How I Use Claude AI Every Day — 3 Automations That Actually Work

In my day-to-day work, Claude AI is no longer just a chatbot I open whenever something comes to mind. It’s a system that sends me my daily schedule by 6:30 a.m., sent emails to 1,168 people on my behalf last week, and uses LinkedIn to find and connect with people worth adding to my network. I’ll show you how it’s set up—and what they never tell you about these automations.

This time last year, I still thought that AI in my work meant nothing more than being able to write emails faster. I’d open a window, ask it to put my thoughts into words, skim through it, and send it off. It’s useful—but it’s pretty much like pushing a Ferrari because I don’t know how to start it.

The change came about because I got tired of it. I got tired of spending my first hour of the morning reading emails, checking my calendar, scrolling through Jira, and then mentally piecing together what I needed to focus on that day. It’s the same routine every day, with the same sources and the same results. If a person always does this the same way—a machine can do it too.

That’s how Gloster’s AI automation project got started, and we’ve since expanded it to include 23 skills, 7 custom MCP connectors, and daily operations.

Why Claude AI in chatbot mode isn't enough—and what most people are missing

90 percent of people use AI this way: question → answer → copy-paste. That’s the bare minimum you can get out of it.

The real breakthrough comes when Claude AI doesn’t just respond— it takes action. It sends emails. It retrieves data from a system. It reads calendars. It updates Jira tickets. It saves files to a folder. To do this, it’s not enough to simply open claude.ai—Claude needs to be integrated with real-world systems.

The name of the interface technology is MCP—Model Context Protocol. This is Anthropic’s open standard for how an AI model can access external tools and data sources. I run these on Cloudflare Workers—a serverless, cloud-based runtime environment that costs around $5 a month and can be connected to any API.

My 7 connectors are currently: Outlook email, Outlook Calendar, Jira, WordPress, LinkedIn, Tempo (time tracking), and a general Webhook. These are the “channels” through which Claude AI can access my actual work environment.

AI isn’t valuable because it provides smart answers to smart questions. It’s valuable because it does the work that I’ve been doing up until now—every morning, reliably, without complaint.

1. Automation: the morning briefing, which is in my inbox every morning

I’ll explain how this system works—because most articles leave this out, which is why it seems so mysterious.

What exactly happens at 6:30?

A scheduled task is triggered (managed by the Cowork desktop app). This launches Claude AI, which retrieves the following in parallel via the MCP connectors: today’s calendar events from Outlook Calendar, emails from the past 48 hours from Outlook, and open Jira tickets (assignee = me, status not Done, sorted by priority).

Claude compiles all of this into a single, coherent briefing and then sends it to me via Outlook using the email connector. It’s in Hungarian and is designed to take no more than two minutes to read.

The structure of the briefing is always the same:

• #1 Priority — one sentence explaining what’s most important today and why
• Today's schedule — meetings listed by time, who is attending, and what the topic is
• Email priorities — up to 5 emails that require a response or action (filters out automated notifications and newsletters)
• Jira — List of open issues sorted by priority
• Recommended Actions — Top 3 Concrete Steps to Start Your Day

How much are the actual savings?

Doing the same thing by hand takes me 20–25 minutes every morning. With a 5-day workweek, that adds up to 100–125 minutes. That’s about 85 hours a year—roughly two full workweeks—that I’ve spent gathering information that a machine can gather for me in ten seconds.

The system has been running flawlessly for six weeks now. There was one instance when the Outlook API slowed down, causing the briefing to be delayed by 40 minutes. That was the only incident.

2. Automation: 1,168 personalized emails — in 3 days, without human intervention

This is the project where I first felt that I was dealing with something qualitatively different.

For one of Gloster’s campaigns, we needed to process a contact list of 1,168 people and reach out to them via email. Previously, three people had done this over the course of two weeks: they took turns sending 50–100 emails a day, manually copying names and company names and personalizing the salutations.

The AI-based version looked like this:

Input: an Excel table containing name, email, company name, and industry.

Processing: Claude AI goes through the list, inserts the appropriate salutation for each contact based on their name (Mr. Kovács / Petra / Dr., etc.), tailors the text based on the company name, and selects one of the three prepared email templates based on the industry. This is not a mass mailing—every email is unique.

Output: The Outlook MCP connector sends emails one at a time, in batches of 20, with 3- to 5-minute intervals (to avoid triggering the spam filter).

The result: 1,168 emails sent in 3 business days, with 0 hours of manual work during the sending phase. Preparing the entire campaign (writing the template, cleaning the list) took about 4 hours.

