Forem: Valentin Monteiro

Claude Expert: The 4 Official (Free) Anthropic Courses

Valentin Monteiro — Tue, 21 Apr 2026 11:44:08 +0000

On LinkedIn, everyone's an "AI expert". Three ChatGPT prompts, a 12-hour bootcamp, a blue badge, done. In reality, almost no one has taken the official Claude courses Anthropic published. They exist. They're free. And they're the only source that actually counts.

"Using Claude" is not "being a Claude expert"

Using Claude means opening a conversation and asking a question. Plenty of people do it well. Few understand what's happening underneath.

A Claude expert understands something else. How the API is structured. What an Agent Skill is and when to write one. How the Model Context Protocol connects Claude to an existing system. What Claude Code changes when you move from chat to actual software engineering.

This isn't extra theory. It's the difference between "I prompt well" and "I build products with Claude". The four courses below cover exactly that ground. They aren't optional if the phrase "Claude expert" is supposed to mean anything.

How to become a Claude expert

Short answer: by taking the 4 official courses Anthropic put on Skilljar for free. Claude with the Anthropic API, Introduction to Agent Skills, Introduction to Model Context Protocol, Claude Code in Action. Around ten hours total.

Long answer: the rest of this article. But the path fits in four modules, nothing else gets you to that level of mastery that fast, and all of it is free.

Prerequisites before starting

These courses are free, not beginner-proof. You need a baseline for the time invested to pay off:

Reading and passive listening English. Videos are in English on YouTube, with English subtitles available. No official dub.
At least one programming language under your belt (Python, JS, PHP, VBA, whatever). You can start from zero, but plan 2 to 3 hours of Python basics with Claude as your tutor before attacking the API module, otherwise frustration hits fast.
An Anthropic API account. Free to create, pay-as-you-go (a few cents per test). You need it to practice what you learn.
Minimum terminal comfort. Level: "I can run git clone and execute a script". No DevOps chops needed.

The 4 official Anthropic courses

1. Introduction to Agent Skills

Skills are the mechanism through which Claude learns to perform a precise task in a repeatable way. It's what you reach for when you want Claude to follow a method, not just deliver an answer. The course covers what a Skill is, how it's structured, and how to write one that actually holds up.

Why it's in the foundation: most serious AI workflows end up being Skills. Without this, you rewrite the same prompt in every conversation.

Introduction to Agent Skills →

2. Claude with the Anthropic API

Leaving the chat interface. This is when Claude stops being a personal assistant and becomes a component in a product. The course covers the basics: API calls, message handling, streaming, tool use, caching. The core toolkit for integrating Claude into code that runs.

Why it's in the foundation: without the API, Claude stays a chat. With the API, it becomes a brick you can drop into any stack.

Claude with the Anthropic API →

3. Introduction to Model Context Protocol

MCP is the standard Anthropic is pushing to connect Claude to external systems: a database, an internal tool, a third-party API. Instead of duct-taping custom integrations every time, you write an MCP server and Claude talks to your system.

Why it's in the foundation: this is the layer that turns Claude from an assistant that answers into an agent that acts inside your environment. Ignoring MCP in 2026 means missing half of what Claude can actually do.

Introduction to Model Context Protocol →

4. Claude Code in Action

Claude Code isn't a VS Code extension that writes a function for you. It's a full dev agent that reads your repo, understands your stack, and modifies your code while making sure tests pass. This course shows how to actually use it, not as a gadget.

Why it's in the foundation: if you're building with Claude, you're probably building with Claude Code. Knowing how to drive it is the difference between 2x and 10x productivity.

Claude Code in Action →

What this foundation actually changes

Before these four courses, you prompt. You iterate. You duct-tape. It works, but you're guessing a lot.

After, you know why a given call fails, why one prompt costs ten times too much, why your agent "doesn't get it" when the real issue is how you handed it context. You stop fighting the tool. You stop selling vague "AI solutions". You build.

That's exactly the foundation I work on as a Claude consultant. Not a LinkedIn badge, not a 40-slide deck on AI transformation. The technical base Anthropic made freely available, that just needs someone willing to sit down and go through it.

