Completely agree. It’s a huge jump forwards in my opinion.

The interface is coming on in leaps and bounds.

And finally when I’m feeling like the tasks require a little less oversight during coding, I’ll boot up 2 x Claude Code agents (which can use got worktrees) - each on a different issue.

Final tip - also run a Cursor agent for cleanup tasks, security audits, wrote docs etc as you go 💪🤙

To date this is my most preferred workflow. It allows you:

🔸Space to think about the product
🔸 To stay excited (very overlooked)
🔸 To keep a clean codebase
🔸 To use architectural patterns that you decide
🔸 Flexibility to rollback mistakes
🔸 To stay in control of design and UX
🏎️ To move SOOOOO fast!

I allow MCP access to browser and backend logs (check the MCP code). But it’s great this way as you’ll often see the correlation between the two and it’s much faster to bug fix that way.
When designing front end, hack first if it’s just an idea but I often add code snippets for the exact look.

Get the agent to clarify it has a solid understanding and ask further questions. Explain that I want small chunks at a time and after testing, linting etc, git add. Very important step: I keep an eye as it’s coding but read ALL code before committing. Make changes if necessary and commit.

My current preference is to write very, very detailed issues in GitHub/Linear. I’ve ditched .PRD files, feature files, task files etc. Have a handful of .MD files explaining the product and how I expect it all to run in simple detail. Then feed in the issue and start a new feature branch.

The daily toolset is jumping between Claude, ChatGPT, Claude Code running in Cursor, and hooking up to tools I need. But what I have found about 1.5 years into going all in on this is you find your own personal workflow. And tools are becoming incredible. The world of TUIs is blowing up and powerful

Also started building a solution to some of the marketing challenges we face in the companies I’m in. Started a few weeks ago. Already 50% of a production grade MVP. Lots of AI tools in it, tests, a full image editor, positioning assistant, billing, workspaces, permissions, CSP, webhooks + more 🚀🚀

And if you want to really have tight ai design control you can use systems like www.aura.build

Hope this can be of interest. During the Covid lockdowns I found Yancey Strickler’s Bento framework quite a nice lens through which to re-evaluate things. The balance of self-interest vs collective interest plus timeline was great for finding energy again. 👍

www.iftf.org/insights/ben...

Although I think from a code perspective I had around 3-4 months feeling completely bored by it all. And then I took a break, reintroduced AI into my workflows in a way that suited me, and now it feels like a power tool again.

Sometimes you need a hammer. Sometimes you need a nail gun.

It feels a bit like being asked to design something with no constraints. When everything is possible it’s just noise.

💯

bsky.app/profile/jimh...

Also think it’s hard to realistically understand the U.K. public when everything is in London. Like the way Apple always tests their products in sunny Cupertino.

Screens from the Ordaloca app. These were the first version, but most of these stayed around in one way or another.

From GPS delivery points, local loyalty points (geofenced), discovery of businesses nearby, group purchasing and more…

These are the early mockups that we used to initially get customers onboard. These were built into a modular UI system that supports lots of product complexity: from bundled items with SKUs to booking forms, customisation, add-ons, complicated variants and more.

🔥Over 2000 products listed for a town of 35000.
🛍️ Each business with its own storefront.
🧾 All local orders in one place.