Thiago Temple

@thitemple.me

👨‍💻 Staff Developer @Shopify | Building indie apps on the side 🎲 Board game nerd | Dad of 3 | Brazil → Canada 🛠 20+ years in dev — sharing what’s worked for me, not what you should do https://ttemp.link/me

As much as models are great, and they trully are these days, it still takes skill to ship good quality code. Just tried asking GPT 5.4 to prototype a new UX feature in an app. It was very off from the target of what I tried to describe, not even using the current design system.

As I get older I'm coming to increasingly radical views like "you have to do things to get good at them" and "you have to think about problems to solve them"

This was a good read, great to see someone with inside information about it. I should share my reasons for why I chose Svelte, but right now they align with great vibes and DX.

Mainmatter@mainmatter.com · 6mo ago

What makes Svelte the right choice for modern frontend projects? And when is it not? In this blog post, @paolo.ricciuti.me breaks down the advantages and limitations of Svelte, and what that means for your next product decision. Read the post 👉 mainmatter.com/blog/2026/02... #svelte #sveltekit

I must be really bad at this OpenClaw thing. Everybody seems to be building companies overnight and I'm struggling to make an agent give me straight answers on how I can improve my product.

When I fix a prompt, it helps once. When I fix a boundary definition in my project docs, it helps on every future task. That's the real shift with AI coding — spend less time re-explaining context, more time refining the structure the agent works within

"Write better prompts" is a dead end if "better" has no definition. The real leverage isn't in the prompt — it's in giving the AI structured boundaries to compare its output against. Define what good looks like, and consistency follows.

I’m 100% sold on the benefits of adding skills for AI agents, they add a small cost to the context without the overhead of starting subagents. eg: I added a skill to « svelte-developer » to consult the docs using the mcp and to not use on:click style event handlers. Done. That is solved

I spent almost the whole day starting threads in the new Codex app. I'm in the $20 plan, and I only reached the limit by the end of the day. I guess the 2x credits helped. I almost finish we building a URL shortner app. Will be done with it tomorrow.

Write first pass with @claudecode.bsky.social , reviews the code, "I was absolutely right", asks GPT to review it and find a swiss cheese of issues. I still think GPT codex 5.2 on high is the best at being detailed and less error prone.

Thiago Temple@thitemple.me · 7mo ago

Started working on a restaurant related side-project. My brain went straight to try to solve the data model for the domain (using @convex.dev). Of course, it is AI assisted, but I wonder what's best with AI these days. To try to build a robust first version of it, or iterate and "fix" along the way

Started working on a restaurant related side-project. My brain went straight to try to solve the data model for the domain (using @convex.dev). Of course, it is AI assisted, but I wonder what's best with AI these days. To try to build a robust first version of it, or iterate and "fix" along the way

The more I dive into the details of @svelte.dev and SvelteKit the more I enjoy using it. It has been my framework of choice for any personal projects for a while but I'm not even looking at anything else right now. I was trying to get the details of all the possible page options for my next YT video

Building a new app to work new brain-muscles, this time an app to generate quizes for online courses. The new thing for me is using @convex.dev. So far the best part has been the great docs specifically the AI Docs feature. Nice onboarding!