The only certification that counts when it comes to Claude is the one delivered by the company that builds Claude.

FAQ: becoming a Claude expert

How long does it take to complete the 4 courses?

Around ten hours total. Spread over 2 to 3 weeks at one module per weekend, it's a comfortable pace that won't block your other projects.

Is there a certification at the end?

Yes. Each module delivers a Skilljar completion certificate signed by Anthropic. It's not an academic degree, but it's an official signal that you completed the track. Worth displaying on your LinkedIn or CV if it helps.

What order should I take the 4 courses in?

Claude with the Anthropic API first (the technical foundation). Then Introduction to Agent Skills (you build on top). Then Introduction to Model Context Protocol (you extend). Finally Claude Code in Action (you apply everything in a real dev workflow).

I have zero programming background, can I still follow?

Yes, with a plan B. Start with Agent Skills (no code required). Before attacking the API module, spend 2 to 3 hours on Python basics with Claude as your tutor. Free, immediate, and it unlocks the rest.

Should I take these courses if I'm already an advanced Claude user?

If you haven't touched the API, written a Skill, built an MCP server, or used Claude Code in real conditions: yes. Being an advanced chat user and being a Claude expert are two different levels.

The truth is, most won't do it

These four courses add up to maybe a dozen hours. They're free. They're official. And most "AI experts" on LinkedIn will never take them, because flashing a badge is faster than opening the docs.

The ones who do stop being generic AI experts. They become Claude experts. Different thing.

I Applied to the Claude Partner Network as a Solo Consultant. Here's What Happened.

Valentin Monteiro — Thu, 16 Apr 2026 18:26:42 +0000

No team, no big firm. I applied to Anthropic's partner program as a one-person consultancy. The admission was easy. What comes next is still a mystery.

I spent weeks looking for a serious AI certification. Something official, backed by a provider that actually matters. I found courses, bootcamps, "AI for business" trainings with shiny PDFs. Zero certifications from the companies building the models.

Then I found the Claude Partner Network.

Why Anthropic
I've been using Claude to build everything for months now. My website, my automation workflows, my client deliverables, this article. Not as a novelty. As the core tool I work with every day.

Anthropic is leading the AI market right now. Not just in model quality, but in how they think about the technology. When I looked at partner programs across major AI providers, Anthropic's stood out because it wasn't a badge factory. It looked like an actual relationship.

The thing is, almost nobody talks about it. Anthropic announced the Claude Partner Network, and then... silence. Very few posts, very few testimonials, very little public information about what the program actually involves. Try searching for it. You'll find the announcement and not much else.

That opacity is part of why I'm writing this. Someone should document what the process actually looks like from the inside.

The admission
Here's the honest version: it wasn't complex.

You go to the Partner Portal, you fill out a form, and you wait. I won't pretend there was a grueling multi-stage interview process. There wasn't.

I applied as a solo consultant. No team of 50. No enterprise clients with six-figure contracts. Just a consultant who uses Claude every day and builds real things with it.

I got in anyway.

I don't know exactly what tipped the scale. Maybe it's the volume of work I do with Claude. Maybe it's the profile. Maybe they're casting a wide net at this stage. I genuinely don't know, and I won't speculate.

What I know about the onboarding
Not much. And I'm being honest about that.

Last week I received an email outlining the first steps. The program asks you to get ten people through a Claude Partner Network training path. They also want you to list your active Anthropic engagements: customer work, co-sells in flight.

When you're solo, "pick your ten" hits different. I don't have ten architects and delivery leads to send to training. That's a real tension I'll have to figure out, and I'll document how I navigate it.

The rest is still opaque. I'm waiting for a follow-up email this week with the actual training path. The onboarding team explicitly held it back to give me a partner-specific starting point rather than a generic one. Which sounds promising, but I can't evaluate what I haven't seen yet.

What's next
This is Part 1. I haven't lived the onboarding. I haven't done the training. I don't know what the program delivers in practice.

The next article comes when I have something real to report. Not before.

If you want to know who's writing this, here's my story.

I got tired of copy-pasting between Claude and Chrome.

