This repository contains comprehensive notes, analysis, and synthesis from the AI Engineer Code Summit 2025 held November 19-22, 2025 in Midtown Manhattan, New York. These notes cover the Leadership Sessions (Nov 20) and Engineering Sessions (Nov 21).
In a spasm of uncharacterist orgnanizatioal enthusiasm, I took it upon myself to capture notes on every talk. It turned out to be an excellent way to keep engaged and locked in.
First I used Comet, went to [https://www.ai.engineer/schedule] and asked it to:
go through each of the schedules talks today and give me a brief bio of who is speaking, what their company does, and what to focus on
I then copied that resulting markdown into my daily Obsidian note, and I started typing furiously. Typing with complete disregard for what keys I was mashing, not even attemption to correct spelling, just capturing whatever stuck out.
I was also snapping photos on my phone.
At some point I thought you know I'm at an AI conference, lets do some image manipulating, so /while I was taking notes/ I also fired up claude code and vibe coded an image processor to extract slides from an image, also looking in the metadata to rename the file as a date with a time stamp.
At the end of each day, I asked claude to intereest the image slides into the right section of the file -- the talks had the time in there -- and that's how we get
Now I started asking claude to figure out how to program itself.
Prompt: conference-analsys.md
This was something like
i want to build a command that helps me make sense of the themes in this talk, ask me questions about what i want one at time until you know what to do.
I use that pattern a lot, one question at a time. I used to say something like "ask me the question that will give you the most important information first" but now it just knows.
Prompt: theme-summary.md
refine the themes even more, making them into a pyrmind, first a summary, then a one pager,and then a presentation
The problem is that prose it produces is so stultifyingly boring, but this was and interesting thought process to fiddling with the information architecture.
This helped figure out which talk really had to do with which theme for example.
Prompt: speaker-decorator.md
This is a fun one that you does a bunch of biographical research on people, what they're known for, their socials etc. I also use it to download their photos. You can run it using something like
go through all of the speakers that we don't have bios for yet and run the speaker-decorator in a subagent 10 at a time
And it will do all the magic it does and then writes out the bios. They generally seem pretty good! I learned a lot about the speakers this way.
For this I used Gemini 3 on aistudio to mock up a few apps.
I was going for a cybernetic, research lab report vibe.
I tried a couple of things on, found one that I liked. Downloaded it, put in in the web folder, and told it to convert the implementation to astro, and go crazy.
Created a nano-banana SKILL.md that I used for the theme photos. And you once you work out a style, you can do things like
ok spin up a subagent for each of the remaining theme files in @2025-11-20/themes/ and @2025-11-21/themes/
Now to actually write everything. It's "written" but the only thing I'm happy with is really the themes and overall structure. Starting with the exective summary, and then the about page.
From "slop" to quality: How 25 sessions revealed the path to thoughtful AI systems
This repository contains comprehensive notes, analysis, and synthesis from the AI Engineering Code Summit held November 20-21, 2025. Across two days and dozens of sessions from leaders at Anthropic, OpenAI, Google DeepMind, Meta, Netflix, and more, a clear architecture emerged: quality AI engineering requires human thinking at the center, environments as universal abstraction, specialized models through RL, capability over scaffolding, and open data as infrastructure.
About: about.md - Conference details, author bio, and repository overview
- Day 1 (Nov 20): Theme Overview - 6 major themes
- Day 2 (Nov 21): Theme Overview - 5 major themes
- Executive Synthesis: Cross-conference insights (run
/executive-synthesisto generate)
Use /theme-summary to generate concise executive summaries from detailed theme analyses:
- 2025-11-20-summary.md - Executive summary (~670 words, ~3 min read)
- 2025-11-20-presentation.md - Slide deck (15 slides, ready for styling with Marp/reveal.js)
- 2025-11-20-one-pager.md - One-pager (~450 words, ~2 min read)
Individual Theme Files:
- Trust, Role Changes, and Infrastructure
- Context, Memory, and The Harness
- SDLC Evolution: Amplified Practices
- ROI Reality Check
- Proactive vs Reactive UX
- Economics of AI Engineering
- 2025-11-21-summary.md - Executive summary (~1,074 words, ~5 min read)
- 2025-11-21-presentation.md - Slide deck (20 slides, ready for styling with Marp/reveal.js)
- 2025-11-21-one-pager.md - One-pager (~528 words, ~2 min read)
Individual Theme Files:
- Context Engineering & Human-AI Collaboration
- Environments as Universal Abstraction
- Reinforcement Learning for Specialized Models
- Model Quality Over Scaffolding
- Data Collection & Quality as Bottleneck
Prepare speaker schedule:
- Copy the conference schedule into a markdown file (e.g.,
2025-11-20.md) - Split into sections by session
- Include time frames for each session (format:
HH:MM am/pm)## 9:00am - 9:20am | Speaker Name, Company - Topic: Session title
Setup photo capture:
- Have your phone/camera ready for slides
- Create a
photos/directory to store raw photos
Take notes with timestamps:
- Write notes in real-time under each session header
- Don't worry about perfect formatting - just capture the key points
- Include timestamps when ideas/quotes happen (optional but helpful)
- Note any particularly important slides
Capture slides:
- Take photos of important slides throughout each session
- Don't worry about organizing - just capture everything
- Photos should include timestamps (most phones do this automatically)
Extract slides from photos:
Use nano-banana-cli to process photos:
# Process all photos and extract slides with timestamps
nano-banana extract-slides photos/ -o slides/This will:
- Extract slide content from photos
- Apply Gemini enhancement for clarity
- Add timestamps to filenames (format:
YYYY-MM-DD-HH-MM-gemini-edited.jpg) - Save to
slides/directory
Run conference analysis:
# For each day, run the conference analysis command
/conference-analysisThis automated workflow will:
- Insert photo references into your notes based on timestamps
- Split notes into individual session files (one per speaker)
- Identify major themes across all sessions
- Perform deep exploration to find additional patterns
- Collaborate with you to refine theme structure
- Launch parallel agents to write detailed analysis for each theme
- Generate a comprehensive conference overview document
The command will ask for:
- Path to your conference notes file
- Initial theme ideas (or let it find them automatically)
- Feedback on theme refinement (merge/split/adjust lead voices)
Output files:
YYYY-MM-DD/MM-DD-HH-MM-speaker-name-company.md- Individual session files with notes and slidesYYYY-MM-DD/MM-DD-day-overview.md- High-level themes from that dayYYYY-MM-DD-themes.md- Detailed analysis at top level (~6,000-7,000 words)
After processing all conference days, create a cross-conference synthesis:
/executive-synthesisThis will:
- Analyze all theme overview documents (YYYY-MM-DD-themes.md files)
- Identify meta-patterns across the entire conference
- Generate executive summary with strategic insights
- Create synthesis document:
executive-synthesis.md
.
