AI Copilots and the Dev Journey: Claude, Gemini and the Rest

I’ve been passionate about tech for over 10 years. I love the craft of building things — the problem solving, the architecture decisions, the moment when something finally clicks. That hasn’t changed. What has changed is the tooling.

AI coding assistants have genuinely shifted how I work, and after using them daily across real projects, I have opinions. Not hype. Opinions.

Table of contents

• The Big Three
  • Claude (Anthropic)
  • Gemini (Google)
  • ChatGPT / Codex (OpenAI)
• What AI Copilots Actually Unlock
  • A Full CMS in Under 25 Hours
  • A Video Game with Complex Interactions
  • The Meta-Article Experiment
• Stop Reading. Start Creating.
• The Future: Claude Mythos and What’s Coming
  • What Happens to Junior Developers?
• My Honest Take

The Big Three

The market has consolidated around three main players for developers. Each has a different flavor.

Claude (Anthropic)

Claude is the one I reach for when I need to think through a problem, not just get code generated. It’s strong at reasoning, explains why behind solutions, and handles large context windows well — useful when you dump an entire file and ask “what’s wrong here.” It also pushes back when you’re wrong, which I respect.

Claude Code (the CLI/agentic version) is particularly interesting: you can give it a task and let it navigate files, run commands, and make changes autonomously. That’s a different category from chat-and-paste.

Gemini (Google)

Gemini has the advantage of Google’s ecosystem. If you’re working with GCP, Firebase, or anything Google-adjacent, the integration is tight. Gemini 2.5 Pro with its massive context window is genuinely competitive for large codebase work. It’s also baked into Android Studio and VS Code if that’s your setup.

ChatGPT / Codex (OpenAI)

ChatGPT is where most people started, and for good reason — it’s versatile and the ecosystem is huge. Codex powers GitHub Copilot, which is still the dominant autocomplete tool in the market. If you want tab-completion inline in your editor, Copilot is the benchmark to beat.

The tools are getting closer together with each release. Pick the one that fits your workflow and use it seriously — that matters more than which one is technically “best” at any given moment.

What AI Copilots Actually Unlock

I’ll skip the theoretical. Here are real things I’ve done with AI assistance that would have taken significantly longer — or that I wouldn’t have attempted alone.

A Full CMS in Under 25 Hours

I built a complete CMS connected to an external API in under 25 hours. From scratch. With an admin UI, content modeling, authentication, and a live preview.

Without AI, a project like that is a multi-week effort. With it, the work shifts: instead of hunting documentation and writing boilerplate, you’re making architectural decisions, reviewing generated code, and testing. The AI handles the repetitive scaffolding; you handle the judgment calls.

This isn’t about AI doing the work for you. It’s about collapsing the gap between “I know what I want to build” and “I have something running.”

A Video Game with Complex Interactions

I prototyped a video game with non-trivial interactions — collision systems, state machines, entity behaviors. Game dev is a domain I’m comfortable around but not deeply expert in.

The AI let me move fast through the parts I knew less well, while I focused on the design and feel. You still need to understand what the code is doing. But you don’t need to have memorized every API to ship something real.

The Meta-Article Experiment

This one is a bit funny. I asked Claude to update this very site, then write this article — in both French and English — about itself, and generate the OG image for it. Autonomously.

It worked. Not perfectly, but it worked. The article you’re reading went through a human revision pass (clearly, since Claude wouldn’t call itself out the way I’m doing right now), but the first draft came from a prompt.

That’s a strange experience to live through as a developer. You’re both the author and the director. The line gets blurry.

Stop Reading. Start Creating.

The biggest shift AI assistants enable isn’t raw speed — it’s lowering the activation energy to start something new.

The mental barrier that said “I don’t know enough to start that” has mostly disappeared. Now you know that when you hit something unfamiliar, you won’t be stuck for long. That changes which projects feel within reach.

Concrete examples:

• Setting up a CI/CD pipeline you’ve never touched? Describe what you need, iterate from the generated YAML, understand what each step does before you merge it.
• Unfamiliar with database migrations? Start the project, hit the first error, ask for an explanation — not just a fix.
• Want to try a new language or framework? Build the first small thing with AI assistance, then read the docs for the parts that don’t make sense. Learning by building is faster.

The key word is understand. AI assistants generate code faster than you can review it if you’re not careful. Slow down enough to know what’s happening, or you’ll accumulate debt you can’t debug.

The Future: Claude Mythos and What’s Coming

Anthropic and its competitors are on aggressive release cycles. Each generation brings better reasoning, more context, and more agentic capability — the ability to take actions, not just generate text.

Names like Claude Mythos are floating around as next-generation candidates. The trajectory points toward models that can hold the full context of an entire codebase, understand the project history, and proactively identify issues. Less “answer this question” and more “review my PR and flag anything that looks wrong.”

That’s genuinely exciting. But it raises a question I don’t have a comfortable answer to.

What Happens to Junior Developers?

I’m going to be honest here, because I find this genuinely concerning.

A senior developer with AI copilots can now take on significantly more work. Realistically — 3x, 5x the throughput on certain types of tasks. So the question becomes: why hire a junior developer when a senior can absorb that workload?

That’s not a hypothetical. It’s a calculation companies are already making. Entry-level positions in tech are drying up. Bootcamp grads are struggling to find placements. Junior roles that used to exist as training grounds are getting harder to justify on a headcount spreadsheet.

The classic career path — junior → mid → senior — assumed that junior work was its own category of value. AI is blurring that line.

I don’t have a clean answer. Maybe junior developers will specialize in things AI is bad at: interpersonal skills, stakeholder management, domain knowledge that’s hard to articulate in a prompt. Maybe the path to seniority just gets steeper and more competitive. Or maybe the market corrects and new types of roles emerge.

What I’m sure of: if you’re early in your career, relying on AI tools to compensate for depth is a trap. The people who will thrive are the ones building genuine understanding and using AI to move faster — not the ones using it as a crutch to avoid learning.

My Honest Take

These tools are force multipliers, not skill replacements. They take what you know and let you do more with it. They’re most useful when you already have the foundation.

Compared to each other:

• Claude — my go-to for reasoning-heavy tasks, complex debugging, and agentic workflows
• Gemini — solid choice for Google ecosystem work and large context analysis
• ChatGPT / Copilot — ubiquitous, great for autocomplete and general-purpose queries

All three hallucinate. All three are confidently wrong sometimes. Keep your judgment active.

The dev journey isn’t shorter with AI — it’s different. The parts that require craft, decision-making, and genuine understanding are still yours. The parts that required memorization, repetitive scaffolding, and documentation hunting are largely automated now.

That’s the deal. Use it.

Stop reading about what’s possible.

Start building it.