6 Things to Know about AIE World's Fair 2026

What should a newcomer prioritize at AI Engineer World’s Fair, and what has changed for someone returning? In this preview of World’s Fair 2026, co-founder Shawn “swyx” Wang introduces the choices behind the program. He describes a 15-person organizing team preparing the show. The guide is for both first-time attendees and veterans who might otherwise assume they already know the format.

The preview advertises $38,000 in sponsor offers, rather than cash or guaranteed savings. Eligibility and redemption terms are not explained. Swyx organizes the guide around six changes for an event planned around 6,000 people: its size, the expo, research participation, leadership programming, AI verticals and side events.

More choice is the purpose of scale. Swyx presents this as the largest AIE yet, but acknowledges that a conference can become too large. His preferred analogy is a buffet: organizers curate a range of subjects, and attendees choose the parts that matter to them. That approach has developed across successive AIE events, alongside gatherings in Melbourne, Miami and Singapore and plans for Paris and Chicago.

In discussing YouTube growth, swyx describes Europe as roughly 10× the size of World’s Fair 2024, without identifying the precise metric. This is not an attendance comparison. For the program itself, his stated balance is approximately 50% evergreen topics and 50% newly relevant replacements. That keeps useful foundations while making room for work that was not a distinct track the previous year.

The changes include:

• Autoresearch: a new track for a subject that did not have its own place last year.
• Inference, post-training and data quality: a more specific division of material previously grouped under GPUs, with pre-training discussed through the data-quality track.
• Memory & Continual Learning and Verticals: additional areas reflecting changes in what engineers are building.

The preview also announces an extra day of content compared with the previous year. Its practical travel advice is to stay through the July 2 programming rather than plan to leave that day.

Recorded talks make in-person conversation more valuable. The World’s Fair format is intended to support an entire day spent in the expo and hallway conversations. The preview describes four expo stages for shipping updates, launches and questions. Recorded presentations remain available later; the informal exchange with someone building a product is harder to reproduce.

The entry-level Expo Explorer ticket is presented as a way to wander the floor and catch up on AI engineering through exhibitors and conversations. That is a specific access proposition, not a promise that the cheapest ticket includes every conference benefit.

Swyx estimates the expo at roughly 4× its 2025 size in the preview. The subsequently updated event page says 3×; neither comparison defines a measurement method, so these should not be read as measured floor-area growth. The physical plan brings the event together at Moscone West, with entertainment as well as business programming. Streets receive names such as Attention Avenue, Backdrop Boulevard, Context Crescent and Diffusion Drive. Playful water-station details and robots add to an expo intended to reward exploration, while broad exhibitor participation makes it a place to survey the field.

Expanding the audience also means bringing research and engineering closer together. Swyx recalls Greg Brockman describing people moving between those roles during their previous World’s Fair conversation. The boundary inside a lab can be more fluid than it appears from outside, and AIE aims to make that exchange accessible to its own community.

AIE positions itself as an industry counterpart to academic conferences such as ICML and NeurIPS. Its partnership with the ACM Conference on AI and Agentic Systems brings selected research into the industry program. The partnership concerns selected work, rather than every ACM paper: the official description identifies accepted papers designated Industry Spotlight or Operational Experience for presentation at both venues.

Poster sessions provide a different discovery mechanism from a scheduled talk. Someone who has spent substantial time on a problem presents the work; an attendee walking past can notice something unfamiliar and ask its creator about it. Swyx contrasts that serendipity with feeds that repeatedly recommend variations of something a person once liked. A carefully curated physical community can introduce work that would never have appeared in that person’s usual stream.

The experiment extends beyond conventional papers. Swyx notes that a paper can already be six months old when it reaches the conference floor. Published research remains a valuable artifact, but useful work also appears in blog posts, products, talks and tweets.

The proposed poasters format—with an A—puts those less formal contributions into the same setting. Print a tweet or hot take, place it alongside conventional posters, and stand there to defend it. The mechanism preserves the useful part of a poster session: a concrete claim and an available person who can explain, qualify or debate it.

The leadership program serves CTOs, VPs of AI and others hiring engineers, managing teams and setting up agentic organizations. The preview reserves level three for leadership, described as about 50% of the breakout floor space. Alongside talks from fellow leaders, it introduces a more specific way to bring people with comparable operating concerns together.