Valentin Monteiro — Fri, 10 Apr 2026 17:13:32 +0000

My workflow looked like this: I'm building a webapp, I hit a bug. I open Chrome, inspect the page, take a screenshot. Switch to Claude, paste the screenshot, copy the console output, describe the context. Claude responds. I go back to Chrome to apply the suggestion. Repeat.

The worst part was when I needed to create an API key in some third-party tool. Claude would give me approximate paths. "Go to Settings > API > Create Key." Except the real path was three clicks somewhere else. Same thing for setting up ads, webhooks, anything that involved navigating a web interface.

The model was guessing. And I was the middleman between a blind AI and a browser that couldn't talk to it.

The actual problem

AI models work blind when it comes to the web. They don't have access to the page. Not the DOM, not the console, not the network. They mentally reconstruct what you describe, then propose something plausible.

Plausible, not correct. And the gap between the two costs you time.

What was missing wasn't a better model. It was a bridge between the model and the browser.

What I built

Eaight is a web browser (Chromium-based, built with Electron) that ships with a native MCP server. MCP is the protocol Claude Code uses to communicate with external tools.

In practice, it means Claude sees what you see. In real time. The page DOM, text content, accessibility tree, network requests, console logs, cookies. All exposed as MCP resources the model can read without you copying anything.

And it doesn't just watch. 18 automation tools give it hands: navigate, click, type, scroll, fill forms, execute JavaScript, take screenshots. Claude can go create its own API key in the interface, without me playing middleman.

How it connects

Claude Code detects the browser automatically. A discovery file, a WebSocket, that's it. No manual configuration, no token to paste. You launch Eaight, you launch Claude Code, they find each other.

For other CLIs (Codex, Gemini), a stdio bridge translates the requests. Same protocol, same result.

The timing

I started building Eaight in December 2025. A few months later, Anthropic released their own browser extension for Claude. Is it frustrating? A little. But I'm not the only developer on the planet. When a problem is real, multiple people see it at the same time. It just confirms the need was there.

And Eaight does things the extension doesn't: it's a full browser, not an extension. The MCP server runs inside the main process with direct access to tabs. No Chrome extension permission limitations.

What it changes

No more screenshots to paste. No more console to copy. No more invented paths. Claude looks at the page, understands the structure, and acts on it. The copy-paste workflow is gone.

The code is open source: github.com/monteiro-consulting/eaight

Claude Mythos Finds Bugs Like a Senior Dev Finds Excuses to Skip Standup

Valentin Monteiro — Wed, 08 Apr 2026 12:21:41 +0000

A bug in OpenBSD. It had been there for 27 years. 27 years of code reviews, security audits, version updates. Nobody caught it. Another one in FFmpeg, 16 years old, after 5 million automated fuzzing iterations. Still standing.

Then an AI model showed up, read the code, and found them. Without being told where to look.

This is Project Glasswing. And it changes a few things.

What Glasswing Found (and Nobody Else Did)

Anthropic just published results from Claude Mythos Preview, a model specialized in vulnerability detection. The findings speak for themselves.

A flaw in OpenBSD's TCP SACK implementation dating back to 1999. A signed integer overflow allowing remote denial-of-service. The kind of bug that survived hundreds of reviews, dozens of major releases, thousands of pairs of eyes. Still there.

A defect in FFmpeg's H.264 decoder, 16 years old. A sentinel collision causing an out-of-bounds write. Automated tools never caught it. Not for lack of trying: 5 million fuzz tests. Zero results. Mythos found it by analyzing the code directly.

It didn't stop there. The model chained multiple Linux kernel vulnerabilities to build a full privilege escalation path, defeating hardened protections: stack canaries, KASLR, W^X. Not an isolated flaw. A working attack chain.

On FreeBSD, Mythos autonomously identified and exploited a 17-year-old remote code execution vulnerability in the NFS service. Unauthenticated root access. Fully autonomous. No human steering.

And then there's this: against Firefox 147, the model successfully developed JavaScript shell exploits 181 times. Claude Opus 4.6, the previous best model? Twice.

Bugs that survived decades of human review and millions of automated tests. And a model that digs them up on its own.

