Not all models are trained equal, and you should acknowledge that. Some are better than others depending on the workload. Part of the agentic engineering skill is knowing what model to use for what, exactly like how some engineers are stronger in some areas than others.
The process is simple. Try different models for different workloads. Planning, UI, backend, ops, image generation. See where each one shines and pick your favorites.
The other knob worth tuning is the effort level.
Here’s my current combo:
<ul>
<li>GPT 5.5 High: my default for non-UI coding. Backend, CLIs, data work, even frontend as long as it doesn’t involve designing the UI directly. The moment I let it touch actual UI design, it generates slop time after time. Learned that one the hard way.</li>
<li>Opus 4.8 High: the best I’ve tried for design work. I also lean on it as my general-purpose model, mostly for cost. I never hit the rate limit on my Claude subscription, whereas Codex always does since I keep it busy.</li>
</ul>
What’s your combo?
<hr>
In case you missed, <a href="https://daily.dev/posts/1As9RA7f4" target="_blank" rel="noopener nofollow">tip #1 is here.</a>

Everything about agentic engineering and vibe coding 🤖
Tools, tips, tricks, workflows, best practices, you name it.
Let the vibes guide you

Agentic Engineering

A practical tip on model selection for agentic coding workflows. The author recommends matching AI models to specific workloads: GPT 5.5 High for backend, CLI, and data work, and Opus 4.8 High for UI/design tasks. Key insight is that effort level is also a tunable parameter, and different models have different strengths depending on the task type.

Agentic Tip #2: Pick your models

Sonnet for most of the coding tasks- much cheaper

Opus+sonnet is the primary choice for backend development. If it fails to achieve the desired results, switch to Antigravity Designs. Claude is not a suitable option.