The Token Billionaire program takes its cue from Ryan Lopopolo’s AIE Europe talk. Swyx describes approximate qualification as consumption on the order of one billion tokens per month. He also says some users consume 10 trillion tokens per month, without identifying their workloads or accounting method. The lounge is compared to an Amex Platinum lounge: a place where substantial usage creates common ground for a different kind of conversation.

Those conversations need not point in the same direction:

• Spend more intelligently: identify useful work that justifies additional model consumption.
• Spend less: reduce usage while preserving the value the organization needs.

Swyx expects both groups to participate. He credits Alex Volkov of Weights & Biases with a “ZL spectrum” reference after Mario Zechner, using it to describe the contrast between these spending attitudes.

Leadership programming is also moving from broad role-based grouping toward daily themes. The request driving that change is concrete: attendees want to see real workflows from peers who are not trying to sell them something. Topics include increasing useful token consumption, reducing unnecessary usage and building company AI factories.

The preview proposes two routes for off-the-record discussion:

• A networking room: a planned booking system for meetings with speakers and fellow attendees.
• Daily McKinsey sessions: conversations about what is happening inside attendees’ companies and how to approach it.

The announced McKinsey subjects include token maximizing and agentic organizations; a third subject remains unannounced in the recording. These are the preview’s planned arrangements, not a report that the meetings took place.

Previous additions such as design engineering and product management organized the program around broad professional roles. Those remain relevant, but the new emphasis follows AI into particular deployment settings. Partly inspired by Chris Lovejoy, the selected areas are forward deployed engineering, agentic commerce, healthcare, finance and GTM, or go-to-market work.

Finance receives special emphasis through the announcement of AI Engineer New York 2026. Swyx describes it as the third New York conference and the first entirely focused on finance. The current event description is narrower about that exclusivity: finance is the mainstage focus, alongside broader programming.

Drawing on his sell-side trading and buy-side hedge-fund background, swyx forecasts finance as the application area closest to code in its potential for takeoff, followed by healthcare, law, FDE and GTM. This is a thesis about where adoption may accelerate. The vertical tracks create places to examine the distinct work in those settings, rather than treating deployment as one generic subject.

The website demonstration gives a simple route to the surrounding gatherings:

1. Open the World’s Fair website.
2. Select Side events.
3. Browse the listings in the embedded Luma calendar.

At recording time, swyx reports 41 listed side events, with more being added. The listings are intended to help attendees continue conversations outside the main day’s programming.

The highlighted World Cup gathering is described in the preview as a confirmed Team USA event. It is a distributed opportunity, not one gathering for the entire conference: swyx says no single viewing venue can accommodate everyone. Attendees can choose an individual event and watch alongside other AI engineers.

Feedback from AIE Europe prompted the first New Engineer Orientation, or NEO, particularly for people attending on their own. The usual team strategy is to divide the program and regroup: swyx describes 10–12 simultaneous tracks, with teammates each following a different track and sharing what they learned afterward. He compares the process to MapReduce.

NEO supplies a social starting point for people without that ready-made team. Planned for the night before the first day, it welcomes solo newcomers as well as returning attendees willing to meet people and help. Swyx reports 300 orientation signups at recording time and projects that participation could reach 1,000. The projection motivates a request for help; it is not a final attendance count.

A chatbot exists to answer questions, but swyx expects newcomers to ask humans anyway. That is an expectation about onboarding, not measured chatbot adoption. Even at an AI conference, meeting someone who can answer a basic question is part of the reason to attend.

Neo4j is organizing a children’s event where attendees can bring their kids and spark their interest. Swyx explicitly says he does not know its rules, so the mention does not establish eligibility, supervision arrangements or childcare provision. The broader intention is to make participation more welcoming to families.

Personal relationships also count among the community outcomes the organizers value. Swyx mentions couples who met at AIE and announces a dating event in the expo on opening night, despite its absence from the listings shown in the preview. The conference’s social purpose extends beyond recruiting, selling and exchanging technical knowledge.

The closing credit goes to MCs, track hosts and the people who choose to contribute. Swyx connects that collective work to the continuing effort to resist low-quality AI content: a meaningful event depends on people bringing care, judgment and substantive work into the same room.

The final invitation mentions a discount promotion and suggests sponsor credits can offset the ticket price, without supplying a code or redemption conditions. Those are promotional benefits to evaluate on their terms. The invitation itself is straightforward: spend time with the community and leave room for the full program during the week of July 2.