Datasette Agent adds split-panel chat for live database app iteration
Source published ·Modelwire updated
Original coverage: Simon Willison ↗·How Modelwire adds context
The development
Datasette Agent 0.5a0 introduces a split-panel interface that positions an LLM chat alongside live database views, enabling real-time application modification through natural language. The architecture bridges agent reasoning with datasette-apps, allowing users to iteratively prompt database-backed interfaces into existence and refine them conversationally. This represents a meaningful shift in how agentic systems interact with stateful data layers, moving beyond isolated chat toward embedded, context-aware tool use where the agent's outputs directly reshape the UI it observes.
Modelwire’s AI-generated summary of coverage from Simon Willison.
Modelwire analysis
Analyst takeOur AI-generated reading of the wider context and the next developments to watch.
The split-panel interface is less about UI novelty and more about acknowledging a hard constraint: agents need to see their own outputs in real time to reason effectively about what they've changed. This surfaces a design lesson from production agentic systems that most chat interfaces still ignore.
Willison's release cadence across datasette-llm (late October fixes), datasette-apps (November 4th integration), and now Agent 0.5a0 shows a deliberate stacking strategy. The earlier datasette-llm 0.1a10 fix (model filtering) and the 0.1a9 configuration inheritance work were load-bearing: they removed friction that would have broken Agent adoption. This mirrors the architectural lesson from the travel-booking agent study (late October), which found that stateless context-per-turn approaches collapse to 46% correctness while deployed systems maintaining server-side state hit 92%. Willison's split-panel design embeds that insight: the agent sees the live database state it modified, not a stale summary.
If datasette-apps usage metrics show that conversational refinement cycles (prompt, observe, re-prompt) complete in fewer turns than traditional SQL-writing workflows, that validates the state-visibility hypothesis. If they don't, the UI is cosmetic. Watch Willison's agent.datasette.io logs over the next 60 days for turn-count distributions on successful vs. abandoned tasks.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·arXiv cs.CL
Deployed agent outperforms stateless baseline by 46 points on structured data delivery
A production travel-booking agent reveals a critical architectural vulnerability in how multi-turn LLM systems handle structured data. Researchers compared two delivery patterns for tool outputs: one that re-emits full record sets each turn (SYNTHESIS) versus one that maintains state server-side and filters deterministically (BYPASS, the deployed system). Across 339 scored turns, the stateless approach collapsed…
MentionsDatasette Agent · Simon Willison · datasette-apps · Datasette
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. Simon Willison originally reported this story as “datasette-agent 0.5a0”. The full content lives on simonwillison.net. If you’re a publisher and want a different summarization policy for your work, see our takedown page.