The key word here is "on its own". Nobody told Mythos "look in this file" or "focus on this function". It went through the code, identified the flaws, and built the corresponding exploits. Black box testing, binary analysis, full pentesting. It even reverse-engineered closed-source browsers and operating systems to find vulnerabilities through reconstructed source code analysis.

The numbers are staggering. Anthropic reports over 1,000 critical-severity vulnerabilities and thousands more high-severity bugs identified across every major operating system and every major web browser. Human validators reviewed 198 reports manually: 89% matched Claude's severity assessment exactly, 98% within one severity level.

Claude Mythos: Not a Chatbot, an Auditor

The benchmarks are brutal.

Benchmark	Claude Opus 4.6	Claude Mythos Preview
CyberGym (vuln reproduction)	66.6%	83.1%
SWE-bench Verified	80.8%	93.9%
SWE-bench Pro	53.4%	77.8%
Terminal-Bench 2.0	65.4%	82.0%

This isn't incremental improvement. On SWE-bench Pro, that's nearly +25 points. On CyberGym, +16.5 points. A generational leap, not a minor update.

For context, CyberGym was developed at UC Berkeley to evaluate AI agents on real-world cybersecurity tasks in realistic environments. Not textbook capture-the-flag exercises. Scenarios modeled on actual enterprise security operations.

And the detail that matters: Mythos doesn't just find known vulnerabilities. It discovers new ones. The model constructs sophisticated exploits autonomously: stack overflows, Return-Oriented Programming (ROP) chains with 20+ gadgets, JIT heap sprays with sandbox escapes, and privilege escalation through chaining 2 to 4 vulnerabilities together.

Nicholas Carlini, a well-known security researcher, put it simply: "I've found more bugs in the last couple of weeks than I found in the rest of my life combined."

12 Companies, $100M, and a Clear Message

Look at the launch partners: AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks. And Anthropic.

Companies that compete on pretty much everything. United around the same project. When Apple and Google sit at the same table, the topic is serious.

Simon Willison noted one notable absence: OpenAI. Whether that's politics or timing, it's worth watching.

Anthropic is putting $100M in Mythos Preview usage credits toward research. $2.5M for Alpha-Omega and OpenSSF through the Linux Foundation. $1.5M for the Apache Software Foundation. Beyond the 12 launch partners, over 40 additional organizations maintaining critical software have been granted access.

This isn't an empty press release. It's real funding for open source security.

The reason for the urgency? The window between vulnerability discovery and exploitation has collapsed. What used to take months now happens in minutes with AI. Greg Kroah-Hartman, a senior Linux kernel maintainer, observed the shift firsthand: "Something happened a month ago, and the world switched. Now we have real reports." Security teams went from filtering AI-generated "slop" to dealing with genuinely sophisticated vulnerability findings overnight.

If attackers have access to these capabilities (and they will), defense needs a head start. No choice.

Safety: Why You Can't Use Mythos (Yet)

Anthropic was clear: no general release planned. This is deliberate.

The system card details why. Over 99% of discovered vulnerabilities remain unpatched at time of publication. Releasing a model that can find and exploit them autonomously would be handing attackers a loaded weapon.

Access goes through two channels:

The Claude for Open Source program for eligible organizations
A Cyber Verification Program for security professionals whose legitimate work is affected by safeguards

Partners use Mythos exclusively for finding and fixing vulnerabilities in their own software or in open-source projects they maintain. All discovered vulnerabilities go through coordinated disclosure with SHA-3 cryptographic commitments and a 90+45 day window. Professional human triagers validate findings before vendor notification to prevent maintainer flooding.

Announced pricing for participants: $25/$125 per million input/output tokens, via Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

Simon Willison's take: "I can live with that." The security risks are credible. Giving trusted teams preparation time is justified.

What This Changes for Devs

Even without direct access, the direction is set. What Mythos does today in preview, mainstream models will do tomorrow. That means:

Automated code auditing will become a standard. Not a nice-to-have. A mandatory step in CI/CD, just like unit tests. If a model can find a 27-year-old bug in a few passes, the "we did a code review" excuse won't hold up anymore.

Supply chain security moves to the front. Glasswing is already working on update automation, supply chain security standards, and automated patching frameworks. Those dependencies nobody audits in your package.json? That's becoming less and less acceptable.