├── README.md # This file
├── 2025-11-20.md # Day 1 raw notes (top level)
├── 2025-11-21.md # Day 2 raw notes (top level)
├── 2025-11-20-themes.md # Day 1 analysis (top level)
├── 2025-11-21-themes.md # Day 2 analysis (top level)
├── executive-synthesis.md # Cross-conference synthesis (top level)
├── 2025-11-20/ # Day 1 directory
│ ├── 11-20-day-overview.md # Day 1 theme summary
│ ├── 11-20-HH-MM-speaker-name-company.md # Individual sessions
│ └── themes/ # Individual theme files
│ └── theme-slug.md # Theme with frontmatter
├── 2025-11-21/ # Day 2 directory
│ ├── 11-21-day-overview.md # Day 2 theme summary
│ ├── 11-21-HH-MM-speaker-name-company.md # Individual sessions
│ └── themes/ # Individual theme files
│ └── theme-slug.md # Theme with frontmatter
├── photos/ # Raw photos from conference
├── slides/ # Extracted slides with timestamps
│ └── YYYY-MM-DD-HH-MM-gemini-edited.jpg
└── .claude/
└── commands/
├── conference-analysis.md # Main analysis workflow
└── executive-synthesis.md # Cross-conference synthesis
Organization principle:
- Top level: Source notes (YYYY-MM-DD.md), theme analyses (YYYY-MM-DD-themes.md), executive synthesis
- Day directories (YYYY-MM-DD/): Individual session files and day overview
- Supporting assets: photos/ and slides/ directories
- 2025-11-20-themes.md - Day 1 deep dive (6 themes, ~6,500 words)
- 2025-11-21-themes.md - Day 2 deep dive (5 themes, ~6,500 words)
- executive-synthesis.md - Executive summary across both days (run
/executive-synthesis)
Day 1:
- 2025-11-20-summary.md - Executive summary (~1,150 words, ~3 min read)
- 2025-11-20-presentation.md - Slide deck (15 slides)
- 2025-11-20-one-pager.md - One-pager (~540 words, ~2 min read)
Day 2:
- 2025-11-21-summary.md - Executive summary (~1,074 words, ~5 min read)
- 2025-11-21-presentation.md - Slide deck (20 slides)
- 2025-11-21-one-pager.md - One-pager (~528 words, ~2 min read)
- 2025-11-20.md - Day 1 raw conference notes with timestamps
- 2025-11-21.md - Day 2 raw conference notes with timestamps
Individual session notes are in day directories: YYYY-MM-DD/MM-DD-HH-MM-speaker-name-company.md
Example: 2025-11-21/11-21-09-11-barry-zhang-mahesh-murag-anthropic.md
Each file contains:
- Session time and speaker info
- Raw notes from the talk
- Embedded slide references
- Key quotes and insights
Multi-step workflow to analyze a conference day:
- Inserts slide references based on timestamps
- Splits notes into individual session files
- Identifies and refines themes collaboratively
- Generates comprehensive theme analysis
Creates strategic synthesis across multiple conference days:
- Analyzes all theme overview documents
- Identifies meta-patterns and convergence
- Generates executive-level insights
Generates concise executive summaries from detailed theme analyses:
- Creates executive summary (~1,150 words, ~3 min read)
- Generates presentation slide deck (15 slides, markdown, ready for styling)
- Produces one-pager (~540 words, ~2 min read)
- Focuses on pyramid structure: apex insight → supporting pillars → actionable takeaways
- Optimized for stakeholder sharing and quick executive briefings
Note-taking:
- Include session headers with exact times
- Capture direct quotes (mark with
"...") - Note speaker reactions and audience moments
- Mark particularly important points with
*or**
Photo capture:
- Take photos of all key slides
- Get clear shots (lighting matters!)
- Don't worry about duplicates - nano-banana will handle it
- Capture diagrams, code examples, and data visualizations
Theme refinement:
- The
/conference-analysiscommand is collaborative - engage with it - Suggest mergers when themes overlap
- Identify strong "lead voices" for each theme
- Push for first-class treatment of important patterns
Quality output:
- The "think hard" instruction in analysis agents matters
- Request concrete examples and quotes
- Ask for subsections in complex themes
- Ensure themes interconnect in the final document
- nano-banana-cli - Extract and enhance slides from photos using Gemini
- Claude Code - Multi-agent analysis and synthesis via custom slash commands
- Task agents - Parallel processing for theme analysis
Conference notes and analysis are for internal use and learning.