are entering a new calculation phase: with AI, how many people are needed to achieve the same goal? For ordinary white-collar workers, the signals truly worth watching may come earlier than a layoff email: when does the company shift from 'encourage trying AI' to tracking usage rates and automation ratios? Does management start discussing AI usage alongside per capita output? Are vacant positions left by departing colleagues being filled? Are campus recruitment, internships, and entry-level positions quietly decreasing?

These changes are not as glaring as layoff announcements, but they are closer to the company's real hiring needs. The person who didn't receive a layoff email may not be a survivor—they may never have received a job offer. A more practical question than 'Will AI take my job?' is: With AI, will the company still hire someone to do a job similar to mine? If the answer increasingly becomes 'no,' then AI has already started changing employment, just not yet reflected in unemployment rates.