Logan Kilpatrick, an engineer related to Google AI Studio and Gemini API, posted that by 2027, we will see agents or mini-companies that can autonomously generate revenue (profit?) on a large scale; currently, this is happening at a smaller scale with many rough edges. The question mark in parentheses regarding "profit" was added by him—revenue and profit are written separately.

This is not the first time he has linked agents and revenue on the same curve. In January 2025, he mentioned that agents would have their own laws of expansion, with metrics including multi-agent expansion, revenue generation, and task completion. He is currently involved with Gemini, Google AI Studio, Gemini API, and Kaggle, and the post is marked as his personal opinion. Previously, he worked in developer relations at OpenAI and joined Google in 2024, aiming to make AI Studio the best home for developers.

Internally, Google writes agents as a product line. At the input-output conference, Sundar referred to the phase as the agentic Gemini era, and Kilpatrick explained: Gemini first connects dozens of unconnected products, and the next step is for the agent layer to connect them again, allowing products to take actions on behalf of users. The company trains models in conjunction with an agent framework called anti-gravity, where tool invocation and orchestration are written into the model from the start, rather than being added later. He also acknowledged that coding investment should have been an order of magnitude earlier; the image model Nano Banana attracted consumer-side traffic, but the frontier was marked by labs that had prioritized coding and agents earlier.

Mini-companies are simplified in the engineering circle as "agents plus a payment key." Some replies suggested this is a weekend project rather than a year; Kilpatrick insisted the difference lies in scale and rough edges: being able to occasionally collect payments is not the same as being able to stably generate revenue or profit without human oversight. He personally appears as an angel investor in a list of shareholders for several developer tools and agent-related startups, betting in the same direction as his public judgments, but this does not constitute evidence of scaled profitability.

The revenue curve is another measure of this prediction. Some security researchers have extrapolated laboratory revenue to trillions of dollars and compared it to the automation of the entire economy, while Kilpatrick provided a narrower product timeline: by 2027, moving from sub-scale rough edges to observable autonomous revenue-generating agent clusters. What Google aims to do is make Gemini API and AI Studio the default models and consoles for these mini-companies; once agents can pay for themselves and hire agents, the billing interface will shift from individual developer accounts to regular expenses for the company.

In market mechanisms, the buyers are developers and small teams who can collect payments and invoke tools, while the sellers are model interfaces, payment tracks, and agent frameworks. The event-driven aspect is that coding agents can now deliver runnable services, but stable accounting is lacking. Funds enter the model company from interface fees, and payment processing fees enter card organizations and stablecoin tracks; whether profits can remain on the mini-company's books is still a question mark. The beneficiaries are those who first become the default cash registers and reasoning platforms for agents; the ones under pressure are applications that only sell chat and cannot be settled by machines as suppliers, as well as the need to invent regulations for "agent companies" regarding taxes, liabilities, and bank accounts.

Source: Public Information