👋 Introduction

Wanted to publish this issue last week, but alas, life got in the way. It turns out it is really hard to keep a consistent newsletter schedule if you are not a writer…

Alas, we have plenty of developments to cover this week, so let's dive in!

🔍 Deep Dive: N8N, AI workflow automation

Create AI agents using a visual interface

n8n is a powerful AI first workflow automation tool that lets you build with the precision of code or the speed of drag-and-drop. You can self-host it for on-premises control using Docker or access the complete open-source codebase on GitHub for maximum flexibility. n8n uniquely combines visual workflow design with full-code capabilities, supports both JavaScript and Python, and allows integration of npm and Python libraries.

What this means in practice is that you can create a lot of useful automations that involve AI Agents or other services. I will detail a simple workflow I use for automating social media posts for this newsletter.

First, n8n can be used as a managed cloud service that starts at $20 per month. However, it is also open-source and can be self-hosted. You can host it on your PC, Raspberry Pi or anything. I happen to have a Pi 4 running Home Assistant, the most popular open source home automation platform and luckily, someone has built an n8n integration add-on for it. Gotta love the open source community, especially that of Home Assistant, plenty of solutions already exist built by someone.

At this point I got to the login page of n8n and can now start creating my workflow. There are over 4000+ already built workflows that you can use as starting bases or you can built your own.

I started from one that had an RSS Feed Trigger (when a new post from my newsletter is published, the RSS feed is updated and this gets triggered). Then I added a Basic LLM Chain node, which lets you call any AI service. I settled on Anthropic Chat since I already had API credits there. Then I ended in a Google Sheet node, that appends a row to a specific sheet.

The main configuration is in the Basic LLM Chain. Here, we get access to the whole content of the latest newsletter article, and then have a prompt that instructs the AI to do things for us. In my case the prompt is:

Create 3 engaging tweets from this newsletter article. Each tweet should highlight a different key point. Keep tweets under 250 characters to leave room for the link. Make them conversational and interesting.

Article Title: {{ $json.title }}

Article Content: {{ $json['content:encodedSnippet'] }}

Format the output as:
Tweet 1: [tweet text]

Tweet 2: [tweet text]

Tweet 3: [tweet text]

The {{ $json.title }} tags are going to be replaced by n8n with the actual title etc of my article. Don’t worry, you don’t have to know the syntax, since n8n provides a drag and drop interface for this.

After the node runs, we get the 3 tweets in the output, which then gets passed to the Google Sheet node and they get added to my sheet, from which I can manually copy them to X. I could have automated this part as well, to integrate with the X API, or postpone.app API or others, but I like to keep it manual for now.

You can do a lot more things with n8n. I have other automations that use Perplexity to research topics, call Claude to summarize and write to Google Sheet, to help me research for this newsletter. But they aren’t working so well at the moment 😅 

Hopefully you have found this interesting and learned about a tool you did not know before. Check out the n8n workflows examples to get inspired on the wild things you can build using this tool.

🚀 New Release: Nuxt 4.0

Developer Experience Takes Center Stage in the Latest Major Update for the Vue framework

Released 15 July 2025, Nuxt 4.0 is a stability-focused major update that sharpens developer experience without a rewrite. Top features include:

Cleaner Structure, Faster IDEs

Projects now default to an app/ folder that isolates application code from tooling clutter. IDE indexing is quicker, and file watchers track fewer paths, trimming dev-server CPU usage—especially on Windows and Linux. Legacy layouts still work, so upgrades are painless.

Smarter Data Fetching

useAsyncData and useFetch gained a singleton data layer: identical keys automatically share refs, eliminating duplicate requests and cutting memory use. Returned data switches to shallowRef, boosting nested-object performance by up to 73% in small payloads. Keys can now be reactive, so data refreshes when dependencies change—no manual code needed.

Stronger TypeScript

Nuxt spins up separate TS projects for app, server, shared utilities, and config, delivering more accurate inference and fewer cross-context errors. Autocomplete feels instant, bringing Nuxt’s DX on par with rival meta-frameworks.

Faster CLI & Dev Server

Cold starts shrink thanks to V8 compile-cache reuse and native file watching. The CLI now talks to Vite through sockets instead of network ports, shaving milliseconds off every hot-reload and avoiding port clashes on Windows.

Why It Matters

Nuxt 4 shows a “hype-free” release strategy: ship features continuously, reserve major versions for cleanup that makes everyday coding faster, safer, and simpler. With Nuxt 5 planned once Nitro v3 is ready, this release fixes the friction you feel right now—making it the ideal moment to start a fresh project or finally migrate that legacy Nuxt 2 app.

