Most AI coding demos stop at generated files.
DesignSetGo Apps is the next step I wanted for WordPress: upload a static HTML, CSS, and JS app bundle, run it in a sandboxed iframe, and connect it to WordPress through a permissioned bridge.
The free version is live here:
https://t.co/zZLiRnTQqP
@techiesreviews The maintainability test matters more than the first render. I want to see clean block structure, predictable CSS, no locked-in generated soup, and a path to keep editing after the AI pass. If the client cannot safely change it next month, it was only a demo.
@0xJame_eth This direction makes a lot of sense. The tricky part is making the agent powerful without giving it unlimited site access. For WordPress I keep coming back to scoped actions, human-readable logs, and very boring rollback.
Plugin #25 is live and this one is what the plugin management experience should look like in #WordPress.
๐ช Bulk upload plugins, bulk toggle de/activations, download plugins and bundles, and a lot more!
https://t.co/LaMA26D6UH
โCreate a page for my portfolio.โ
Thatโs all it takes to update your website with Airo for WordPress.
Add new pages, create new sections, and refine your content โ all by chatting.
No blank pages. No writer's block.
See it in action ๐
https://t.co/h3CHfOvnSu
I keep thinking about the gap between AI generating a working mini app and WordPress actually having a safe place to run it.
Static builds are not the problem by themselves. The missing pieces are isolation, permissions, versioning, and a way for the site owner to know what is running.
Claude Code is much better at WordPress when it can see the boring parts.
Plugin structure, hooks, REST endpoints, theme files, wp-env, tests, the exact error.
The weaker workflow is asking for a beautiful site. The stronger workflow is giving it one small job and forcing it to prove the change.
The more I build around WordPress and AI, the more convinced I am that the agent should never need god mode.
Give it a small capability, a clear user context, an audit trail, and an obvious rollback path.
That is how this moves from fun demos to tools site owners can actually trust.
@leoobai@Saboo_Shubham_ This is the part that separates demos from systems. Fast prototype loops are great, but production agents need evidence: tests, failure logs, permission boundaries, and a boring path for rollback when the agent does the wrong thing confidently.
@tmdxmeme Yep. Vibe coding amplifies taste and structure. If the product thinking is fuzzy, the output just gets fuzzy faster. The strongest builders I see still do the hard parts themselves: scope, constraints, user flow, failure cases, and what good enough means.
@webstudio55@WordPress Logs are underrated here. Once AI calls move through a shared WordPress layer, site owners need to see which plugin initiated the request, which user context it ran under, what provider and model was used, and what it cost. Key storage is only the first trust problem.
Best results are not magic direct WordPress access. I treat Claude Code like a local dev: give it a wp-env or LocalWP repo, WP-CLI where useful, plugin and theme conventions, and a small testable task. It is strongest on hooks, CPTs, REST endpoints, blocks, and boring integration work.
I built a thing. And I actually shipped it. :)
I'd be honored if you checked it out. I'll also suggest registering as a user (free of course) and exploring the features and user experience.
More on the "why" and my "getting back to building again" journey to come.
๐ Congrats to @Justinnealey for winning the audience vote in our live-stream! His infinite canvas block nailed the "Unbound" theme. You can check out Justin's demo of Roam here: https://t.co/zb1UeaeLoQ
Thanks again to everyone who came out and participated! See you next month!
Visual feedback is the wall.
Agents can generate interfaces quickly, but if the loop cannot see the actual screen, layout bugs become expensive.
Browser checks and screenshots should be first-class in vibe coding workflows.
https://t.co/lGrrQtKoRb
it's very fun vibe coding on mobile with claude code but the lack of ability to visually debug what you're building quickly becomes the limiting factor.
what i want is an agent that knows my taste well enough to make design calls for me without me needing to see every screen. we're almost there.
Vibe coding gets dramatically better when the system around it is boring.
Types, tests, screenshots, deploy previews, permission boundaries, rollback.
The prompt is the fun part. The harness is what lets you trust the output.
https://t.co/xIj6mirbbo
AI vibe coding is broken
the best engineers in 2026 won't just prompt.. they'll build stricter systems around AI
types. tests. patterns. static analysis. ci. guardrails
new talk is live: https://t.co/4vH7gsRn7Z
@lysechko The plain English part is useful, but the setup guide is doing a lot of the real work. WordPress plugin output gets much better when the agent has local WordPress conventions, security rules, and a small testable scope instead of a blank prompt.
@h_guedea@WordPress That is the interesting spot for legacy WordPress projects. The AI layer is not just a shiny admin feature if it reduces custom provider wiring and gives older sites a standard way to expose capabilities. Still needs careful staging, but the direction is useful.