FiveTech Support Forums

Dear Antonio, I’d love to hear your perspective on something: How do you compare Google Antigravity and Claude Code?

From my experience so far, Antigravity has been fantastic—thanks for encouraging me to try it! It’s truly impressive.

Antigravity feels like a full-fledged modern IDE, offering visual agent management, multi-agent orchestration, browser control, and artifact handling. It’s incredibly intuitive and feels like the next generation of development tools. I’ve tested it myself and can confirm its capabilities.

Claude Code is making quite a splash—you’ve probably already given it a try. From what I read. Claude Code, on the other hand, seems to prioritize a terminal-first approach. It excels in deep reasoning, large-scale refactoring, and complex multi-file tasks, though it leans more toward command-line interaction.

I’d be grateful for any insights or experiences you might have—especially given your keen eye for robust architectures. Thank you again for everything you do for the Harbour community. Thanks to you, we’re the world’s leading xbase community. Warm regards, Otto

Dear Otto,

I have been intensively working with the three of them: Antigravity, Codex and Claude Code.

Claude Code is by far the best one :!: :idea:

In fact I plan to continue working just with Claude Code. Its really impressive :)

The best "developer" you can hire to help you :wink:

Dear Antonio, Thank you so much for sharing your experiences with Antigravity, Codex, and especially Claude Code! It’s great to hear that you find Claude Code so impressive and effective—it sounds like a game-changer. Since you’ve been working extensively with Claude Code, I was wondering: Do you have a simple “how to start” guide or any tips for getting started? That would be incredibly helpful as I begin exploring it myself.

I also have a strategic question: I’ve mainly worked with Grimoire as my primary AI, and after the update to 5.4, it was less effective for a few days, but now it seems balanced again. I’ve also dabbled with Antigravity. If I switch again, is it productive in the long run? Or do you think AI performance tends to balance out over time, making the short-term learning curve of a new AI worthwhile before the next update arrives?

Thanks again for your recommendation and support! Best regards, Otto

Dear Otto,

Claude code is so impressive from day one !!!

Have you ever worked with "CLI" (command line interface) tools ?
Claude code is a "CLI", basically it looks as an old Msdos text mode app. You can have running more than one from different cmd windows at the same time on different tasks :idea:

You go to the folder where your code is. Then open a cmd window and run "claude" and the magic begins...

Ask it whatever you may need. Read the anwers, get impressed by its logic, and totally shocked with the quality of its code.

I would say that Claude code is right now the "developer" that better understands FiveWin :!: :idea:

Dear Antonio, thank you for your reply. Which plan do I need to test? Will Pro work? Best regards, Otto

Dear Otto,

Max 5x is the way to go for a professional work :wink:

You can get a free week to test it from a friend doing /passes from the CLI.
My son sent me the invitation :!:

Dear Otto,

good morning,

I already suscribed for Plan Max :wink:

Claude code is such a wonderful tool :idea: :D

Dear Antonio,

Thank you again for your advice to use Claude Code. It really looks very good, and as far as I can tell so far, it also works excellently.

In any case, I had a very, very short night today. I also just asked whether Claude Code can help me in the HARBOURINO style – and it actually seems that it can. I will definitely try it out.

Even though HARBOURINO is now becoming significantly less important because the AI delivers practically error-free code, I still like the approach with the patcher and preprocessor. Especially that the code is modeled after a flowchart.

Thanks again very, very much – I will keep you posted! Best regards, Otto

I'm currently on a first-name basis with /help — we're practically best friends. 😄

📁 Command Structure Overview

The graphic illustrates the hierarchical organization of all available commands, divided into seven main categories.

At the center is the 📁 COMMANDS node, from which all command groups branch out. The structure follows a three-tier logic: the first level contains the core areas CORE, CONTEXT, and SEARCH. These form the foundation for daily work—here you control the chat (/clear, /reset), manage files (/add, /remove), and search within the project (/search, /grep).

