According to the Financial Times, in recent weeks Amazon has begun a large-scale rollout of its internal product “MeshClaw”, allowing employees to create AI agents that can connect to company software and perform tasks on a user's behalf. Employees use this tool to automate non-essential tasks to show managers that they use the technology frequently. Not to increase productivity, but to artificially increase the use of AI. Employees say the company tracks AI token consumption, encouraging quantity over quality when it comes to the technology.
Read also: The giant is close to a trillion-dollar valuation. Is this a bubble already?
Tokenmaxxing
Some employees say the software can be used for, too automation of additional, unnecessary AI activity, just to increase the consumption of tokens – units of data processed by models. In their view, this reflects pressure to implement technology after Amazon introduced goals according to which over 80 percent programmers will use AI weeklyand at the beginning of this year it started tracking the consumption of AI tokens in internal rankings.
“The pressure to use these tools is enormous. Some people use MeshClaw just to maximize the use of tokens,” one Amazon employee told the FT. Such practices have been named tokenmaxxing.
tokenmaxxing
|
AI
Amazon has informed employees that AI token statistics will not be used to evaluate performance. However, according to employees, managers monitor this data and it is not certain whether, based on such data, they do not evaluate the employee against the background of others who develop “AI” standards.
“Managers are looking at it. Tracking AI use creates unfair practices, and some people are very competitive at it,” another Amazon employee said.
This is why there is so much emphasis on using AI
Silicon Valley groups are pushing for increased adoption of generative AI tools, a companies are struggling to demonstrate a return on massive investment in AI infrastructure and embed this technique more deeply into everyday work – often forced, without understanding and against the actual needs of employees.
When a company implements a new solution, it usually creates one KPIs (key performance indicators). When adapting AI, this takes into account, among others: the number of automated processes, the percentage of employees using AI or time and costs saved.
Read also: They named the ships after Trump. Now they will produce 15 of them
In practice, it often looks like this: a manager is given a specific goal/KPI to implement AI in the team, so he pressures people to use AI – even if the tool does not yet work well or generates a risk of errors. This makes employees feel that AI is being pushed, and the KPIs measure the use of AI rather than its actual, practical value or usefulness. However, safety and quality become secondary to KPI and reporting goals.
Amazon (illustrative photo)
|
FP Creative Stock / Shutterstock
It is estimated that this year Amazon will allocate $200 billion. for investments, the vast majority of which will go to AI and data center infrastructure. The e-commerce group has published statistics on its employees' use of AI, but recently restricted access to them so that only employees and managers can view them. Managers are discouraged from using tokens to measure performance, according to a person familiar with the matter.
What is MeshClaw
MeshClawa tool that some Amazon employees also use to inflate AI use, draws inspiration from OpenClawanother AI tool known for its promises of huge productivity improvements but also the potential for huge risks. OpenClaw, by creators' promises, clears the inbox, sends e-mails, manages the calendar, checks in for flights, all from WhatsApp or Telegram.
OpenClaw – clears your inbox, sends e-mails, manages your calendar, checks in for flights. All from WhatsApp, Telegram or any chat app you already use.
|
openclaw.ai
Unlike other AI models, OpenClaw and MeshClaw run locally on user devicesthus ensuring yourself unprecedented permissions and access to user data and applications.
Similar practices of the so-called tokenmaxxing Meta employees also used it to improve their position in internal rankings. Earlier this year Summer Yue's story went viral, Director of Security and Compliance at Meta Superintelligence Labs, when OpenClaw deleted almost her entire inbox. This shows the danger of granting too broad access to AI (and shows what people create solutions that must later be used by millions of users).
OpenClaw – at least he admitted it
|
twitter/@summeryue0
Likewise, Amazon's MeshClaw can initiate code deployments, sort emails or even use applications like WhatsApp or Slack. The company said early in a statement that the tool enabled “thousands of Amazon employees to automate repetitive tasks every day” and was one example of allowing teams to experiment and implement AI tools.
“We are committed to the safe, confident and responsible development and deployment of generative AI for our customers,” the company added.
According to internal documents, more than thirty Amazon employees worked on the tool. A recent note describing the bot said: “The bot dreams throughout the night to consolidate what you've learned, monitors your deployments while you're in meetings, and sorts your emails before you wake up.”
Read also: This is the dramatic situation on the helium market. Poland is benefiting from the crisis
Employee concerns
Many Amazon employees have expressed concerns about security risks associated with a tool that has been given permission to act on a user's behalf. This may result in situations in which an AI agent may make a mistake or take unintended actions – contrary to the employee's intentions.
Amazon employees are abusing AI for unnecessary tasks
|
AI
“The default approach to security scares me. I'm not going to let it run itself and do whatever it wants.” – said one of the employees. This shows that employees often see risks and technical problems earlier than the management, but the business is motivated by the pace of implementation and the pumping of KPIs and performance indicators.
