Skip to main content
AI Jul 7, 2026 · 62 views

Copilot vs Claude: What I Actually Learned After Using Both Every Day

A breakdown of how Copilot, Claude Code, and Claude Cowork each earned distinct roles in my workflow, from fire‑and‑forget GitHub tasks to deep, iterative coding and personal assistant automation.

AI Trends
Copilot vs Claude: What I Actually Learned After Using Both Every Day

I said I was leaving Copilot. I didn’t. Here’s what I actually learned.

A month ago I wrote what was basically a breakup letter to Copilot. I was frustrated, convinced I was about to be hit with a usage bill far higher than I expected, and ready to jump ship entirely.

But I didn’t leave.

I’m still paying for Copilot. I’m still using it. And I’m also using Claude Code, a tool I’ve grown to trust in ways I didn’t anticipate. The real story isn’t “I switched,” it’s that I finally stopped pretending these tools were interchangeable. They aren’t. They do different jobs, and they earn different kinds of work.


Claude Code: The coder I actually trust

I don’t know if it’s placebo, model quality, or just the way it interacts with my workflow, but Claude Code feels noticeably more reliable as a coding partner.

With Sonnet 4.6, Sonnet 5, or Opus 4.8, I can hand it a task and walk away with a reasonable expectation that when I come back, the work will be done to a good standard. Not perfect, not magical, but solid. It feels like working with someone who learns from the environment, adapts, and doesn’t repeat the same mistake twice.

Copilot, by contrast, often struggles with the basics of operating inside my workspace. I’ve had it fail to run commands because of UNC path issues, then get stuck in loops retrying the same command over and over instead of applying the workaround it already discovered. Claude doesn’t do that. It adjusts. It remembers. It moves forward.

That difference... the sense that Claude is learning within the task... is what’s earned my trust.


The one thing Copilot still does better

And yet, I haven’t cancelled my Copilot subscription.

Because Copilot still nails one thing Claude doesn’t: fire‑and‑forget GitHub issue assignments.

There’s a very specific kind of productivity Copilot gives me:

  • I write a clear issue on GitHub.
  • I assign it to Copilot.
  • I close my laptop.
  • I go live my life.
  • Thirty minutes later — or the next morning — there’s a PR waiting.

It’s rarely ready to merge. It often needs significant cleanup. Sometimes it needs multiple re-prompts. But the feeling of explaining a problem from my phone, assigning it, and then forgetting about it until a chunk of work appears... that’s valuable.

Especially on busy days when I’m away from my PC, juggling commitments, and trying to keep momentum on my projects. Copilot integrates with GitHub in a way that fits into the cracks of my life. That’s why the $10/mo stays.


Claude Cowork: The missing layer I didn’t know I needed

The biggest surprise of this whole journey wasn’t Claude Code — it was Claude Cowork.

I was hesitant at first. I’ve spent most of the AI era building my own personal assistant tools: systems for tracking events, goals, courses, fitness, diet, and incoming messages from my personal and corporate websites. These all feed into a web-based assistant that sends me a morning digest with updates and a “phrase of the day” in whatever languages I’m dabbling in.

I thought I already had a good setup.

But Cowork turned out to be the missing layer.

By adding each of my existing tools into Cowork: some via custom MCP servers, and others via standard REST APIs, I suddenly had a unified assistant that could actually manage my day-to-day life. Not just surface information, but act on it. Not just store context, but use it.

It’s the first time I’ve felt like my personal assistant system wasn’t something I built for the AI, but something the AI could genuinely work with.

That (and setting up those fiddly custom MCP servers!) deserves its own blog post, because the learning curve, and the payoff, has been huge.


Where I’ve landed

I didn’t switch. I specialised.

Claude Code
My in‑the‑loop coding partner. Iterative. Contextual. Adaptable. The one I trust for complex tasks where judgement matters.

Claude Cowork
My personal assistant layer. The glue between my tools. The thing that finally made my long-running personal systems feel cohesive.

Copilot
My fire‑and‑forget GitHub agent. Perfect for small, well-scoped tasks I want progress on even when I’m away from my PC.

Three tools. Three different jobs. No replacements, just roles.

This is the workflow I’m settling into. And honestly, it feels good.

If you want to compare notes, I’m @anderslearmonth.bsky.social on Bluesky.

Share this post

Help others discover this content

© 2026 Anders Learmonth. All rights reserved. This article is original content and may not be reproduced, copied, or republished without written permission. Please link back with attribution if you reference or quote this work.

Keep reading

Related Posts

Perfect is the enemy of good enough
AI Mar 30, 2026

Perfect is the enemy of good enough

Embracing "Good Enough": How AI is Revolutionizing My Creative Process

Read more
Embracing Github Copilot to reduce tech debt in personal projects
AI Apr 17, 2026

Embracing Github Copilot to reduce tech debt in personal projects

The whole is greater than the sum of its parts, and AI is allowing me to finish off more parts than ever.

Read more
I've still never tried Claude Code
AI Apr 22, 2026

I've still never tried Claude Code

Am I the only person not using Claude?

Read more

Enjoyed this post?

Get more like this
in your inbox

No spam, ever. Just occasional updates when I publish something worth reading.

No spam, ever. Unsubscribe any time.

This site uses essential cookies to keep you securely logged in and to protect against CSRF attacks. No tracking or advertising cookies are used. By continuing to use this site you consent to these essential cookies. Read our Privacy Policy for full details.

Learn more