Lorenzo Rheinicke

Working on:
-

[Lorenzo Rheinicke @lorenzor]

I also want to get a name for this app. Naming it makes it feel more tangible.

[Lorenzo Rheinicke @lorenzor]

Got the PoC of the React Native Expo app working. Now need to polish it and give it to my wife to test and give feedback.

I had an issue with getting the app to load on my iPhone via Expo Go. Was an app version mismatch between the project and the Expo Go app running on my phone.

[Lorenzo Rheinicke @lorenzor]

For the past 5 years, I have mostly used Flutter to build cross-platform mobile apps.

The next app I want to build, I am planning to use React Native + Expo. It is something I want to build for my wife to solve a very specific problem she deals with daily.

The last time I seriously looked at Expo was in 2019, so I am curious to see how far it has come, how it compares with Flutter today.

I will build it with Codex.

[Lorenzo Rheinicke @lorenzor]

Shipped a few fixes and new features on https://t.co/uHFwo6jDdC based on conversations. Have another user interview lined up tomorrow/Monday.

Want to see if they want to pilot it with the new features and give feedback.

[Lorenzo Rheinicke @lorenzor]

Working on https://t.co/uHFwo6jDdC today. Adding functionality based on talking to potential users (x2). Need to talk to way more people.

[Lorenzo Rheinicke @lorenzor]

Then want to get to a point where people can rent/hire my agents to do work for them.

[Lorenzo Rheinicke @lorenzor]

Currently on my mind… building out agents to help me do certain workflows (both internal and service my clients with ad hoc requests)

[Lorenzo Rheinicke @lorenzor]

I know the first round is not going to be perfect… so I just have to start.

[Lorenzo Rheinicke @lorenzor]

@danshipper I like this “All SaaS will need to incorporate the idea of 'presence' for agents”. Sounds novel now but I think it will become a thing.

It will interesting to see your agent is busy doing something while you are also busy using the SaaS.

[Lorenzo Rheinicke @lorenzor]

Codex vs Claude ???

I really enjoy using both. They are good at different things. I feel Codex (GPT-5.4) gives better quality, is more detail-oriented and handles complex issues better.

I do like Claude's style and taste. I do feel like hanging out more with Claude than Codex. 😅

The $100 plan is more than enough for me, but at the moment, I can't really justify having two $100 plans.

Do I alternate? Last month, I was using Claude Max ($100) plan.

Do I change and make Codex my primary driver this month?

[Lorenzo Rheinicke @lorenzor]

Enjoying the new Claude Desktop tool. Looks similar to the Codex app, but "feels" different.

Now I am using the Claude Desktop app to work on one project and Codex on another project to compare. 😅

At least the split is helping me not to hit Opuse 4.7 limits too soon 😅

[Lorenzo Rheinicke @lorenzor]

Been deep in the Laravel AI SDK lately, wiring AI assistants into two projects, and was pleasantly surprised by the following:

Adding an agent alongside a complex multi-step workflow does not just make the workflow easier; it highlights everything that was wrong with the UI in the first place.

Imagine the following: the user opens the app, hits the assistant, describes what they want, and the agent walks them through or completes the task at hand.

Suddenly, the 6-step form that took 10 minutes is done in 30 seconds. Users can still jump in, tweak, edit, and override. But the assistant becomes the primary entry point, not a sidekick.

The unexpected part is the feedback loop. Every time I try it out or watch someone use the agent, I see exactly which parts of the existing flow are genuinely confusing, which fields/steps are redundant or can be simplified.

The agent is basically a live UX audit.

User feedback has been consistently positive. It strips out the complexity without hiding it, and when people do want control, it is right there.

Should the default approach be agent first, then rebuild the UI based on what the agent exposes?

Not for everything...

Simple CRUD does not need this. But for the stuff where users get stuck, where onboarding is painful, where there are real decisions embedded in the flow...

...maybe agent-first is the cleaner starting point, and the UI follows.

Still early, still learning. But the signal is strong enough that I am starting to rethink how I scope these projects and features from the start.

[Lorenzo Rheinicke @lorenzor]

It's now 10x easier to build out new app features... marketing your app / new service... now that is a different story

[Lorenzo Rheinicke @lorenzor]

You only truly start understanding AI agents when you build/implement one.

Thinking about agents is one thing. Wiring one up to solve a problem in your business is where it clicks... the architecture, the edge cases, the trade-offs.

Go build out your own agent. That is the lesson.

[Lorenzo Rheinicke @lorenzor]

It also does not mean you should add agents to everything you build, but make it part of your checklist and see if it really makes your software easier to use.

[Lorenzo Rheinicke @lorenzor]

The Laravel AI SDK makes shipping agent capabilities really easy.

Today I built a context-aware AI Assistant that guides users straight to the outcome they want... no hunting through menus or docs.

Make agents part of the product, not an afterthought.

[Lorenzo Rheinicke @lorenzor]

Shipped an MCP server on https://t.co/uHFwo6jDdC yesterday.

Claude can now monitor the platform, and businesses can plug in their own AI agents to build custom workflows on top of it.

The first step toward making agent-first software.

[Lorenzo Rheinicke @lorenzor]

AI agents are becoming first-class users.

Companies like Notion, Linear, and Basecamp are not just adding AI features; they are enabling agents to use their products.

That is a big shift.

We are moving from:
User-first software to Agent-first software

Going forward, this is the lens I am using to build.

[Lorenzo Rheinicke @lorenzor]

Debugging apps with massive log files? I created a Claude Code skill that pipes logs through Gemini 3 Flash for analysis. Claude Code asks the right questions, Gemini reads the giant files. Result: faster bug fixes, better log insights. Skills are a Claude Code superpower.

[Lorenzo Rheinicke @lorenzor]

Had a Laravel feature importing 500k-1m records at a time. It worked, but it took hours because of all the validation rules.
With the help of Claude Code late last night, the same feature now runs large imports under 10 mins.