Jibran’s Perspective

I keep coming back to Amp Code

My first thread in Amp is from 7 months ago, July 2025. Back then, Claude Sonnet 4 was the model used by Amp’s smart mode. Today that is Claude Opus 4.6. Much has changed in that time - the models have become smarter, and I’ve leaned into AI assisted software development more.

What remains constant is how much I like the Amp experience - even after honestly trying out Claude Code & Codex (OpenAI). I’m not the only one. Here’s a Tweet linked from their home page.

I’ve been using Claude Code (Max $100/m plan) since Jan 2026, and recently started using Codex ($20/m plus plan with the 2X usage limits offer). I’ve used Opus 4.6 & GPT-5.3 Codex since they launched. They are very capable models.

I like the Codex app - especially the diff UI & the ability to comment on changes inline and have the agent act on those.

Despite that, when I started comparing the 3 on the same tasks, I consistently enjoyed working with Amp better.

Amp is more expensive - I spent $20 over 2 days in Jan and was on track to spend $10/day based on how much I was using it. That’s when I switched to the Claude Max $100/m plan, and I’ve used it a lot since then without having to worry about additional costs.

Yet I want to start using Amp again, even if it’s more expensive.

I can’t point to one thing that makes Amp feel better. It’s a combination of its speed and the way it interacts with me. It just feels like a better tool.

In no order, here are the reasons why I started out and stuck with Amp.

• Amp code has usage based billing. This was the big one for me. I couldn’t bring myself to pay a monthly subscription when I wasn’t using agents for coding that frequently.
• When I discovered them, they talked up their philosophy of working with the agent instead of having it just do stuff. This was most visibly demonstrated by their VSCode extension (the CLI came much later) using the enter/return key for new lines by default. Every other extension defaulted to sending the prompt on enter, and you had to use shift+enter to add a new line. This one was big for me. It meant that they expected you to add the right context upfront instead of YOLOing it.
• The people behind Amp, and their communications, radiated a love of the craft. They didn’t “move fast & break things”. They move deliberately, and they remove features that don’t seem to be working.
  • Offer a limited set of models to choose from. Those models however are very well integrated, both with the agent & each other
  • Removed Bring-Your-Own-Keys
  • Removed Compaction
  • Removed Amp Tab (auto-completion in VSCode)
  • and recently it seems they are going to deprecate the VSCode extension as they feel the CLI is the better UI for a coding agent (I agree).

Amp just offers a more curated/opinionated experience from people whose taste I share. I’m going to start using it again.

Given the value I’m getting out of AI assisted software development, I’m thinking of sticking with Amp until the costs start hurting, which feels like it would be around the $200/m point.

Claude & Codex are very capable and get the job done equally well, but the experience of Amp is just better.