SpaceXAI Engineer Lauren Tan Reveals Running Over 30 Programming Agents with pstack
Lauren Tan, an engineer at SpaceXAI's Grok Bot, revealed that she runs over 30 programming agents using pstack in a loop, rather than managing a single chat window. The widely circulated claim of '250 applications, two years with no hires, and then earning $700,000 with 30 agents' did not appear in her resume, interviews, or this conference, and was altered in English retellings to 'he' and $850,000.
What aligns is the production claim. In a widely circulated hour-long meeting and posts, she stated that she runs over 30 agents with /pstack, merged about 1,000 pull requests last month, and plans to double that this month; by the time she wakes up, 20 have already landed on the main branch. Another version of the edit states 2,000 per month, or nearly 800 by the 12th of the month. She defines her current work as building a team of agents rather than writing code herself.
pstack is her open-source collection of skills, principles, scripts, and sub-agents, licensed under MIT. Neuron noted in a statement on September 10 that at the time, plugin version 0.15.1 had 23 task scripts and 23 engineering principles, with 47 skill entries and 2 dedicated sub-agents publicly available. The Cursor engineering team had called her personal skills 10,000 times within a week of her open-source announcement. The slogan is to go deep quickly.
She places validation before generation. One agent writes functionality, another reviews, and a third operates on a separate branch; validation skills allow agents to self-check and continue until the task is complete, which she calls critical infrastructure, not just one of many skills. Based on an estimate of 2,000 pull requests per month, manually reviewing each would leave about five minutes per request, so the review layer must run unattended and in parallel.
Her public resume is continuous, not a two-year gap. The summary states she previously worked at Cursor, Meta, and Netflix, and was involved in the core team of the React compiler. In response to a post that turned her office into a promotional image, her reply was, 'You decorated my office,' without confirming the job application numbers.
There is no verifiable $700,000 hire. What is being bought is the workflow: open-source plugins, hour-long meetings, and paid courses for traffic. Retellings have pegged the course prices at $500, $1,000, or $1,500. Beneficiaries are the agent orchestration and validation tools, while the workload model still treating a chat window as a full-time nanny is under pressure. The numbers drift in retellings, serving as traffic rather than quotes.
Source: Public Information
ABAB AI Insight
Tan's path is to turn personal habits into a callable stack. She worked in engineering management at Netflix, as an employee engineer at Meta while advancing in the React compiler core, then moved to Cursor, and now to SpaceXAI to work on Grok Bot. pstack is not a job-seeking project; it is a skill package already in use during her time at Cursor: at the time of the open-source announcement, the Cursor engineering team called it 10,000 times in a week. There is no English material supporting the two-year gap, 250 applications, and zero hires.
Resource mobilization is in the review authority, not in headcount. Over 30 agents submit requests at night, and what is seen in the morning are requests that have already landed on the main branch; her own definition is building a team, not writing code. Validation skills are singled out as infrastructure because generation speed has already surpassed reading speed. External callers use the MIT-licensed plugins, while internally the same set of scripts constrains the agents. If money occurs, it happens in tool subscriptions and courses, not in the salary numbers of that post.
The analogy is Google using error budgets to shift publishing authority from human caution to machine thresholds, and Toyota writing stop-line authority as an Andon, rather than another 'sea投逆袭' story. The software industry is in a reconstruction phase: the scarcity of writing code is decreasing, while the scarcity of proving that changes can be merged is increasing. Cursor sells the entry to agents, while pstack supplements the scripts and validation behind the entry.
This is a restructuring of the industrial chain. The output of individual engineers is no longer counted by keystrokes, but by how many parallel agents they can maintain without putting bad requests into the trunk. The mechanism is that review bandwidth is exhausted before coding bandwidth: if there are thousands of merges per month still relying on humans to read, five minutes per request will reach the limit. Thus, the role shifts from writer to scheduler, and the premium moves from language proficiency to who owns the validation stack. The $700,000 figure has no invoices, while the scheduling authority already has plugins and call counts.
ABAB News · Cognitive Laws
As generation becomes cheaper, validation begins to be priced.
Humans no longer write code; humans begin to audit fleets.
Compensation in retellings often lacks contracts.