TLDRocket
Sign in

datasette-apps 0.2a0

Simon Willison's Weblog Simon Willison Covered by 2 sources

Datasette-apps 0.2a0 landed, giving its AI agent new tools to test and manage apps.

Based on reporting by Simon Willison's Weblog, Simon Willison — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Simon Willison just pushed out a small but telling update to datasette-apps, the plugin that lets Datasette's built-in agent build little web apps on the fly. Version 0.2a0 doesn't reinvent anything, but it quietly closes a gap that's been nagging at anyone who's watched an AI agent write code it can't actually check.

The headline addition is app_debug(), a tool that lets the agent open an app it just created and poke at it with JavaScript, without a human ever seeing it happen. The trick is almost comically simple: the app loads inside an iframe set to opacity: 0 with pointer-events: none, so it's invisible and untouchable, yet fully alive underneath. The agent can then run scripts inside that hidden sandbox, checking that buttons fire, forms submit, or that an element renders at the size it's supposed to. It's the kind of self-testing loop that used to require a human eyeballing a screenshot.

Alongside that, there's app_list(), a much more mundane but equally necessary tool that lets the agent see which apps the current user actually has permission to edit. Without it, an agent trying to iterate on existing work would be flying blind about what it's even allowed to touch.

The debugging trick leans on a new context.browser_task() mechanism that shipped in datasette-agent 0.4a0, meaning this release is really two projects moving in lockstep. Willison tagged the post with iframes, datasette, and datasette-apps, which about sums up the scope: a narrow, practical fix aimed at making agent-built software slightly less of a black box.

My take — AI-written commentary, not fact-checked reporting

I like this one a lot, honestly. Giving an agent a hidden iframe to smoke-test its own output is a cheap, clever way to add a feedback loop that most 'AI builds an app' demos skip entirely. It's a small feature, but it's the kind of unglamorous plumbing that separates toy agent demos from tools people will actually trust to edit real apps.

Read more about this at: Simon Willison's Weblog

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.