AI Engineer

AI Engineer organizes developer conferences and distributes free technical talks and workshops about building AI applications. Cofounded by Shawn “swyx” Wang and Benjamin Dunphy, it is operated by Software 3.0 Inc. Its programs cover agents, AI infrastructure, developer tools, retrieval, and application design; its own software includes conference submission tools and an agent-accessible knowledge library.[6][7][12][8][9]

The conference series began in 2023, before these software interfaces, around the emerging practice of building applications with pretrained foundation models.[6][1][3]

AI Engineer traces its origin to Wang’s June 2023 essay, The Rise of the AI Engineer. Wang and Dunphy launched the first Summit in San Francisco in October 2023. Contemporary programming framed this work around using pretrained foundation models without collecting and labeling data to train those models. The inaugural Summit combined stage talks, hands-on workshops, sponsor booths, and breakout sessions covering agents, AI UX, infrastructure, code generation, LLM operations, retrieval and vector databases, open-source AI, and multimodal APIs.[6][1][12]

Wang and Dunphy developed the World’s Fair as a larger, multi-track format. In his organizing retrospective, Wang explained his preferred cadence: “I was pretty set on us doing at least 2 conferences a year due to the pace of AI.” He recalled launching fully refundable blind-bird tickets at the Summit before fixing dates or a venue. While ticket sales continued, the organizers maintained soft holds on venues; Dunphy eventually found the Marriott Marquis through an existing contact.[10]

The first World’s Fair took place in San Francisco in June 2024. A June 25 announcement said it was sold out. AI Engineer’s retrospective timeline reports more than 3,000 attendees, 18 tracks, over 150 sessions, and an expo of more than 50 companies, following a first Summit that the same timeline describes as drawing 500 engineers with a 10:1 applicant ratio.[2][6][7]

The conference directory records a New York Summit in February 2025, another San Francisco World’s Fair in June, a Paris edition in September, and a New York Code event in November. The first London event, AI Engineer Europe, concluded in April 2026 after three days of online programming, workshops, and, according to the organizers’ recap, more than one hundred in-person talks. A further San Francisco World’s Fair followed in June 2026.[7][4]

The May 2026 World’s Fair call for speakers sought work on autoresearch, memory, world models, tokenmaxxing, agentic commerce, and vertical AI in law, healthcare, GTM, and finance. These were programming interests, rather than research attributed to AI Engineer itself. The organization also distinguishes its flagship conferences from independently organized partner events. Its published partner policy offers selected events marketing, operational, and speaker support, excludes financial support, and calls for good content to be shared free on the AIE YouTube channel.[5][7]

In April 2025, AI Engineer announced an open-source Model Context Protocol (MCP) server that let users chat with the conference and submit talk proposals from clients such as Cursor, Windsurf, and Claude Code. The aie-mcp repository documents a Cloudflare Worker with an SSE endpoint and KV storage for authentication and submissions. It carries an MIT license and credits @threepointone for help with the Cloudflare setup.[3][8]

Accepted speakers using that MCP workflow still had to complete their records in Sessionize. In the April 2025 announcement, the organizers explained that “Sessionize doesn’t have a write API” and that they needed a single source of truth. The interface added a proposal channel while retaining Sessionize as the required completion system.[3]

By September 2026, the public interface documentation described a broader knowledge library: six public JSON and CSV collections, advertised as approximately 16.7 million tokens, supporting discovery of conferences, speakers, talks, and transcripts. It listed a Streamable HTTP MCP server at ai.engineer/mcp, covering public data and private CODE attendee applications. Attendee applications are distinct from speaker proposals, which remain a Sessionize workflow.[9]

In his AI Engineer Europe 2026 talk, Wang described using Claude Cowork to connect Devin with Figma, then having Devin implement the website. A designer could continue iterating through annotated screenshots after Wang went to bed. He also described moving conference content into code as its source of truth, with Devin processing speaker changes from emails, screenshots, and spreadsheets: “this entire schedule, uh, is managed by Devin.”[11]

AI Engineer reports more than 1.2 million unique viewers and 10 million views of its talks, workshops, and livestreams in 2025, and more than 100,000 newsletter subscribers as of September 2026. At the Europe 2026 event, Wang described nine full-time people running a business of more than $9 million; he did not specify the financial measure or accounting period.[7][6][11]

As of September 2026, the official leadership page lists Wang as cofounder and CEO, Dunphy as cofounder, VP Sales and Creative Director, and Lia McBride as General Manager responsible for global operations, partnerships, and international expansion.[6]

AI Engineer turned its own conference operations into a test of AI-assisted software development after receiving a $43,000 annual quote for speaker-management software across four events. The team opened a challenge for builders to create a replacement, judged against the practical needs of the events staff. The Latent Space episode reports 72 submissions from people or teams.[13]

The workflow was more complicated than a simple talk-submission form: it included separate speaker and sponsor portals, headshots and slides, reminders, schedule changes and people with multiple event roles. Contestants received a brief, a product walkthrough and 42 screenshots; the challenge offered a $10,000 winner’s prize and token reimbursement for serious attempts.[13]

The organizers also built an AI judge that logged in as different roles and checked 20 user scenarios. It helped builders find missing functionality, but the events team made the final decision. The eventual winner ranked 19th on the automated leaderboard, illustrating why the team used hands-on human testing alongside the automated evaluation.[13]