AI is here to stay, and it's getting sharper with every model update. This blog post is about Meridian, a full task list application built entirely inside Grok without ever leaving the chat. It reached a working version in 19 minutes, and one error nearly derailed the lot. Here's how it went.
The power of Grok
Grok is the model people now stack up against Claude and ChatGPT. With Cursor now part of the same group, the coding side will only become stronger. The test was simple: ask Grok to build a task list, and see what the model could really do.
The odd start
My Grok journey began differently from every other tool I've tested. I opened a general chat, downloaded the mobile app, and spent a bit of time getting a feel for the model. Then Grok went on autopilot, without being asked to do anything.
A notification arrived on my phone. I opened it, and Grok was already typing on its own. It wrote itself a prompt for a fun image of a corgi surfing. A whole paragraph I hadn't asked for, which was quite bizarre.
The graphic it created was fantastic. You can see it as this post's cover image, and it also lives on the Digi Creative About page. It captures something unimaginable, which is exactly how AI feels at the moment. Too good not to share.
Building the task list on Grok's free plan
I started on Grok's free plan with a brief of 225 words: build a task list to manage a working week. Just over 19 minutes later it came back with a fully working app, Google sign-in included, and an option to deploy it publicly. Grok built the pieces and connected them.
For context, when I built a client onboarding tool, I had to join the dots myself. Supabase for the data and Cloudflare for hosting, that took several hours. Here the whole stack stayed inside the conversation, which is what makes this so impressive.
What Meridian actually does
At heart, Meridian is a simple task list for keeping you on track through the week, so nothing important slips. Plan ahead the moment you think of something rather than trying to hold it in your head.
There's also a built in teams area, which I will explore in the post at a later date.
Built for desktop and mobile
Its important to build different layout versions for Desktop and mobile, as how you interact with each will be completely different. I always build tools on my laptop first, I then review the design and adapt for mobile.
You can also store the app on your home screen for easy access. Losing the browser bar and gaining a proper icon makes a surprising difference, and it takes about ten seconds.
The complex stuff
Grok reached for tools I've never touched, and the setup sounds more complicated than anything I've used with Claude or ChatGPT. For most people this will read as a load of strange talk, but it's worth including, because it shows what a model reaches for when it's building the whole thing itself. Here's what it chose, and what each part is responsible for.
| What Grok used | What it does |
|---|---|
| React | The framework the whole app is built on. |
| TanStack Start | Organises the pages and how they load. |
| Vite | Packages the code up so it runs quickly in the browser. |
| Tailwind | The styling. It keeps cards, spacing and colours consistent throughout. |
| Better Auth | The login, either through a Google account or a standard email and password. |
| Postgres | The database. Every task and account is stored here rather than in your browser, which is what lets your list follow you between devices. |
| date-fns | Handles the dates. It's why the week runs Monday to Sunday and dates read the British way round. |
| PWA | Short for progressive web app. It means you can add it to your phone's home screen and it behaves like a normal app. |
Disaster struck: the source code disappeared
After the first prompt I moved onto Grok's first paid tier to build the app out properly. Everything stayed inside Grok, and partway through development it lost the source code entirely. A simple update request rebuilt the app from scratch, in a completely different shape from the one that had taken days to get right.
Thankfully I spotted it before publishing, and spent time walking the new version back towards the original. That recovered some of the work, but it took a lot of back and forth. I added a few more features on top, so it ended up a delay rather than a complete disaster.
On a client build, that gap would have been expensive and awkward to explain. To stop it happening again, the source code now lives in a GitHub repo, and I download a hard copy regularly as an insurance policy.
After some investigation, I think the cause was the plan's limit on chat length. It would have been nice for Grok to say so.
Grok or Claude for a client tool?
For a paying client right now, I'd still start with Claude. The architecture feels more solid once it's built. With more experimentation, that could easily change.
There's one more thing to weigh up. Because Grok builds everything for you inside the chat, it burns through your usage far more quickly.
Recommendations
- Start with a tight brief. 225 words was enough to get a working tool.
- Give it a design example from Dribbble, and set your brand colours via Coolors.
- Connect a GitHub repo early as a backup, and push to it regularly.
- Ask Grok to recommend features you should add.
- Test every new feature as it lands, rather than at the end.
Try Meridian
Why not try the task list for yourself: meridian.digicreative.co.uk. Meridian will help keep you organised so you can finish the week in style. Clear cards, tick them off when you're done, or slide the leftovers to tomorrow. Grok got there in one sitting, and that's what impressed me most.
Join the conversation