Dense Slides: Katelyn Lesse – Evolving Claude APIs for Agents, Anthropic
Source Video
Katelyn Lesse – Evolving Claude APIs for Agents, Anthropic
Method
This deck is slide-only. The existing captured video frame set supplies candidate frames, then local OpenCV rejects sponsor/title/speaker-only frames, crops visible slide surfaces, deduplicates, and saves the cropped slide images.
Cropped Visible Slides

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
Slide text:
How the Claude Developer Platform maximizes intelligence
Harness Claude's capabilities
Manage Claude's context
Give Claude a computer & let it cook

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: advanced OCR
rapidocr-live/bright-screen/contrast. - OCR decision: ready — Contains a code screenshot with small text; OCR should handle it better than direct transcription.
Slide text:
response model=°claude-sonnet-4-5-292S8929 ltent.tessoges.
Thinking Extended max tokens= thtnktng=f:.suoyo)"lobpnq. enabled":True,
messagcsa
Content": *Help re debug why this systen fails under load" User
ANTHROPIC
AIE / LEAD EVOLVING CLAUDE APISFOR AGENTS
Google DeepMind PRESENTEDBY KATELYN LESSE/ Engineering Leader ANTHROPIC

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: advanced OCR
rapidocr-live/bright-screen/contrast. - OCR decision: ready — Dense multi-column diagram text is better suited to OCR than manual transcription.
Slide text:
Context engineering for agents
Possible context to give model Contextwindow
Doc Doo System prompt
Tool Tool Memory file Tool Too! Memory file Doc Doc2 messoge Assistant
instructions Comprehensive Tool1 Tool2 Toolcall
Domain knowledge Usermessoge
Memory file Doc Metsoge histon
Tool
Messoge history
Toolresult
ANTHROP\C
AIE/ LEAD Brian Fioca OpenAl Samir Mody NLW Superintelligent Stanford YegorDenisov-Blanch University DX Justin Reock I:: T Arman
Google DeepMind PRESENTEDBY stigg Weights & Biases byCoreWeave SENTRY Daytona ^arize Xgoose

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: advanced OCR
rapidocr-live/bright-screen/opencv-adaptive. - OCR decision: ready — Contains a code screenshot with small text; OCR should extract it more reliably than manual reading.
Slide text:
MCP response clentit:arltr:.rn0 node1- ttaude sonne1 4-5 2025e929 0 ncp servers=! max tokens? (ur: +pcD/github-server. (ur! a ocp//sentry scrver. 0 numo 'sentry) name o 'glthub+
messages=[0 (oyids 4subien @ 6snt) foidhp s1ul oip Am. 6 du3juo2. role o user'0
ANTHROPIC
AIE/LEAD eligent Stanford Yegor Denisov-Blanch University DX Justin Reock Tenex Arman Hezarkhani Lei Zhang Bloomberg Gene Kim REYOLUTIC
Google DeepMind Presented eY ne04j autho Workos *sourcegraph stigg::: Weights & Biases by CoreWeiv

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: advanced OCR
rapidocr-live/full. - OCR decision: ready — Contains a code screenshot with small text; OCR is appropriate.
Slide text:
Memory response =client.messages.create( max_tokens=4e96, messages=[{ model="claude-sonnet-4-5-20250929"
prefers circuit breaker pattern logic Ln Paymentcltent.java SAARS "Paymentservtce uses 3os tus
ANTHROP\C
AlEngineer Engineering the future of Al
Classification audit: raw/sources/slide-ai-classification/dense/aqW68Is_Kj4/audit.json