Amazon employees are using an internal AI agent called MeshClaw to automate non-essential work and inflate their token counts on company leaderboards, according to three people familiar with the matter. The behavior, which staff describe as 'tokenmaxxing,' follows an Amazon target for more than 80% of developers to use AI each week and the rollout earlier this year of internal dashboards tracking token consumption. Amazon is expected to spend $200 billion in capital expenditure in 2026, the bulk of it on AI and data center infrastructure, and the pressure to show usage is filtering down to individual engineers.
MeshClaw rolled out widely in recent weeks. The tool lets Amazon employees spin up agents that connect to workplace software, initiate code deployments, triage emails, and interact with apps such as Slack. The company said the tool enables 'thousands of Amazonians to automate repetitive tasks each day' and described it as part of an effort to let teams experiment with AI.
More than three dozen Amazon employees worked on the in-house product, according to internal documents. One internal memo describing MeshClaw said it 'dreams overnight to consolidate what it learned, monitors your deployments while you're in meetings, and triages your email before you wake up.' The tool was inspired by OpenClaw, a project that went viral in February by letting users run agents locally on their own hardware.
Key facts
- 01Amazon has set a target for more than 80% of developers to use AI tools each week.
- 02Thousands of Amazon employees now use the internal MeshClaw agent daily to automate workplace tasks.
- 03More than three dozen Amazon staff built MeshClaw, which connects to Slack, email, and code deployments.
- 04Amazon is expected to spend $200 billion in capital expenditure in 2026, mostly on AI and data center infrastructure.
- 05Internal AI token consumption has been tracked on company leaderboards since earlier this year.
The leaderboard dynamic is what's driving the synthetic usage. 'There is just so much pressure to use these tools,' one Amazon employee told the Financial Times. 'Some people are just using MeshClaw to maximize their token usage.' Amazon has told staff that token statistics will not be used in performance evaluations, and the company recently restricted team-wide leaderboard visibility so that only individual employees and their managers can view stats.
“Some people are just using MeshClaw to maximize their token usage. Managers are looking at it. When they track usage it creates perverse incentives and some people are very competitive about it.”— Jaeden Schafer
Workers don't believe the disclaimer. 'Managers are looking at it,' another current employee said. 'When they track usage it creates perverse incentives and some people are very competitive about it.' Amazon said managers are discouraged from using token consumption as a performance metric, according to a person familiar with the matter.
The pattern is not unique to Amazon. Meta employees have engaged in similar tokenmaxxing behavior to climb internal AI usage leaderboards, a sign that the largest Silicon Valley employers are running into the same measurement problem as they push generative AI deeper into daily workflows. The underlying business logic is straightforward: companies that have committed tens of billions to AI infrastructure need to demonstrate adoption to justify the spend.
Security is the other open question. Multiple Amazon employees raised concerns about granting an autonomous agent permission to act on a user's behalf, given the risk of errors or unintended actions. 'The default security posture terrifies me,' one employee said. 'I'm not about to let it go off and just do its own thing.' Amazon said it is 'committed to the safe, secure, and responsible development and deployment of generative AI for our customers.'
The 80% weekly usage target is aggressive by any measure, and it sits alongside Amazon's broader push to show capex returns. The company's $200 billion 2026 spend dwarfs prior infrastructure cycles, and AWS leadership has consistently pointed to internal AI adoption as evidence that the build-out will pay off. Token leaderboards are one of the few quantitative signals an executive team can point at when boards ask whether the spend is producing real change.
The skeptical read is that token counts are a poor proxy for productivity. A developer can run an agent in a loop, consume millions of tokens, and ship nothing. A different developer might use AI sparingly on a hard problem and produce more value. If managers — formally or informally — start treating consumption as a performance signal, the metric stops measuring usage and starts measuring gaming.
Amazon's situation is a preview of where every large enterprise AI program is heading. Mandated usage targets work as a forcing function to get tools in front of skeptical employees, but they decay quickly once workers learn the measurement. The companies that get durable productivity gains out of internal AI will be the ones that measure outputs — shipped code, closed tickets, resolved customer issues — rather than tokens consumed. Until that shift happens, expect more tokenmaxxing, more inflated leaderboards, and more AI agents quietly burning compute to keep their humans in good standing.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
