Reconstructed Slides: The emerging skillset of wielding coding agents — Beyang Liu, Sourcegraph / Amp

Source Video

The emerging skillset of wielding coding agents — Beyang Liu, Sourcegraph / Amp

Method

This deck is reconstructed from the existing video frame captures by detecting likely slide regions with OpenCV, cropping/upscaling those regions, deduplicating similar crops, and OCRing the cropped slide images locally. It is a cleaner companion to the full-stage frame deck.

Reconstructed Slides

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: agent_vision.
OCR decision: ready — Multi-column slide with embedded social screenshots and small text.

Slide text:

Are coding agents good or slop?

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.97
Text source: agent_vision.

Slide text:

You are using Cursor AI incorrectly...

- Using Cursor as a replacement for Google Search.

- Underspecification of prompts, not knowing how to drive outcomes and using low-level thinking of "implement XYZ, please".

- Treating Cursor as if it is an IDE, instead of it being an autonomous agent.

- Blissful unawareness of the concept that you can program LLM outcomes.

- Unnecessary usage of pleasantries ("please" and "can you") with it as if it were a human. If it fucks up, swear at it - go all caps and call it a clown. It soothes the soul.

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.99
Text source: agent_vision.

Slide text:

Top mistake:

Using coding agents like you were using AI coding tools 6 months ago

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.99
Text source: agent_vision.

Slide text:

GPT-3 → GPT-4, → Claude 4

Claude 3,

Gemini 2.5

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.98
Text source: agent_vision.

Slide text:

GPT-3 → GPT-4, → Claude 4

Claude 3,

Gemini 2.5

Copilots → RAG bots → Agents

What would a coding agent designed to unleash tool LLMs look like?

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.99
Text source: agent_vision.

Slide text:

Controversial design decisions

- Just make edits - don't ask me

- Minimal UI - no VS Code fork

- No "choose your own model"

- No fixed pricing, no token limit

- Unix philosophy over vertical integration

- Don't build on top of a RAG bot - existing code AI UX is now outdated

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.96
Text source: agent_vision.

Slide text:

RAG context UIs

Agent UI

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.98
Text source: agent_vision.

Slide text:

2 simple clients

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.97
Text source: agent_vision.
OCR decision: ready — Dense docs/code and embedded UI screenshot; OCR is likely useful.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.96
Text source: agent_vision.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.96
Text source: agent_vision.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: agent_vision.
OCR decision: ready — Multi-column code and tool output are small and dense; OCR should do better than manual transcription.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: agent_vision.
OCR decision: ready — Dense assistant output, TODOs, and issue lists are better handled by OCR.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: agent_vision.
OCR decision: ready — Dense issue list and code-like UI; OCR is appropriate.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: agent_vision.
OCR decision: ready — Code-heavy split view with small labels and UI text; OCR is appropriate.

Slide text:

Amp

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.96
Text source: none.
OCR decision: ready — Dense code/editor screenshot with small labels and diff text.
Slide text: not surfaced (none by AI classifier).

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.96
Text source: none.
OCR decision: ready — Dense code/editor screenshot with small labels and diff text.
Slide text: not surfaced (none by AI classifier).

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: none.
OCR decision: ready — Dense code/editor screenshot with explanatory side-panel text and small code diff.
Slide text: not surfaced (none by AI classifier).

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.98
Text source: agent_vision.
OCR decision: ready — Product UI screenshot with multiple labels and dense interface text.

Slide text:

External Connectors

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.98
Text source: agent_vision.
OCR decision: ready — Product modal screenshot with dense form labels and JSON configuration text.

Slide text:

Edit Connection

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.95
Text source: none.
OCR decision: ready — Dense code diff plus explanatory bullet text in side panel.
Slide text: not surfaced (none by AI classifier).

Recreated text/layout view: open HTML recreation
AI slide classifier: content_slide confidence 0.91
Text source: none.
OCR decision: ready — Dense collage of social posts and testimonial screenshots; OCR is better than manual transcription.
Slide text: not surfaced (none by AI classifier).

Hidden Non-Slide Evidence

`slide-001.jpg` — sponsor_logo confidence 0.99; Sponsor/logo wall, not a content slide.
`slide-023.jpg` — demo_video confidence 0.99; Non-slide demo interstitial with only a demo label on a black screen.

Classification audit: raw/sources/slide-ai-classification/reconstructed/F_RyElT_gJk/audit.json