Local Developer Burning 38 Million AI Tokens A Month Still Bravely Has No Product, Users, Or Clear Idea What He Is Building

SAN FRANCISCO – Software developer Tyler Ng, 29, confirmed that despite spending roughly $4,300 a month across AI coding subscriptions, inference providers, vector databases, agent frameworks, GPU credits, and something he called "autonomous orchestration tooling," he has still not technically shipped anything another human being can use.

Ng, who describes himself online as a solo founder building at the edge of human-AI collaboration, reportedly begins each morning by opening ChatGPT, Claude, Cursor, Perplexity, GitHub Copilot, Replit Agent, three terminal windows, and one open-source tool he installed from a Discord after seeing a screenshot with 11 fire emojis.

The man has absolutely entered the zone.

Friends say Ng now spends about 14 hours a day generating large codebases for products that are abandoned as soon as he thinks of another product with slightly stronger moat potential.

"He's built eleven startups this month if you count folders," said former coworker Matt Reynolds. "If you count customers, he has built zero. But the folders are incredible."

According to browser history reviewed by VanFlip, Ng's recent projects include an AI-powered second brain infrastructure layer, an autonomous content repurposing platform, a vibe-coded personal CRM memory system, a founder accountability co-pilot, and one repo named `new-new-real-one-final`.

None currently have users.

One briefly had a landing page, although the waitlist form was connected to a database Ng later deleted during a migration he described as "not technically necessary, but it made the repo feel cleaner," which is how men talk when they have lost contact with commerce.

Industry analysts say developers like Ng represent a fast-growing category known inside AI infrastructure companies as infinite prototype men, a market segment defined by enormous token consumption, aggressive stack switching, and a near-religious belief that the next rewrite will reveal the product.

"These founders consume unbelievable amounts of compute," said AI infrastructure analyst Claire Morton. "One man spent $11,000 generating a TypeScript app that displays the weather, but with more agent memory."

Ng insists progress is real.

"People focus too much on shipping," he said from a standing desk surrounded by energy drinks, supplement bottles, and a whiteboard that still said "MVP" under six layers of arrows. "Right now I am refining the orchestration layer."

Asked what the orchestration layer did, Ng paused for several seconds, looked at his token dashboard, and replied, "It lets agents talk to agents."

His apartment now resembles a command center for absolutely nothing. One monitor shows token usage climbing like a hospital chart. Another displays a diagram of agents routing tasks to other agents that appear to route the tasks back to the first agents. A third exists only to monitor API spend in real time, which Ng calls "momentum."

"He watches the costs rise and nods," Reynolds said. "It is like seeing a man warm his hands over a burning invoice."

Ng's monthly recurring revenue remains $0. His user count is technically one, if you include Ng testing the app at 2:17 a.m. while logged in through a broken admin account. A bot signed up once during local testing, then crashed after being asked to choose a pricing tier.

Still, the validation has been enough to keep the project alive. Ng recently received 14 likes on a build-in-public thread, one "this is huge" reply from an anime avatar, and a newsletter mention describing agentic workflows as the next SaaS abstraction layer.

He has been eating off that for nine days.

Experts say AI tools have created a strange new productivity loop where developers can generate software faster than reality can reject it. Before AI, a useless startup required months of lonely effort, a regrettable logo, and at least one Stripe integration. Now a man can build three dead companies before lunch and still have the afternoon free to pivot into memory architecture.

Ng reportedly burned through 9 million tokens last weekend attempting to build a self-improving autonomous code generation ecosystem before realizing midway through that he had reinvented Trello, except slower and more expensive.

He took the discovery well, immediately opening a new repo called `trello-killer-but-ai`.

At press time, Ng was announcing a complete strategic pivot toward AI video agents after seeing another founder post, "text is dead," followed by a graph with no labeled axes.