Nan Yu Announces Departure from Linear to Join OpenAI, Will Work on Codex and ChatGPT Products
Nan Yu Joins OpenAI to Lead Product
Nan Yu announced his departure from Linear to join OpenAI, where he will participate in the product work for Codex and ChatGPT; he stated that he worked at Linear for four years and hopes to bring the software product experience he accumulated there into the next phase.
This public statement confirms his destination and product responsibilities but does not disclose Nan Yu's official title, reporting relationships, start date at OpenAI, or the specific division of responsibilities between Codex and ChatGPT, nor does it mention whether he will directly manage the two product lines.
Nan Yu was an early product lead at Linear and previously served as its Head of Product; he was also the Vice President of Product at Mode and held the position of CTO at Everlane, with a career path spanning engineering, data products, consumer internet, and collaborative software. Existing public search results do not provide independent verification of this background in English, so it should be regarded as his existing public career background rather than confirmed content from this appointment announcement.
Linear emphasizes small teams, product judgment, and engineer autonomy within its product system. Nan Yu has previously discussed the organizational methods for involving engineers in product decisions; this background signals that his joining OpenAI is not just about hiring a PM, but also about bringing a collaborative software product approach aimed at developers and knowledge workers into Codex and ChatGPT.
The timing overlaps with Linear's progress on its Agent product. Linear launched Loops in July, advocating for Agents to continuously execute team tasks based on schedules or event triggers; OpenAI is advancing Codex from code completion and single-task execution to agent-based products that can handle development workflows over longer time scales. In publicly verifiable information, Nan Yu himself did not mention Loops or "Persistent mode" in his departure announcement.
From a market mechanism perspective, OpenAI is not competing for a single product manager but for the daily usage entry point of Agent products: developers and teams are willing to pay for tools that can understand project context over the long term, continuously advance tasks, provide reviewable outputs, and integrate into workflows. Beneficiaries include Codex, ChatGPT, OpenAI enterprise subscriptions, and the developer ecosystem; Linear loses a core leader deeply involved in its product methodology and Agent direction. Ultimately, funding will flow to products that can translate model inference into stable task completion rates, team retention, and charges per seat or per task, rather than just to vendors that lead in model capability testing.
Source: Public Information
ABAB AI Insight
The product influence of Linear comes from its reverse selection against the complexity of Jira-style project management: it emphasizes speed, keyboard-first, restrained interfaces, and engineer-led workflows, rather than piling every collaboration need into generic enterprise software. Nan Yu's path from engineer, Everlane CTO, Mode product head to Linear Head of Product illustrates that he has long dealt not with abstract model research but with how to make high-complexity tools voluntarily used by teams every day. OpenAI's absorption of such talent fills the gap in the "model capability to work habit" product transformation stage.
The capital path is shifting from pre-training investment to application layer retention. Training foundational models relies on GPUs, data centers, and electricity, but revenue must be realized through ChatGPT subscriptions, enterprise seats, API calls, and developer tools. If Codex is merely a one-time code generator, users can switch between multiple models; however, if it integrates into codebases, task queues, testing, code reviews, deployments, and team collaboration processes, the switching costs will become embedded in context, permissions, history, and workflow dependencies. By placing product leaders in Codex and ChatGPT, OpenAI is effectively putting consumer-grade and developer-grade entry points within the same commercialization loop.
Historically, the value of GitHub lies not just in code hosting but in the workflows formed by pull requests, CI, permissions, issues, and developer social graphs; Slack did not win solely on chat features but through channels, bots, and app integrations that entered organizational collaboration. Linear's Loops represent the next stage: software no longer just records task status but allows Agents to directly push tasks forward around events and plans. OpenAI is currently in the phase of expanding Agents from "single-turn generation" to "long-term execution, recoverable, and reviewable" products.
Essentially, this is about technological substitution: models will not directly replace developers or knowledge workers; instead, tasks such as task distribution, information retrieval, status updates, repetitive coding, draft generation, and process follow-up are the first to be substituted. Whoever can string these segments into a continuously operating Agent while retaining final approval rights for users will be able to convert inference costs into software revenue. The importance of product leaders thus increases: models are responsible for generative capabilities, while product systems handle permissions, memory, failure recovery, collaboration boundaries, and user trust; without the latter, even the strongest models will struggle to become tools that users open daily.
ABAB News · Cognitive Laws
Models create capabilities, workflows determine retention
Agents replace tasks but will not automatically replace responsibilities
The stronger the technology, the more the product determines who gains users