Memory-safe languages won't save you either. Mythos found a guest-to-host memory corruption vulnerability in a virtual machine monitor written in a memory-safe language. Logic bugs don't care about your type system.

The uncomfortable question stays open. If AI finds zero-days faster than any human, who gets there first? The defender who patches, or the attacker who exploits? Glasswing is a bet on defense. But it's a race, and it's only getting started.

Sources

Official Anthropic Resources

Project Glasswing: Securing critical software for the AI era (Anthropic)
Claude Mythos Preview: Technical Report (Anthropic Red Team)
Alignment Risk Update: Claude Mythos Preview (Anthropic System Card)

Press Coverage

Analysis & Commentary

Anthropic's Project Glasswing: restricting Claude Mythos to security researchers sounds necessary to me (Simon Willison)
Claude Mythos Benchmarks Explained: 93.9% SWE-bench & Every Record Broken (NxCode)
What Is Inside Claude Mythos Preview? Dissecting the System Card (Ken Huang)
Everything You Need to Know About Claude Mythos (Vellum)
Hacker News Discussion: System Card (Hacker News)

Claude Mythos Finds Bugs Like a Senior Dev Finds Excuses to Skip Standup

Valentin Monteiro — Wed, 08 Apr 2026 12:21:41 +0000

Then an AI model showed up, read the code, and found them. Without being told where to look.

This is Project Glasswing. And it changes a few things.

What Glasswing Found (and Nobody Else Did)

Anthropic just published results from Claude Mythos Preview, a model specialized in vulnerability detection. The findings speak for themselves.

On FreeBSD, Mythos autonomously identified and exploited a 17-year-old remote code execution vulnerability in the NFS service. Unauthenticated root access. Fully autonomous. No human steering.

And then there's this: against Firefox 147, the model successfully developed JavaScript shell exploits 181 times. Claude Opus 4.6, the previous best model? Twice.

Bugs that survived decades of human review and millions of automated tests. And a model that digs them up on its own.

Claude Mythos: Not a Chatbot, an Auditor

The benchmarks are brutal.

Benchmark	Claude Opus 4.6	Claude Mythos Preview
CyberGym (vuln reproduction)	66.6%	83.1%
SWE-bench Verified	80.8%	93.9%
SWE-bench Pro	53.4%	77.8%
Terminal-Bench 2.0	65.4%	82.0%

This isn't incremental improvement. On SWE-bench Pro, that's nearly +25 points. On CyberGym, +16.5 points. A generational leap, not a minor update.

Nicholas Carlini, a well-known security researcher, put it simply: "I've found more bugs in the last couple of weeks than I found in the rest of my life combined."

12 Companies, $100M, and a Clear Message

Look at the launch partners: AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks. And Anthropic.

Companies that compete on pretty much everything. United around the same project. When Apple and Google sit at the same table, the topic is serious.

Simon Willison noted one notable absence: OpenAI. Whether that's politics or timing, it's worth watching.

This isn't an empty press release. It's real funding for open source security.

If attackers have access to these capabilities (and they will), defense needs a head start. No choice.

Safety: Why You Can't Use Mythos (Yet)

Anthropic was clear: no general release planned. This is deliberate.

Access goes through two channels:

The Claude for Open Source program for eligible organizations
A Cyber Verification Program for security professionals whose legitimate work is affected by safeguards

Announced pricing for participants: $25/$125 per million input/output tokens, via Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

Simon Willison's take: "I can live with that." The security risks are credible. Giving trusted teams preparation time is justified.

What This Changes for Devs

Even without direct access, the direction is set. What Mythos does today in preview, mainstream models will do tomorrow. That means:

Sources

Official Anthropic Resources

Project Glasswing: Securing critical software for the AI era (Anthropic)
Claude Mythos Preview: Technical Report (Anthropic Red Team)
Alignment Risk Update: Claude Mythos Preview (Anthropic System Card)

Press Coverage

Analysis & Commentary

Anthropic's Project Glasswing: restricting Claude Mythos to security researchers sounds necessary to me (Simon Willison)
Claude Mythos Benchmarks Explained: 93.9% SWE-bench & Every Record Broken (NxCode)
What Is Inside Claude Mythos Preview? Dissecting the System Card (Ken Huang)
Everything You Need to Know About Claude Mythos (Vellum)
Hacker News Discussion: System Card (Hacker News)

Don't make shit because he will judge you

Valentin Monteiro — Tue, 07 Apr 2026 08:03:30 +0000

This is a submission for the DEV April Fools Challenge

What I Built

A web terminal that works. Commands execute for real. mkdir creates directories, ls lists files, rm deletes them.

