How I Built an AI Agent to Read the News for Me for Free: The Complete (and Messy) Story – No Code

I have a confession: I can’t keep up with the news. Between tech, world events, and industry updates, my brain feels like a browser with too many tabs open. I wanted a personal news assistant, but I didn’t want to pay for one. So, I decided to build my own.

This post is the raw, unfiltered story of how I built an autonomous AI agent to be my personal news summarizer. It scours the internet for the latest articles, uses Google’s Gemini AI to write a perfect one-sentence summary, and neatly organizes everything for me in a Google Sheet.

But this isn’t a perfect, polished tutorial. This is the real story. You’ll see my initial plan, the frustrating errors that made me want to quit, the “aha!” moments, and the crucial tweaks that finally led to a breakthrough. By the end, you’ll not only have a working AI agent of your own but you’ll also have learned the most important skill in tech: how to debug your way to success.

What You’ll Build

You will create a fully automated workflow that connects three powerful services:

The final result is a Google Sheet that magically fills itself with the latest news, perfectly summarized, every single day. In fact, here is the live output from the very agent we’re about to build. This sheet will continue to update as long as the agent is running!


Click here to open the live results in a new tab.

The Toolkit: Everything You Need is Free 🛠️

Before we start, make sure you have accounts for these three services. All of them have generous free tiers that are perfect for this project.

  1. A Make.com Account: Our no-code “workshop” where we’ll assemble the agent.
  2. A Google Account: For our spreadsheet and the AI’s brain.
  3. A Google Gemini API Key: Get this from the Google AI Studio. This is the key that starts the engine.

Step 1: The Plan and The Workspace

The plan was simple: connect three modules in a chain. An RSS Feed to find the news, a Gemini AI module to think about it, and a Google Sheet to write it down. Easy, right?

First, I set up the workspace. I created a new Google Sheet named “AI News Summaries” and added three headers in the first row: Title, Summary, and Link.

Step 2: Building the Agent (The “Easy” Part)

In Make.com, I started a new scenario. I added the RSS module (“Watch RSS feed items”) and pointed it to the BBC Technology feed. I initially set it to grab 3 articles at a time—this was my first mistake.

Next, I added the Google Gemini AI module (“Create a Completion”) and connected my API key.

Finally, I added the Google Sheets module (“Add a Row”) and mapped the fields. It looked perfect. My agent was built.

Step 3: The First Disaster – The “429 Quota Exceeded” Error

I clicked the “Run once” button, filled with excitement. And then… disaster.

ERROR: [429] You exceeded your current quota…

The workflow stopped at the Gemini module with a glaring red error. My agent had a brain, but it refused to think!

Tweak #1: The “Maybe I’m Too Fast” Tweak

My first thought was that I was hitting a per-minute rate limit. So, I added a “Sleep” module to make the agent pause for 5 seconds between each article. A clever fix, I thought.

I ran it again. Same error. My frustration grew.

Tweak #2: The “Aha!” Moment – It’s Not Speed, It’s Weight!

I realized my other Gemini apps were still working, so my account wasn’t blocked. The problem was specific to this workflow. The culprit? Token usage. The Description field from the RSS feed was huge. I was asking the most powerful AI model (Gemini 1.5 Pro) to read a novel for each article. This was too “heavy” for the free tier.

This led to the breakthrough: **Optimization.**

Step 4: The Fix – A Leaner, Smarter Agent

I went back into the Gemini module settings and made two crucial changes.

The Two-Part Optimization Fix

  1. A More Efficient Model: I switched the AI Model from the powerful Gemini 1.5 Pro to the fast and efficient Gemini 1.5 Flash. It’s designed for quick summary tasks like this.
  2. A More Efficient Prompt: I completely rewrote the prompt. Instead of sending the entire article description, I only sent the much shorter Title and Summary fields.

Here is the final, optimized prompt I used. It uses far fewer tokens and gives the AI a clearer, more focused task:

Based on the following title and summary, write a single, ultra-concise news headline (less than 15 words).

Title: [Map the 'Title' from the RSS module here]
Summary: [Map the 'Summary' from the RSS module here]

Step 5: Victory! The Agent Works

With the optimized module in place, I clicked “Run once” one more time. I watched as the little bubbles flew from module to module… and completed without a single error.

I opened my Google Sheet, and there it was. A new row, perfectly summarized by my AI agent. You can see the live results for yourself in the embedded sheet at the top of this article!

Final Step: Full Automation

To make my agent truly autonomous, I clicked the scheduling toggle at the bottom of the screen and set it to run at a regular interval. Now, it works for me in the background, a silent, efficient news assistant, constantly adding to the sheet we’re sharing here.

What I Learned

This project taught me more through failure than it ever could have through instant success. The real lesson wasn’t just how to connect APIs, but that **optimization is often a better solution than upgrading**. Before you pay for more power, see if you can use the power you have more wisely.

So, go ahead and build this. And when you hit your first error, don’t get frustrated—get curious. That’s where the real learning begins.