AI Engineer World's Fair 2026: 6,000 badges, 29 tracks, one follow-up window.
Four floors of expo, 29 tracks, 6,000 practitioners. The badges scan clean. The context behind them is what you actually lose. Here is how to work the room.
The expo floor runs four levels deep
The Marriott Marquis expo hall stacks across multiple floors, and by the second afternoon you have shaken hands with the founder building the eval harness you needed last quarter, two VPs of AI hiring for roles you could staff, and a small team looking for design partners before their Series A closes. Every badge scanned clean. The context behind the badge did not. By the flight home, the person you logged as "the RAG-eval one, sharp, wants a pilot" has gone fuzzy, and the thread you meant to open the next morning still has not opened. Six thousand engineers in one building is a deal venue, not a lecture hall. The deals are made in the hallway and lost in the inbox.
This year's story is consolidation, not novelty
The Fair runs June 29 to July 2 at the Marriott Marquis in San Francisco: 6,000-plus AI engineers, founders, and VPs of AI across roughly 29 tracks and about 100 expo partners. The headline this cycle is not a new model drop. It is the shakeout. Agent platforms that raised on a 2024 thesis are now being judged on whether anything shipped, evals tooling is collapsing from a crowded field toward a handful of names people actually run in CI, and the open-source-versus-frontier-API argument has turned from religion into a cost-and-control spreadsheet. Read that way, the expo floor is a map of who is about to get acquired, who is hiring aggressively, and who is quietly looking for their next round. The people worth your three days are standing next to those signals.
Who to find, and the question to actually ask
Skip "how do you see the future of agents." Everyone here has answered it forty times. The keynote and track lineup tells you who set the agenda; the expo tells you who is buying it. On the agent-platform booths, ask what they cut from the roadmap this year, not what they added. To the eval-tooling teams, ask which metric their design partners stopped trusting. To anyone hiring, ask what role has been open longest and why, because that is where you or someone in your network fits. Each of those questions returns something specific you can act on. None of them survives a week in your head without being written down with the name and the moment attached.
What nobody on the schedule will say out loud
The quiet part: a practitioner-dense, badge-heavy crowd is the worst possible place for follow-up discipline, and everyone knows it. The density that makes the Fair valuable is the same density that buries every promising conversation under thirty more by dinner. The people who close from this event are not the ones with the most scans. They are the ones who can still tell you, two weeks later, why a specific person mattered and what they promised to send. Conference ROI is decided after you leave the building, in a follow-up window most attendees miss because the context evaporated before the calendar invite went out.
Capture the room while you are still in it
The fix is not a better notes app you open later. It is capturing the person and the moment in the same breath as the handshake, so "wants a pilot, mentioned their eval metric broke in March" is attached to the face before the next conversation overwrites it. Met's Event Mode is built for exactly this density: scan the card, hold the room, keep moving, and let the context stay pinned to the contact instead of your memory. Storage stays in your own iCloud, so a list of who you met at the Fair never becomes someone else's data set. The badge scan is the easy part. The thread you can still open in July is the deal.
Get Met before you pack for San Francisco, and walk the Fair knowing every conversation comes home with its context intact.
Met is built for the operator who leaves a conference with a hundred half-remembered conversations and needs every one of them to still mean something next week. Capture is fast enough to keep up with a packed expo floor; context stays pinned to the contact, not your memory.
