How to Code With AI · Part 4 · Engine Selection · Lesson 4.1

Every model has a personality. Learn the room.

Top vibe coders pick the engine for the job, not the job for the engine.

The lesson

The frontier models are not interchangeable. Each has a personality that shows up reliably across thousands of sessions. Knowing the personality of the room lets you pick the right engine for the move you're trying to make.

CLAUDE (Opus, Sonnet) — strongest at thoughtful refactors, careful edits to existing code, multi-file reasoning, asking clarifying questions. Weakest at cute one-shot tricks. Pick when modifying a real codebase.

GPT (GPT-5, GPT-5 Pro) — strongest at clever solutions to bounded problems, math-y code, one-shot algorithm implementations. Weakest at knowing when to ask. Pick for tight, well-specified problems.

GEMINI (2.5 Pro, Pro Ultra) — strongest at long context, structured data extraction, multimodal moves. Weakest at sustained focus on small change. Pick when ingesting a lot.

GROK — less-filtered alternatives, willing to try unusual approaches. Pick for a second opinion that disagrees.

LOCAL (Llama, DeepSeek, Mistral) — speed, privacy, deterministic short tasks. Pick for volume work or air-gapped.

The pros don't pick a favorite. They pick the right one for the move.

Example

✓ Three-engine workflow

"Use Gemini Pro Ultra to read the whole codebase and
produce a one-page architecture summary. Hand the summary
to Claude Sonnet to plan the refactor. Hand the plan to
GPT-5 to implement individual functions. Hand each function
to local Llama for unit test generation."

✗ One-club golf

"Use whatever model is the default in my IDE for everything."

The equivalent of using a single golf club for the entire round.

Try it · Vibe Trial

Solve the same trial with three engines

~90 seconds. Drops you into the workbench. Coach the engine — no code required.

Try this trial

Watch it · Codigami

Codigami: @kaz's three-engine solve of LRU Cache

Tape coming soon

Back to Part 4 Next lesson