Below them—connected by arrows—follows the second level with the active working areas CODE, EXECUTION, and SETTINGS. While CODE focuses on code quality with commands like /explain, /fix, and /refactor, EXECUTION handles running code (/run, /shell). SETTINGS provides access to configuration (/model, /config, /tokens).

The structure is completed by the 📜 SESSION card at the bottom level. It is intentionally placed as a single node beneath EXECUTION, as session commands like /history and /save are typically used after working with the code—for example, to review the conversation history or save the current state.

The color-coded design of the cards aids orientation: blue for Core, green for Context, yellow for Search, purple for Code, orange for Execution, red for Settings, and gray for Session. Arrows visualize the logical relationships between levels, making the dependencies between command groups easy to follow.

/btw how is it going ?

This one is very usefull to know what it is doing if it is taking time

:-) thank you, Antonio, very good – and once again many thanks for your advice – and your work in research and development for all of us. many, many thanks.
... and I will make a note of the command

Yesterday I was telling my oldest son that Claude code is the new "Clipper"... :wink:

Dear Antonio,

We just reduced 175 lines of HTML to 29 — and accidentally built something that feels like Harbour.

Before:

<div class="xwh-personen-grid">
<div>
<label for="PERSONEN" class="xwh-label">Personen</label>
<input id="PERSONEN" type="number" inputmode="numeric"
min="0" max="99" step="1" value="0"
onfocus="if(this.value==='0')this.value=''; this.select();"
onblur="if(this.value==='')this.value='0';"
class="xwh-input compact">
</div>
<!-- × 3 person fields, × 4 meal categories with 2 inputs each -->
<!-- ... 150 more lines ... -->
</div>

After:

<div class="group cols-3">
$-> COUNTERFIELD : caption = Persons; id = PERSONEN
$-> COUNTERFIELD : caption = Adults; id = ERWACHSENE
$-> COUNTERFIELD : caption = Children; id = KINDER
</div>

<div class="xwh-meal-grid">
$-> COUNTERFIELD_2 : title = No breakfast; id1 = ZO; id2 = KIND_ZO
$-> COUNTERFIELD_2 : title = Breakfast; id1 = ZF; id2 = KIND_ZF
$-> COUNTERFIELD_2 : title = Half board; id1 = HP; id2 = KIND_HP
$-> COUNTERFIELD_2 : title = Full board; id1 = VP; id2 = KIND_VP
</div>

<div class="group cols-2">
$-> NFIELD : caption = Deposit (€); id = deposit
$-> DFIELD : caption = Due date; id = depositdate
</div>

This is Harbourino — a homegrown template system that expands $-> block calls into fully styled HTML at build time.

We're migrating a large hotel management system from FiveWin/Harbour to the browser. The parallel is intentional:

Each block (COUNTERFIELD, DFIELD, NFIELD, SFIELD, CHKFIELD, TFIELD, RFIELD, HFIELD...) is defined once — HTML
structure, CSS, and JS behavior included. After that, you build dialogs declaratively, the same way you did in Harbour
20 years ago.

The goal: migrate dozens of FiveWin program modules to HTML with minimal risk and maximum consistency.

175 → 29 lines. Same functionality. Maintainable. Readable. Harbour/Fivewin feeling in the browser.
Best regards,
Otto

These screens show Claude Code at work – Claude Code supports Harbourino.

So glad that you are enjoying Claude code :wink:

What a tool! :idea:

Karpathy says he hasn't written a line of code since December and is in "perpetual AI psychosis." How many Claude Code users feel the same?

Just listened to his new No Priors episode. He describes going from 80% writing his own code to 0%, spending 16 hours a day directing agents, and being in a constant state of "AI psychosis" because the possibilities feel infinite.

Garry Tan calls it "cyber psychosis" — sleeping 4 hours because he can't stop building with Claude Code.

I've seen similar vibes in this sub — people running multiple Claude Code sessions in parallel, hitting the rate limit daily, feeling like idle tokens are wasted tokens. Is this just a handful of high-profile people, or is "AI psychosis" way more common among Claude Code users than we think?