Swyx

Swyx is Shawn Wang, a developer, writer, and community builder whose AI work spans Latent Space, the AI Engineer conferences, and coding agents. His essay The Rise of the AI Engineer proposed a software-engineering specialization devoted to turning foundation-model capabilities into useful products. He co-founded the conference series with Ben Dunphy, launched smol-developer in 2023, and announced his decision to join Cognition in September 2025. The alias “swyx” comes from the initials of his English and Chinese names.[1][4][2][6]

In his June 30, 2023 essay, Wang argued that applying AI had become a full-time discipline: engineers needed to evaluate models, work with retrieval and agent tools, and productize capabilities that model researchers had not necessarily anticipated. He compared the emerging specialization to site reliability, DevOps, and data engineering, while qualifying his contribution: “I am calling attention to this trend rather than starting it.”[4]

His framing gives product-specific data and evaluations a central place in AI engineering. The boundary between using and developing models is permeable: AI engineers may fine-tune or host models, and research engineers may build atop APIs. Wang also emphasized human-written code that orchestrates or supplements LLM capabilities, alongside agents that generate code. The role therefore extends beyond writing prompts or simply calling a model API.[4]

The essay also announced the AI Engineer Summit. Wang explained that he had community experience but had never run a 500-person conference, so he teamed up with Ben Dunphy, the organizer behind Reactathon. They launched the first Summit in October 2023. Wang credited his Latent Space cohost Alessio Fanelli, Sarah Guo, and Pranav Reddy with reviewing the essay and providing feedback and support, and Dunphy with co-founding the conference series and network.[4][2]

Wang's 2023 experiments included smol-developer, which he later described as a bundle of prompts and a resilient workflow built on Modal. The project's intended loop keeps the developer involved: write an application specification, generate code, run or inspect it, clarify missing requirements, and feed errors back into the prompt. Its README calls this “engineering with prompts, rather than prompt engineering”; the human can take over the codebase when the AI stops adding value.[6][5]

One concrete design problem was consistency across generated files. The project introduced an intermediate shared_dependencies.md plan and reused it during each file's generation to reduce incompatible cross-file references. The README acknowledges that the plan could still miss hard dependencies; specifying exact names in the prompt was a workaround. It credits Modal with Python dependency management, parallel generation, and API retries with backoff, while describing hosted endpoints and attached-storage uses as future directions.[5]

Wang later recalled that, when he retried the same agent one week after launch on the same endpoints, it could no longer perform tasks he had tested. He said this undermined his confidence and led him to other experiments, including AINews. In explaining his subsequent Cognition decision, he described its work as the direction he wished smol-developer had taken.[6]

In the August 19, 2025 Latent Space conversation with Chroma's Jeff Huber, Wang acknowledged Huber's influence on his programming decisions: “Yeah, I killed the rag track partially because of your influence.” He also questioned marketing that used perfect needle-in-a-haystack results to suggest reliable utilization of very large context windows.[3]

Wang endorsed examining and labeling data and building golden datasets for evaluation. He described Chroma's generative benchmarking—using an LLM to create queries from known chunks for retrieval evaluation—as a larger personal insight than its context-rot report because developers could apply it to their own systems. His remarks on memory were more exploratory: he suggested a batch-processing cycle analogous to sleep or garbage collection, while questioning whether human-memory analogies fit AI. He also proposed that reasoning models might revisit context more effectively; Huber immediately mentioned potentially contrary research.[3]

On September 8, 2025, Wang announced that he would join Cognition, the maker of Devin and owner of Windsurf. Most of Smol AI would join as part of the deal, while he would continue AI Engineer and Latent Space independently of Cognition. He credited Lia with stepping up to General Manager of AI Engineer and said he retained the Smol AI domain, his existing code intellectual property, and AINews.[6]

His rationale combined a belief that coding AI would deliver value sooner than general AGI, a preference for product-first “agent labs,” and interest in covering both synchronous IDE work and asynchronous delegated tasks. He favored the asynchronous side: “Slack is the killer agent UI,” he argued, because people already assign work to remote colleagues and communicate while waiting for results. Enterprise usage pricing and the team also shaped his decision. In his account, Scott visited his house, spent substantial time with him, and handwrote an offer while Wang was still considering a job he had not been seeking.[6]

Wang attached limits to those arguments. He acknowledged that the agent-lab thesis lacked a strong answer to “what if OpenAI does this?” and described Cognition's combination of synchronous and asynchronous products as potential that still required follow-through. His decision was a bet on the company and its approach, rather than a claim that those questions had been settled.[6]

Before his AI work, Wang moved from finance into software engineering. Raised in Singapore, he worked as a currency-options trader and then a technology, media, and telecom hedge-fund analyst. His finance work included Python and Haskell applications for quantitative trading, derivatives pricing, and portfolio management. He describes the career change plainly: “Didn't like the trading part, loved the coding part.”[1][7]

His earlier roles included Software Engineer at Two Sigma, Senior DX Engineer at Netlify, Senior Developer Advocate at AWS, and Head of Developer Experience at Temporal; his developer-experience work also took him to Airbyte. He helped run the React subreddit, co-founded Svelte Society, and wrote The Coding Career Handbook for developers progressing from junior to senior. This work supplies the earlier professional and community context for his later AI writing and conferences.[7][1]

His Learn In Public essay, dated June 19, 2018, advocates creating tutorials, cheatsheets, talks, and other resources while learning. Its guiding advice is: “Whatever your thing is, make the thing you wish you had found when you were learning.” Wang urges developers to document solved problems, accept correction, and contribute to the work of more experienced engineers. He presents teaching and public contribution as ways to deepen learning and gain mentorship, rather than judging progress by popularity metrics.[8]