About these notes
This unofficial site covers two main-stage livestreams from AI Engineer NYC 2026: October 13 and October 14. Day 1 and Day 2 here label those two recording dates. The coverage is the Simons Room feed, including recorded talks played during breaks. It does not include every talk in the other rooms.
The engineering field guide compares approaches across both days. Each topic guide explains the evidence limits and proposes a small experiment. Each session summary includes key lessons and a timestamped transcript. Use the day filter to narrow the list, or start with the Day 1 digest and Day 2 digest.
Source recordings
Section titled “Source recordings”| Day | Date | Recording | Video start, Eastern Time |
|---|---|---|---|
| Day 1 | October 13, 2026 | Day 1 on YouTube | 8:47:52 AM |
| Day 2 | October 14, 2026 | Day 2 on YouTube | 8:40:36 AM |
The event schedule describes the planned program. The agenda and stream notes explain recording gaps, timing changes, and talks that aired during scheduled breaks.
How to use the notes
Section titled “How to use the notes”Start with a topic on the home page to compare approaches across days. Open a session for our notes and lessons. Use its source link when you need more context or want to check a claim. Matching Day 1 sessions link to the official reading editions, which include slides and searchable transcripts.
Numbers and results are the speakers’ reported findings. Some describe one experiment, a specific workload, or a future plan. The session notes preserve those limits. The field guide is our editorial synthesis of the two recordings. Comparisons and suggested experiments are our interpretation, rather than procedures prescribed by the speakers.
We checked speaker attribution across the official October 13 library and reviewed selected numerical claims. This corrected the Railway and Flywire speaker names. Vals’ $1.5 million token figure describes the reported value of free access. Railway’s “700 million tokens” line is an aside. GPU security results describe separate experiments, not a general success rate. The original machine transcript text remains available in the session notes.
The Day 2 catalog contains two records of the same Intuit Credit Karma talk. The session browser lists that talk once. Both source records remain available. Announcements, breaks, and alternate records stay in the stream archives.
Topic analysis
Section titled “Topic analysis”The topic explorer compares 24 term families across 47 distinct talks. It shows coverage, normalized rates for each day, source passages, and related concepts. The separate Day 1 question period is counted within its keynote. This explains the difference from the 48 summaries in the session browser.
Term matches are reading aids. They do not establish importance or agreement. Eight excerpts have reviewed labels and explicit limits. The explorer includes the matching rules, recording limits, and downloadable data.
How timestamps work
Section titled “How timestamps work”Each recording starts its video clock at zero. A timestamp includes the video
position and the wall-clock time, for example [01:23:45 | 10:11:37 AM ET].
ET means Eastern Time. Each video link opens the correct day’s recording at the
specified position.
How the notes were produced
Section titled “How the notes were produced”- Download the source audio and split it into chunks.
- Transcribe the chunks and align the text with timestamps.
- Map the transcript to the event program and verify the session boundaries.
- Write session notes, enrich speaker records, and audit the references.
- Synthesize lessons across both days and link each lesson to its source sessions.
Attribution
Section titled “Attribution”The talks belong to their speakers and the event organizers. This site provides unofficial reference notes and links back to their public recordings.