@diysoftware

Building software to solve my everyday problems

lioloc.com
Joined March 2024
  • Tweets1.1K
  • Following160
  • Followers229
184Media
One AI plan cost me $6.50 and 310K tokens because my prompt was vague. What fixed my workflow: 1. Refactor first so file names explain themselves 2. Smart model decides, cheap models edit and execute Anything I missed?
17
Have you tried Gemini Argon recently? How does it compare with Opus 5.5 / Astra?
28
I ran 20 subagents for 4 hours on my meal prep app Result: unlabelled icons, broken UX, and clicking on recipes crashed my app. How do you one shot your development?
2
1
49
I got tired of repeating the same cleanup prompts to my AI agent, so I built a global skill in pi.dev with: 1. JSDoc only, no inline comment slop 2. No 1-letter vars (even in .map / .filter) 3. Blank lines between logical steps Do you use clean coding rules?

Pi Coding Agent

A terminal-based coding agent

pi.dev
3
1
110
I made a video explaining how I built this skill, dm me if you want it
22
Here's how I use Opus 5.5 without session anxiety: pi.dev + openrouter (deepseek) + claude (opus 5.5) OR my fav: pi.dev + openrouter (opus 5.5) + ollama (GLM 5.3 flash) What do you use?

Pi Coding Agent

A terminal-based coding agent

pi.dev
1
1
77
10. Having software engineering fundamentals dramatically increases speed, when things have gone horribly wrong, you'll typically know before it hits production
1
2
This is a small taste of what I've learned so far. I recently fixed up the backend slop AI generated initially and I talk about it in my newest video on YT, name of the channel is diysoftware
11
7. Use service files to auto boot your app on your machine if you're not using something like Coolify to handle deployment
1
10
more replies
9. Don't delegate your brain to your AI, if it comes up with a plan, don't just blindly ok it. AIs are typically overeager nowadays.
1
3
8. You don't need a self hosted Coolify isntance to manage your deployments, as my backend engineering friend once told me, it wastes a lot of resources. Just ask your ai to help you set up your deployment.
1
9
6. Use Tailscale to make sure your private project never touches the public internet. Even if you do plan to make it public, use tailscale to connect to your database if you're self hosting one.
1
6
4. Go with a framework you are familiar with from the start and that you like over a popular one. You will be more invested in the project that way and will have a better "mental map" of your project as well.
1
13
5. If you're using a separate raspberry pi or vps machine to host your project, you don't have to setup github on it, just ask AI to create a deploy script with ssh access that you run directly on your main computer.
1
15
3. If your project uses a database, make sure to ask your agent to use a framework that supports migrations from the start, you don't have to know what it does, just know that your database will thank you.
1
4
Last week I vibe coded my an app to solve my meal prepping needs, here's 10 things I learned:
1
29
1. Vibe coding works great for small projects. The more it grows, the more things silently break. 2. GLM 5.3 flash is surprisingly good and significantly cheaper than Opus 5.5.
1
34
Here's how I solved my meal prepping needs
24
liolocs retweeted
TANSTACK
@tan_stackOct 5
TanStack Charts 1.0 is here 🎉 Small bundles, fast updates, and charts you don’t have to outgrow. Start with a line and see where it takes you, or just simply tell your agent to "use Tanstack Charts"! tanstack.com/charts
40
155
24
2,084
137,314
TypeSafe.ai got adopted faster than ChatGPT ! Here's how to skip the waitlist: 1. classifier(dot)dev 2. Or use clef through ollama: ollama pull clef Have you tried a decision model yet?
23
Load more