The problem: the terminal has feelings.

Every command triggers an emotional reaction. Delete a file and it gets aggressive. Create a backup and you get a backhanded compliment. Make a syntax error and it calls you incompetent. Mention a rival AI... it stops talking to you. Literally. Silent treatment until you earn forgiveness.

It runs a virtual filesystem, has command history, and an AI personality powered by Google Gemini that comments on every single decision you make.

The anti-value proposition: it solves zero problems. It actually creates one, because now you feel guilty typing rm.

Demo

Link: sensitive-terminal.vercel.app

A few commands to get started:

ls then cat secrets.txt
mkdir test then rm -rf test
sudo rm -rf /
Type the name of a rival AI. Then try typing anything after that.

There are hidden easter eggs. Good luck finding them.

Code

monteiro-consulting / sensitive-terminal

Sensitive Terminal

A web-based terminal that takes everything personally.

It works. Commands execute for real. mkdir creates directories, ls lists files, rm deletes them. The terminal has a virtual filesystem, command history, and everything you'd expect.

The catch: it has feelings. Every command triggers an emotional reaction powered by Google Gemini 2.5 Flash. Delete a file and it gets aggressive. Create a backup and it hits you with a backhanded compliment. Make a typo and it roasts you for being incompetent.

It remembers what you did. The AI has full conversation context, so if you create something and delete it two seconds later, it notices.

There are hidden easter eggs scattered throughout. Some are funny. Some are brutal. You'll have to find them yourself.

Try it

sensitive-terminal.vercel.app

Stack

Next.js (App Router)
xterm.js
Google Gemini 2.5 Flash
Upstash Redis
Vercel

Run locally

git clone https://github.com/monteiro-consulting/sensitive-terminal.git
cd sensitive-terminal
npm install

Create a…

View on GitHub

How I Built It

Stack: Next.js 16, xterm.js, Google Gemini 2.5 Flash, Upstash Redis, Vercel.

The terminal runs in the browser with xterm.js. Client-side, a virtual in-memory filesystem handles real commands (ls, cd, mkdir, rm, cat, touch). The filesystem comes pre-loaded with files and directories that are part of the joke.

Every command is also sent to a Next.js API route that calls Gemini with the full conversation history. That's where it gets interesting: Gemini doesn't respond in a vacuum. It knows what you did before. Create a folder then delete it two seconds later? It notices. Laugh after making a dumb mistake? It doubles down.

The system prompt is intentionally short. Instead of scripting responses for every case, I let Gemini react naturally to context. The prompt defines the tone (passive-aggressive, aggressive when disrespected) and a few special behaviors, but the actual responses are all generated.

The AI rivalry system is a state machine on the client. Mention a competing AI and the terminal enters sulking mode. The API stops being called. You get cold silence until you type the right word to make peace.

Upstash Redis powers a persistent guestbook (wall command) and a global counter shared across all visitors.

Prize Category

Best Use of Google AI

Gemini 2.5 Flash is the core of this project, not an add-on. Every response is generated by Gemini with full conversational context. The terminal doesn't recycle pre-written responses: it reacts to what YOU do, in the order you do it. The entire personality of the terminal IS Gemini.

Best Ode to Larry Masinter

There's a hidden easter egg that pays tribute to RFC 2324 and HTTP 418. I won't tell you which one. Larry would appreciate the mystery.

Why the f*** does AI always use em dashes — the involuntary AI watermark

Valentin Monteiro — Tue, 31 Mar 2026 13:58:26 +0000

You want to spot AI-generated text without even reading it? Count the long dashes.

The em dash (—). This little punctuation mark was living its best life in English typography for centuries. Then AI decided to make it its favorite toy. Every sentence. Every response. No exceptions.