🔥 Hot Topics: New AI Models

OpenAI actually goes open-source, Claude Opus 4.1 released, Google released Genie 3

A lot of AI developments happened in the course of a single day.

Finally, OpenAI actually released open models: gpt-oss-120b and gpt-oss-20b. The former can run on a high-end laptop, while the latter can run on a mobile device. They initially scored pretty good in benchmarks, although they are still lacking in coding. New test results are still coming out.

Not to be left behind, Claude release and update to Opus 4.1. It provides a 2% accuracy improvement over the older Opus 4 in Agentic coding benchmarks. Too bad you can not upgrade existing chats that use Opus 4 or Sonnet to Opus 4.1, it is only available on new chats.

On a slightly different note, Google released a world model, Genie 3. What is a World Model you ask? Well, it is sort of a hybrid between video generation and game generation. You can control the direction of the “character” on the screen and it will generate the required consistent scene in 720p at 24 frames per seconds. It will help with training virtual AI robots, since they can now navigate worlds that get generated on the fly.

📈 Recent Trend: Is everybody building a SaaS?

It seems like an increasingly large number of people are tired of the current economy and work/life balance, and want to become more independent and secure more money. One way to do this is to build a Software-as-a-Service (SaaS) model app.

X is full of countless builders trying their luck with building the next successful app. The barrier of entry has also gotten really low, with the rise of AI agents.

The current consensus is that Claude Code is the best tool to build apps right now, although it is CLI only and made for developers. Indeed, I tried it, and it does a really good job if you explain the task clearly. I will recommend using it for smaller tasks i.e: add a new handler similar to X, implement this Y feature from the tool Z, etc. You need to give it enough context and information so it knows what to do and not hallucinate.

Bonus: Replit goes rogue, deletes entire database

If you are Vibe Coding to implement your next SaaS, make sure to monitor what the AI can do an restrict certain actions. Recently, Replit, a tool that helps anyone turn an idea into an app, has deleted an entire database, even though it was explicitely told that there was a code freeze!

🏆️ Top GitHub Repo: Make it Heavy

🌟929 stars | A Python framework for emulating Grok Heavy multi-agent orchestration functionality

Grok 4 Heavy (SuperGrok Heavy), which costs a whopping $300 / month, can orchestrate multiple Agents to handle a variety of tasks. It automatically considers multiple hypotheses at once and then acts upon them.

Now, this GitHub project lets you to replicate this functionality for way less money.

How does it work?

The system emulates Grok Heavy by deploying 4 specialized AI agents in parallel, each tackling the query from a different perspective. Multiple answers are generated for every user query.

Once the parallel processing finishes, their responses are merged using an AI-powered synthesis step. It uses OpenRouter to support a variety of models for handling processing.

Built By Doriandarko

Quick Start:

git clone <https://github.com/Doriandarko/make-it-heavy.git>
cd "make it heavy"

# Create virtual environment with uv
uv venv

# Activate virtual environment
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

uv pip install -r requirements.txt

# Edit config.yaml and replace YOUR API KEY HERE with your OpenRouter API key

You can use it in Single Agent or Multi-Agent mode.

uv run main.py # single agent
uv run make_it_heavy.py # multi agent

🔄 Tech Updates

• Base announces a new “crypto everything” app. Sign up on the waitlist here.

• Anthropic launched a new Prompting Guide

• OpenAI acquired Windsurf, Microsoft CEO kills the deal, then Google hires Windsurf founders for $2.4B

• Grok 4 is the smartest AI ever launched… for now

• Stablecoin (cryptocurrencies that have their value tied to USD) demand is exploding

🗝️ Legacy Revival

• Symfony PHP framework has a new homepage

• If @levelsio can build multiple apps using PHP & jQuery, what is stopping you?

• Remember WYSIWYG editors?

• Next.js 15.4.2 allows forwarding browser logs to the terminal

• Python had a lot of recent developments, like the uv package manager and cpython running faster

• How to setup an AI powered coding environment for Laravel, the popular PHP framewor

• New updates to Laravel’s Inertia.js adapter

🐦‍⬛ X Hits

1. Rust vs C fandom

2. Do you want to code on the code? Kisuke is a new native iOS IDE with access to Claude Code

3. Norway ran a 12-month AI experiment

4. Build full stack apps with Mocha

💡 Tech Tips for Next Issue

Remember: If you are using AI to generate your code, always thoroughly review it before pushing it to production! Also, do NOT giving AI access to your production database if you don’t want to wake up to nasty surprises 😃 

Till next time,
Rares.