Hi everyone.
This is the June community update. You can read May’s update here.
As our engineering team makes our codebase safer and easier for AI Agents to operate in, we’ll be shipping lots of cool features and long-standing requests, particularly from many of the folks most active here in this forum. Tomorrow, Emmanuel and I are going live to demo the first batch. Super excited to show what we’ve been working on. Come with questions! You can RSVP here.
Before I get into what else we shipped, a quick shout-out. @petter — the author of the Bubble Manual, the instructor of our new security course on the Academy, and the person who literally wrote the book(s) on Bubble — is about to launch his Substack newsletter. He’s a great resource on all things Bubble!
Now let’s get into the updates.
What we shipped in May
Phase 1 of native JSON support on Bubble is out. This update helps auto-parse JSON in LLM responses, which previously required complex workarounds. You can read more on the forum here.
We upgraded the Agent’s model to Claude Opus 4.8 in tandem with its release last week and are seeing up to 20% faster generation times.
Compound edits are live in the Agent for everyone, meaning the Agent makes several changes in a row that it can assume it needs to take vs. needing to do one step at a time. This should help builders of all levels move a lot faster.
The Agent can now create and edit option sets, attributes, and values in a single prompt, and you can ask it to create multiple sets. You can also now undo, redo, and revive Agent edits within the Agent instead of having to exit the conversation. This means you don’t have to leave the chat and allows you to more actively see what’s being created or undone as it builds. The revive action is particularly useful: If you dismiss an edit plan, you can restore it from chat instead of re-prompting from scratch.
We’ve gotten a fair amount of feedback about the Agent saying it did something that didn’t actually happen, and we’ve been rolling out a series of targeted fixes to the Agent over the past few weeks: Property changes now apply consistently, and UI components render in the intended order. We’re also releasing improvements to help the Agent check its work before it reaches your apps. That includes giving the Agent access to app screenshots after each edit so it can visually check the UI it builds. That’s out for web apps now, and we’re working on mobile next. It also includes validating the generated JSON to check expressions, data, and option sets and reduce silent errors. This should help with Agent reliability, which the team is working on continuously.
Lastly on the Agent side of things, you can now drag and drop images when showing the Agent an example of what you want to build (a small but hopefully impactful follow-on to the feature we launched last month).
Workflow functionality got two new updates: The first is the ability to disable specific actions within a workflow without deleting them. This is especially useful for testing and debugging. We also added a new Generate with AI option in the workflow dropdown, which allows you to describe what a workflow should do and have the Agent build it. You can read more about those updates on the forum here.
The team shipped a batch of quality improvements to mobile app generation, including using style variables and lowering latency. Mobile apps now generate fast with an upgraded look and feel. There’s also a new workflow action to programmatically dismiss the keyboard on iOS, and we added a data source that tells you whether the keyboard is currently open. iOS doesn’t hide the keyboard automatically, so this was a real friction point for forms and multiline input flows.
Finally, we doubled the cap on API Connector calls from 2.5 minutes to 5 minutes, which is a highly requested feature. This reduces timeouts on long-running LLM calls and other slow external services, although the total max length for a workflow run is still 5 minutes.
For a full list of releases and bug fixes, you can read this sprint’s release notes here: bubble.io/release-notes.
One quick note: This section does not include the features we’ll be announcing at Wednesday’s livestream, so expect more updates soon! We’ll be sharing a recap with the livestream on the forum if you can’t make it.
What we’re working on
-
Expanding Agent access to non-AI apps: We’ve been testing the Agent in existing apps with BAMs and Gold and Silver agency partners, and we’re now working on extending access to existing apps. We’re starting in small groups and will be expanding from there.
-
Agent response UX improvements (Phase 2): Earlier this year we improved the UX for edit requests in the AI Agent. Phase 2 focuses on improvements to non-edit responses — cleaner design primitives in Agent output, Bubble documentation references surfaced in-context, and a more native way to engage with what the agent shows you. Design is complete, and engineering work is underway and on track for this month.
-
Mobile plugin editor beta: The mobile plugin editor is headed to public beta! Developers will be able to build mobile plugins using any React Native library they need. It’s a major unlock for the plugin developer community, and by extension for everyone building for native mobile. You can expect that very soon.
-
Mobile web preview: Outside of general performance improvements, the top priority for mobile is to provide a more accurate web preview experience, especially as more users begin to use mobile plugins. That work is ongoing.
-
Property editor: We’re continuing to make improvements to the property editor redesign while it’s in opt-in beta, and we’ll be rolling those out on a continuous basis.
-
Prompt AI workflow action: The team is working on a simplified AI integration step via a new workflow action. This would cut out the API Connector entirely for users who want a workflow that is essentially “I want my app to talk to an LLM.” We’re aiming for the end of the month for that.
How to get involved
-
One more reminder that Emmanuel and I are going live tomorrow with a session specifically for power users, where we’ll demo some of the new releases. If you want a first look (and a chance to put us on the spot with your questions), RSVP here.
-
We launched V2 of Bubble Is Global yesterday — the new map shows app launches around the globe in near-real time. Check it out here and read more on the blog here.
-
Eric Ries, author of The Lean Startup, joined Emmanuel for The New Build right ahead of his new book Incorruptible. It digs into why successful startups go off the rails and it’s a great listen for anyone thinking about how good companies lose their way. You can find it here.
-
Clément Llehi also joined Abhinav for the podcast. He went from cleaning offices himself to running an enterprise-level cleaning company, Makko, and the way he’s built and scaled it looks nothing like the typical enterprise playbook. Check it out here.
-
We just dropped several new sections in our security course, taught by Petter, which you can take here.
-
Emmanuel hosted his AMA on May 21. The replay and recap are here.
-
A few team members will be at the CreateWith conference in Brighton, UK on June 25. Come say hi in person! RSVP here.
That’s all for this month!
— Josh and Emmanuel
