The Four Google AI Tools Worth Your Time

Antigravity and Stitch got the threads. They are both real and both good, and they are also the two that photograph well: an agent-first IDE and a tool that turns a prompt into a UI are easy to screenshot and easy to argue about.

The four below are harder to make a thread out of, because what they do happens while you are not watching. An agent clearing a pull request backlog overnight has no screenshot. That is the entire reason this list exists.

• Every tool here is live right now, in a browser, no waitlist
• All four are Google Labs or Google products, not wrappers
• None of them needs a card to try

One honesty note before the list. "Fifteen" is the round number going around for the month. It is roughly right and it is not an official Google count. Treat it as "a lot, more than the two you have heard of".

Jules is an autonomous coding agent running on Gemini. You connect it to a GitHub repository and it works asynchronously: it picks up tasks, fixes bugs, writes tests, updates dependencies and opens pull requests, without you sitting in a chat window steering it.

The distinction that matters is asynchronous. A coding assistant waits for you. Jules is given a job and goes away and does it, and you review the pull request afterwards like you would a colleague's.

1. 1
   Connect Jules to a GitHub repo you own. Start with one that is not business critical.
2. 2
   Give it something small, bounded and boring. Dependency bumps and missing test coverage are the ideal first job, because you can tell instantly whether the result is right.
3. 3
   Let it run and come back to the pull request. Read the diff, do not skim it.
4. 4
   Only widen its scope once you have merged two or three of its PRs without edits.

Treat the first week as an interview, not an install. You are finding out what it is good at, and the cheapest way to learn that is on work you could redo yourself in ten minutes

The realistic limit: it is very good at mechanical, well specified, locally verifiable work. It is not a substitute for you on anything that needs a decision about what the product should be.

Opal is a drag and drop builder for AI workflows. You describe what you want in ordinary language, it lays the steps out as a chain you can see and rearrange, and it gives you a working mini app with a link you can share.

If you have used n8n, this is the same job with the wiring done for you. If you have not, the useful comparison is: it is the difference between describing a process and building one.

• Describe the workflow in a sentence, then adjust the steps it produces
• Rearrange, add and remove steps by dragging them
• Publish and share it with a link, so other people can run it
• No code at any point

I get this wrong in the video's voiceover and it is worth correcting here: I call Opal a "video editor". It is not. It is a workflow and mini app builder. Everything else in that line is accurate and the animation shows the right thing, but the words were wrong and it is my error.

The best first project is something you already do by hand every week and resent. A form that arrives, gets sorted, and emails you a summary is the canonical example and it takes about ten minutes.

Mixboard is an AI whiteboard from Google Labs. You drop images, text, notes and scraps onto an open canvas, and you generate new images against that context, repeatedly, until what is on the board matches what is in your head.

Its value is not that it makes images. Plenty of things make images. Its value is that it holds a messy pile of half formed references in one place while you work out what you are actually after.

1. 1
   Start by dropping in what already exists: real photos, screenshots, colours, a competitor's page, a scribbled note.
2. 2
   Generate against that pile rather than from a blank prompt box. The context is the point.
3. 3
   Reject fast and keep going. The rejects are information about what you do not want.
4. 4
   Stop when the board, not any single image, looks like the thing.

This is the tool on the list with the clearest "before" state. Use it when a brief does not exist yet, and stop using it the moment one does.

Pomelli is an AI marketing tool. You give it your website URL. It reads the site and builds a profile of your brand from what is actually there: your tone of voice, your fonts, your colour palette, your imagery. Then it writes and designs social posts and ads that match.

The reason that ordering matters is that every other tool of this kind asks you to describe your brand, and most people describe the brand they wish they had. Pomelli starts from the one that exists on the page.

• Input is a URL, not a creative brief
• It extracts tone, fonts, colour and imagery into a reusable profile
• Output is written copy AND designed assets, on brand, ready to edit
• Best on a site that already looks like you want it to look

The obvious failure mode: if your website is old, off brand or half finished, Pomelli will faithfully reproduce that. Fix the site first, or point it at whichever page is most on brand.