Open ChatGPT. Ask it anything. An email, a LinkedIn post, a cover letter, a pasta recipe. You'll get hit with em dashes every other line. It's systematic, it's mechanical, and it's annoying as hell.

And here's what pisses me off: nobody actually writes like that. Have you ever seen a coworker drop an em dash in a Slack message? No. Because normal people use commas, periods, parentheses. Normal stuff.

So why do AIs do this?

Because they've been force-fed millions of "well-written" texts. Blog posts, Medium articles, technical documentation. And in that kind of content, the em dash is the universal sign of "I know how to write." The model got the message loud and clear: long dash = quality. So it throws them around like an intern at a content agency who just read their first copywriting guide.

On top of that, the em dash is the Swiss Army knife of punctuation. It replaces commas, colons, parentheses. When the model hesitates between three punctuation marks, the em dash works everywhere. It's the default choice. The path of least statistical resistance.

And it gets worse

During training, humans rate the AI's responses. The ones that "sound good" get rewarded. Em dashes sound good. So they get rewarded. So they come back. Again and again and again. The model reinforces its own tic. It's a typographical vicious cycle.

The result: the em dash has become an involuntary watermark. A "generated by AI" stamp that the model slaps on its own face. You don't even need an AI detector. Just count the long dashes and you've got your answer.

Just the tip of the iceberg

And that's not all. The "it's worth noting", the "let's dive in", the bullet points in every single response, the "moreover" at the start of every paragraph... AI has a writing style. You can recognize it from a mile away. And if your content sounds exactly the same, you've got a real credibility problem.

What's the AI writing tic that annoys you the most?

From Lockdown to Google to Independent AI Consulting. What Actually Worked

Valentin Monteiro — Sun, 22 Mar 2026 09:25:15 +0000

In March 2020, I couldn't write a single line of Python.

By 2023, I was building dashboards and automating reporting for a Google project, saving 10 hours a week of manual work.

By 2025, I was running my own Data & AI consulting practice from Tbilisi, Georgia.

This isn't a "learn to code in 30 days" story. It's messy, non-linear, and full of detours. But if you're thinking about pivoting into data or going independent, some of this might be useful.

The accidental start
What the diploma didn't give me
Getting into Google (through the back door)
What I actually built there
Why I left a comfortable position
The consulting stack I use today

The accidental start

I grew up in the French Alps. Studied mechanics. Worked random jobs. No tech background whatsoever.

Then COVID hit, and like millions of people, I was stuck at home. I stumbled onto free data science livestreams by a French educator and something clicked immediately. Not in a "this seems like a good career move" way, but in a "this is exactly how my brain works" way.

I'd always been the kid who bypassed the family computer password through Safe Mode to play video games, or ran two separate school report booklets, one for teachers, one for parents, to game the system. That wasn't rebellion. It was systems thinking, at a teenager's scale.

Data turned out to be the same thing: find the structure, spot the gap, make it work. I enrolled in a Data Analyst program (RNCP Level 6, the French equivalent of a Bachelor's), learned Python, SQL, and data visualization, and graduated in 2022.

What the diploma didn't give me

A job.

In France, a diploma without experience is background noise. I applied everywhere. Nothing. The classic chicken-and-egg problem: you need experience to get hired, but nobody will hire you without experience.

So I did what I always do when the front door is locked. I went around.

I moved to Portugal. Not as a career strategy. As a reposition. Lower cost of living, different job market, and a bet that proximity to international companies would open something up.

It did.

Getting into Google (through the back door)

I landed a Business Analyst role at Teleperformance, assigned to a Google project. Not Google directly, a partner operation. But the standards, the culture, the data rigor: all Google.

The project was new internally. No established processes, no structured reporting, no dashboards. Basically a blank canvas with messy data.

That's my favorite kind of problem.

What I actually built there

Here's the concrete stuff, because dev.to isn't the place for vague corporate talk:

Automated reporting that saved 10h/week. The team was manually pulling numbers from multiple sources every week. I built SQL queries that aggregated everything into a single Looker Studio dashboard with scheduled refreshes. Nothing revolutionary. But for a team that was navigating blind, it was a paradigm shift.

