Agentic AI Summit Notes | From ChatBI to AI Canvas at AskTable (with slides)

Event photo

At the Agentic AI Beijing summit, AskTable presented “Immersive Vibe Analyzing—new interaction for AI-driven analytics.”
Below is a concise recap.

Event photo

Key ideas

Classic ChatBI (one chat, one answer) is hitting a wall. AskTable believes analytics should be about reconstructing the thinking process, not only Q&A. With AI Engine and AI Canvas, we move AI from a dumb query box toward agents that understand business logic.

Why classic ChatBI falls short

Three recurring limits:

1. Fragile answers (hallucinations)

When models invent numbers or logic, enterprise decisions cannot rely on the output.

2. Black-box process

Users see a chart but not SQL, data lineage, or reasoning—hard to trust.

3. Weak on complex work

Multi-table joins, changing metrics, and long analyses do not fit a single chat thread.

AskTable design philosophy

Underlying principle: generate instructions, not hand-waved content.

1. Accuracy: content → instructions (Text-to-SQL)

•AI’s job is to emit SQL executed by your database for ground-truth results.
•Reduces numeric hallucinations and keeps answers reproducible.
•Generated SQL can be tested for semantic equivalence.

2. Interaction: chat → AI Canvas (Vibe Analyzing)

•Infinite canvas for extract → clean → visualize—not one tiny thread.
•Transparent steps: every SQL or Python step is visible and editable.
•Collaboration: analysis paths become templates from creators to viewers.

3. Understanding: semantic layer

•AI must speak business language, not only column names.
•Metadata, metrics, and relationships give models the right context.
•Users ask “last year’s repurchase rate” and the system maps it to real tables.

Product map

Module	Role	Best for
AskTable AI Engine	Text-to-SQL platform, 20+ databases, API/SDK	Builders embedding data agents in FastGPT, Dify, etc.
AskTable AI Canvas	Infinite canvas, smart links, Vibe Analyzing	Analysts exploring data and building dashboards