---
title: "Context Engineering in 2026: Compaction, Memory & Cost"
category: "talks"
date: "2026-06-29"
time: "2:20pm-4:20pm"
track: "Track 6"
room: "Track 6"
speakers: ["Louis-François Bouchard", "Samridhi Vaid", "Omar Solano"]
sourceLabels: ["Official conference schedule", "Public YouTube metadata"]
scheduleTrack: ""
scheduleRoom: "Track 6"
scheduleLabels: ["Track 6", "sponsor", "confirmed"]
---
# Context Engineering in 2026: Compaction, Memory & Cost

## Conference Context
- Date/time: 2026-06-29 · 2:20pm-4:20pm
- Track/room: track TBD · Track 6
- Speaker(s): Louis-François Bouchard, Samridhi Vaid, Omar Solano
- Session type/status: sponsor · confirmed

- Track: track TBD
- Room: Track 6
- Session type: sponsor
- Status: confirmed

## Session Description
Every long agent session eventually breaks: the assistant that swore it would "never push to main" does exactly that forty turns later. The model didn't get dumber — its context did. This workshop is about engineering the context window so that stops happening, shown with Towards AI's open-source AI tutor, which answers questions for students of our AI-engineering courses. Context engineering is deciding what the model sees on every single call — instructions, history, retrieved course content, memory, and tool outputs — and it's the line between a tutor that holds a coherent session and one that forgets the student's setup halfway through. We'll move in three stages, mirroring how the project actually went. The concepts: the two root problems (a finite window, a stateless model), the full compaction toolkit (truncation, trimming, tool-result clearing, summarization, and offloading to files — and when each actually helps), memory that survives across sessions, skills loaded on demand, and production-grade retrieval (chunking, metadata, course scoping, hybrid search, reranking, and evaluating). We'll cover the tutor's architecture, and the evaluation harness we used to measure every run on Gemini — tokens, cost, latency, and memory probes instead of vibe-checks. At real volume, even Gemini Flash got expensive, so we tested whether open and local models could match the quality for a fraction of the cost and match result quality. Everything is open-source and will be shared during the workshop.

## Media Evidence
[Turn 10,994 Notes Into Memory - Paul Iusztin, Decoding AI & Louis-François Bouchard, Towards AI](https://www.youtube.com/watch?v=ZRM_TfEZcIo) (speaker-match related prior/adjacent AI Engineer video; captions: English auto-captions).

- [[youtube-I2cbIws9j10-transcript]] — full cached transcript markdown for the related YouTube source.

- Source video: `youtube-I2cbIws9j10`
- Slide deck: [[youtube-I2cbIws9j10-dense-slides|Dense Slides: WF26: Harness Engineering & Startup Battlefield ft. Garry Tan, Mike Krieger, @t3dotgg , DSPy]] — 11 visible slide image(s); 11 HTML recreation(s).
![[assets/dense-slides/I2cbIws9j10/slide-001.jpg]]
![[assets/dense-slides/I2cbIws9j10/slide-002.jpg]]
![[assets/dense-slides/I2cbIws9j10/slide-003.jpg]]
- Additional slide evidence: [[youtube-I2cbIws9j10-slides|Slides: WF26: Harness Engineering & Startup Battlefield ft. Garry Tan, Mike Krieger, @t3dotgg , DSPy]]
- Slide-derived themes for `youtube-I2cbIws9j10`: context, window, selects, response, facts, retry, coerce, rollback.
- Source video: `youtube-ZRM_TfEZcIo`
- Slide deck: [[youtube-ZRM_TfEZcIo-dense-slides|Dense Slides: Turn 10,994 Notes Into Memory - Paul Iusztin, Decoding AI & Louis-François Bouchard, Towards AI]] — 10 visible slide image(s); 10 HTML recreation(s).
![[assets/dense-slides/ZRM_TfEZcIo/slide-001.jpg]]
![[assets/dense-slides/ZRM_TfEZcIo/slide-004.jpg]]
![[assets/dense-slides/ZRM_TfEZcIo/slide-005.jpg]]
- Additional slide evidence: [[youtube-ZRM_TfEZcIo-slides|Slides: Turn 10,994 Notes Into Memory - Paul Iusztin, Decoding AI & Louis-François Bouchard, Towards AI]], [[youtube-ZRM_TfEZcIo-reconstructed-slides|Reconstructed Slides: Turn 10,994 Notes Into Memory - Paul Iusztin, Decoding AI & Louis-François Bouchard, Towards AI]]
- Slide-derived themes for `youtube-ZRM_TfEZcIo`: obsidian, google, plus, notion, drive, growing, files, month.

## Evidence Graph
This evidence graph is generated from currently linked source material: official schedule text, related video pages, cached transcripts, visible slide text, dense/reconstructed slide pages, and AI slide-classification audits.