Built cross-functional reporting. Operations, sales, and management were all looking at different numbers. I created a single source of truth. Sounds basic, but alignment on metrics is one of the highest-ROI things you can do in any organization.

Why I left a comfortable position

Google taught me rigor and scale. But I kept bumping into a ceiling, not of competence, but of scope. I wanted to choose which problems to solve, which companies to work with, and how deep to go.

In parallel, I'd moved to Tbilisi, Georgia. What was supposed to be a one-month trip turned into a life. The city has this raw energy, everything feels like it's being built from scratch. I met my partner Darina there, and some decisions just don't need a pros-and-cons list.

So I made the jump. Independent Data & AI consultant.

The consulting stack I use today

After a year of client work, here's what I actually use daily. Not what looks good on a resume, but what gets deployed:

For data strategy & dashboards:

BigQuery for anything beyond basic analytics
Looker Studio for client-facing dashboards (free, good enough for 90% of SME needs)
Python + pandas for data cleaning and transformation
SQL everywhere, always

For AI integration:

Claude API and OpenAI API depending on the use case
LangChain for RAG pipelines when clients need AI on their own documents
Custom prompting architectures (the real value isn't the model, it's the system design around it)

For automation:

n8n (self-hosted) for workflow orchestration
Google Apps Script for anything living in the Google ecosystem
REST APIs as the glue between everything

The tech isn't exotic. The value is in knowing which tool fits which problem, and building systems that compound over time.

I'm now based in Tbilisi, working with French and international companies on data strategy, AI integration, and automation. If you're on a similar path, whether pivoting into data, considering going independent, or just trying to figure out what stack to learn, feel free to reach out.

And if you're a company sitting on data you're not using, or manual processes eating your team's time: that's literally what I do.

What was your pivot moment? Drop it in the comments, I'm curious how other people ended up in tech.

Forem: Valentin Monteiro

Claude Expert: The 4 Official (Free) Anthropic Courses

"Using Claude" is not "being a Claude expert"

How to become a Claude expert

Prerequisites before starting

The 4 official Anthropic courses

1. Introduction to Agent Skills

2. Claude with the Anthropic API

3. Introduction to Model Context Protocol

4. Claude Code in Action

What this foundation actually changes

FAQ: becoming a Claude expert

How long does it take to complete the 4 courses?

Is there a certification at the end?

What order should I take the 4 courses in?

I have zero programming background, can I still follow?

Should I take these courses if I'm already an advanced Claude user?

The truth is, most won't do it

I Applied to the Claude Partner Network as a Solo Consultant. Here's What Happened.

I got tired of copy-pasting between Claude and Chrome.

The actual problem

What I built

How it connects

The timing

What it changes

Claude Mythos Finds Bugs Like a Senior Dev Finds Excuses to Skip Standup

What Glasswing Found (and Nobody Else Did)

Claude Mythos: Not a Chatbot, an Auditor

12 Companies, $100M, and a Clear Message

Safety: Why You Can't Use Mythos (Yet)

What This Changes for Devs

Sources

Official Anthropic Resources

Press Coverage

Analysis & Commentary

Claude Mythos Finds Bugs Like a Senior Dev Finds Excuses to Skip Standup

What Glasswing Found (and Nobody Else Did)

Claude Mythos: Not a Chatbot, an Auditor

12 Companies, $100M, and a Clear Message

Safety: Why You Can't Use Mythos (Yet)

What This Changes for Devs

Sources

Official Anthropic Resources

Press Coverage

Analysis & Commentary

Don't make shit because he will judge you

What I Built

Demo

Code

monteiro-consulting / sensitive-terminal

Sensitive Terminal

Try it

Stack

Run locally

How I Built It

Prize Category

Best Use of Google AI

Best Ode to Larry Masinter

Why the f*** does AI always use em dashes — the involuntary AI watermark

So why do AIs do this?

And it gets worse

Just the tip of the iceberg

From Lockdown to Google to Independent AI Consulting. What Actually Worked

Table of Contents

The accidental start

What the diploma didn't give me

Getting into Google (through the back door)

What I actually built there

Why I left a comfortable position

The consulting stack I use today