By way of comparison: using the previous method, this would have taken 80–100 person-hours.

There are some things whose value you only truly appreciate when you don't do them yourself at first. This was one of those things.

3. Automation: LinkedIn networking — data-driven, without manual browsing

This is the trickiest of the three, because LinkedIn actively defends against automation. But there is one area where it works legally and effectively.

The process

Input data: an Excel list containing names and email domains (e.g., kovacs.janos@cegnev.hu). Based on this, Claude AI searches for LinkedIn profiles via a web browser—combining the name with the company name in its search, determines the degree of connection, and categorizes the results:

• 1st-degree connection → already in my network, no need to mark
• 2nd-degree connection → message can be sent; the system will send it
• Not found → missing data; manual verification required
• Uncertain match → the system skips it and flags it for review

Last week's results: 26 contacts processed. 3 nominations sent; 11 were already first-degree connections; 8 could not be clearly identified; 4 were uncertain.

At first glance, this doesn't seem impressive—but keep in mind that performing these 26 checks manually takes about 45–60 minutes, and scrolling through LinkedIn's search results can be pretty tedious. Meanwhile, the AI is doing other things for me.

Areas still under development: name recognition accuracy decreases when the list contains only partial names (e.g., “Dr. Kovács J.”). We are working on this—we can likely improve it with a dedicated normalization step.

What it is NOT suitable for — and this is also important to mention

It’s not exactly something to brag about, but my AI system has failed miserably in a few areas.

I tried to automate the drafting of investor communications. Technically, it worked—Claude generated the responses. But the output was always too generic; it lacked the nuance needed to respond to the specific situation. I’ll continue to do this manually.

I also tried to automate the summaries following client meetings. This worked to some extent—if there’s a transcript, Claude does an excellent job summarizing. But if there isn’t a high-quality recording, the output will be poor. Garbage in, garbage out—that applies to AI as well.

The rule of thumb I’ve come up with is this: AI excels when the task is structured, repetitive, and the output is well-defined. When context, nuanced judgment, or true creativity are required—that’s where I’m still the bottleneck, and I’ll remain so.

The total cost of the system — something no one talks about

• Claude Max: ~35,000 HUF/month (at this usage level, the Pro plan is no longer sufficient)
• Cloudflare Workers (7 connectors): ~2,000 HUF/month
• Cowork desktop app: currently free (research preview)

Total: ~37,000 HUF/month.

So, for that amount of money, I saved myself about 30–35 hours of repetitive work each month. Even using the most conservative estimate, the 37,000 HUF investment yields a return of at least 10–15 times that amount—every month.

I wouldn't recommend it to everyone without reservation. If you don't have the basic technical knowledge to set up MCP connectors, the first few weeks will require a significant investment of time. In our case, two developers from our team helped build the connectors—this time should also be factored into the actual implementation costs.

Frequently Asked Questions (FAQ)

How do I get started if I'm not a developer?

The easiest way to get started: Claude Pro ($20/month) + the Cowork desktop app (free). This lets you immediately start working with files, emails, and your calendar—even without an MCP connector. Custom connectors (Jira, proprietary systems) are only needed if you want to automate those tasks as well.

How much will it cost if I'm starting from scratch?

Realistically: Claude Pro costs 7,500 HUF per month, and Cloudflare Workers costs 2,000 HUF per month. Setting up the initial connectors takes 4–8 developer hours (a one-time investment). If someone on the team knows Python or JavaScript, this can be handled in-house.

What exactly is an MCP connector, and why is it needed?

The MCP (Model Context Protocol) is Anthropic’s open standard that allows Claude AI to “interact” with external systems—read your emails, query Jira, and send messages. Without a connector, Claude only knows what you type into the chat window.

Is it really reliable for everyday use?

After 6 weeks of experience: yes, when the task is well-defined. The morning briefing runs with 99% reliability. During the email campaign, one batch was delayed once (due to an incorrect email format in the list)—that had to be corrected manually. Automation isn’t flawless, but it frees up my attention for more important things.

Summary

Six months ago, I would have said that AI is a useful tool for editing text in my work. Today, I say: it was the best decision I’ve made in our day-to-day operations over the past two years.

Save 8–10 hours a week for just 37,000 HUF a month. With three specific automation tools that anyone can set up—if they have the patience.

AI won't take away jobs. But those who know how to use it will seriously take market share away from those who don't.

If you found this article helpful, follow Viktor on LinkedIn — he regularly shares his thoughts on IT, AI, and business leadership.

---

🔔 Want weekly AI and IT trends delivered straight to your inbox?
Sign up for the Gloster newsletter.