### Media Signals
- `youtube-I2cbIws9j10` — 91,792 transcript words; 7 slide-derived text signals
- Transcript signals for `youtube-I2cbIws9j10`: code, model, back, system, well, first, today, even.
- Slide-derived themes for `youtube-I2cbIws9j10`: context, window, selects, response, facts, retry, coerce, rollback.
- Evidence links for `youtube-I2cbIws9j10`: [[youtube-I2cbIws9j10]], [[youtube-I2cbIws9j10-transcript]], [[youtube-I2cbIws9j10-slides]], [[youtube-I2cbIws9j10-dense-slides]]
- `youtube-ZRM_TfEZcIo` — 9 slide-derived text signals
- Slide-derived themes for `youtube-ZRM_TfEZcIo`: obsidian, google, plus, notion, drive, growing, files, month.
- Evidence links for `youtube-ZRM_TfEZcIo`: [[youtube-ZRM_TfEZcIo]], [[youtube-ZRM_TfEZcIo-slides]], [[youtube-ZRM_TfEZcIo-dense-slides]], [[youtube-ZRM_TfEZcIo-reconstructed-slides]]

### Agent Reading Notes
Use these signals to refine the synopsis, topic links, people/company context, and method notes. If a source is a related external video rather than an exact official recording, keep it framed as supporting evidence.

## Transcript Status
Related video transcript availability: English auto-captions. Treat this as supporting context, not a recording of this exact scheduled session unless later confirmed. Not fetched yet.

## People
- [[louis-fran-ois-bouchard]]
- [[samridhi-vaid]]
- [[omar-solano]]

## Supporting Slides
- [[youtube-ZRM_TfEZcIo-slides]] — extracted from the related public AI Engineer video.

## Slide Evidence
- Slide-only cropped deck: [[youtube-ZRM_TfEZcIo-dense-slides]] (12 viable slide images).
- Related slide/OCR pages:
- [[youtube-ZRM_TfEZcIo-dense-slides]]
- [[youtube-ZRM_TfEZcIo-reconstructed-slides]]
- [[youtube-ZRM_TfEZcIo-slides]]
- Slide-derived terms: `notes`, `obsidian`, `research`, `towards`, `index`, `every`, `database`, `files`, `engineer`, `handbook`, `content`, `courses`, `videos`, `starts`, `zero`, `codex`, `repos`, `course`

## Livestream Segment
- [Watch in livestream at 02:11:16](https://www.youtube.com/watch?v=I2cbIws9j10&t=7876s) — WF26: Harness Engineering & Startup Battlefield (Day 3).
- Match basis: speaker and title; timed captions matched Louis-François Bouchard, engineering.
- Confidence: high automated match; prefer a dedicated cut-video recording when one exists.

## Attendance Visibility
No high-confidence attendance icon signal is shown for this talk. The sampled video evidence was either low confidence, source-proxy-only, or did not expose a clear audience view.

## Synthesis
### Synthesized Breakdown
Mhm. Mhm. Mhm. Ladies and gentlemen, welcome to the AI Engineer World's Fair.

### Speaker And Company Context
- [[louis-fran-ois-bouchard|Louis-François Bouchard]] — CTO & Co-Founder at [[towards-ai|Towards AI]].
- [[samridhi-vaid|Samridhi Vaid]] — Senior Machine Learning Engineer at [[towards-ai|Towards AI]].
- [[omar-solano|Omar Solano]] — AI Engineer at [[towards-ai|Towards AI]].

### Topics Covered
- [[agent-security]]
- [[agentic-search]]
- [[agentic-web]]
- [[ai-sandboxes]]
- [[coding-agents]]
- [[mcp]]

### Derived Links And Source Material
- [[youtube-I2cbIws9j10-transcript]] — transcript markdown; source cache `raw/sources/youtube-livestream-transcripts/I2cbIws9j10.txt` (91,792 words).
- [[youtube-I2cbIws9j10]] — related YouTube source page.
- [[youtube-I2cbIws9j10-slides]] — slide evidence.
- [[youtube-I2cbIws9j10-dense-slides]] — slide evidence.
- [[youtube-ZRM_TfEZcIo]] — related YouTube source page.
- [[youtube-ZRM_TfEZcIo-slides]] — slide evidence.
- [[youtube-ZRM_TfEZcIo-reconstructed-slides]] — slide evidence.
- [[youtube-ZRM_TfEZcIo-dense-slides]] — slide evidence.

### Novel Concepts / Clever Methods
- [[agent-ready-accessibility|Agent-Ready Accessibility]] — Designing for agents and designing for accessibility converge around explicit structure, reachable controls, and understandable state.

### Evidence Boundary
This synthesis uses the official schedule plus cached video transcripts. Official AI Engineer World's Fair San Francisco 2026 livestreams and cut videos are primary event video sources for transcript/slide evidence; external, historical, or speaker-matched videos remain supporting context unless manually verified as exact official event recordings.