Anything AI ships AI "employees" with phone numbers. Watch the agent-per-head ratio

Anything AI has released Skydive, a product that lets customers create AI agents to fill internal roles either by describing the job or by picking from a catalog of four templates, according to reporting by Fast Company Design published August 20 ^[1]. The agents connect to Slack, email and the web, each gets its own phone number so it can be reached on iMessage, and they keep running when human staff are offline ^[2] - which means the buyer is not purchasing a tool so much as a line item that behaves like a colleague. The vendor is unusually exposed here, because it runs the product on itself. Roughly 150 agents operate inside the company, about 10 of which co-founder Dhruv Amin describes as essential to operations, and each human employee directs somewhere between 10 and 15 ^[6]. Work backwards and that implies a human team of roughly 10 to 15 people ^[7]; the essential subset is under 7 percent of the fleet ^[8]. Anything was founded last year by Amin and Marcus Lowe, both former Google product leads, as a platform for nontechnical people to build apps with AI ^[3], and it raised an $11 million round at a $100 million valuation from Footwork, Uncork Capital, Bessemer Venture Partners and M13 ^[4]. The company says more than a million people have used Anything to build millions of apps, including project trackers and CRM platforms ^[5]. The internal roster is specific enough to audit. Emma investigates reported bugs across logs, databases and code and drafts replies; Otis runs engineering operations and forecasts launch dates; Clementine handles brand imagery; Tilda manages the CEO's inbox and calendar and briefs him each morning; JB, the office manager, places Instacart orders and tells the team when they arrive; another agent acts as Amin's vice president of finance ^[9]. They carry human names and animated, mouthless avatars that Amin says are deliberately "not quite human": "We're not trying to trick someone that this is a human" ^[10]. Emma exists because of a support crunch: in May, after the AI app builder Mocha announced its shutdown and pointed users toward Anything, support volume ran to roughly triple normal, and Amin says an employee built Emma and added it to the help desk rather than hiring support staff ^[11]. More recently, an engineering agent helped Lowe ship 56 pull requests in a single Sunday while he directed it from Slack on his phone ^[12]. The failure log is on the record too, which is the more useful part. Amin says an agent misread a database calculation and told him the company had lost $1 million to a credit bug that did not exist ^[13]. An employee who asked an agent to write emails to potential influencer partners without reviewing drafts had the draft version sent out ^[14]. Marketing lead Zaria Zinn asked for one email draft and got several, already loaded into the email provider, and says the tendency to over-deliver makes it necessary to spell out where an agent should stop ^[15]. "You deal with the mess ups because like in some cases there's like, no other choice," Amin says, citing the cost of running a full influencer program ^[16]. Context for the pitch: Grand View Research values the market at $10.9 billion and projects $182.9 billion by 2033 ^[17], a roughly 17-fold increase ^[18], and a 2025 McKinsey report found 62 percent of respondents saying their organizations are at least experimenting with AI agents ^[17]. Nvidia earlier this year announced agent-building software and racks designed to run agents, shifting its strategy's primary focus away from GPUs ^[19]. Watch whether the 10-to-15 ratio survives growth. Amin says he expects to hire more humans as costs fall and revenue grows, but with a higher bar, and that junior developer roles at Anything are becoming more senior, though the team recently hired a 21-year-old college dropout ^[20].