agentic_ai_basics

llm_backbone

LLM (Backbone) — the large language model that serves as the reasoning engine inside an agent, swappable independently of the agent’s architecture.

llm_backbone.py

Runs one fixed harness — same system prompt, same tool, same loop shape as ../tool_use/basic_agentic_tools.py — against the identical task under two different models (claude-haiku-4-5 and claude-sonnet-5), printing each one’s answer, token usage, and time taken side by side. Type your own task, or press enter for the built-in default.

Concepts covered

• The model is a parameter, not a constant — every other template in this repo (starting with ../basics/basic.py) fixes MODEL once at module level. Here, run_with_backbone(model, task) takes it as an argument on purpose, because the whole point is running the same harness against different backbones in the same process.
• The harness stays fixed; only the backbone changes — run_with_backbone is called twice with nothing different except the model string. Any difference in the two runs’ answers, tone, token counts, or latency is attributable entirely to the backbone, because everything else was held constant.
• Usage is per-backbone, not per-harness — response.usage (see ../token_tracking/basic_token_tracking.py for a dedicated deep dive) is accumulated separately for each model, making the cost/quality tradeoff between backbones directly comparable rather than theoretical.

Run

From the repo root:

pip install -r requirements.txt
export ANTHROPIC_API_KEY=your-key-here
python3 Core_Architecture/llm_backbone/llm_backbone.py

Try:

Task to give both backbones (blank for a default demo):
> A store is offering 30% off a $85 jacket. What's the final price?

Running the same harness with 2 different backbones on:
  A store is offering 30% off a $85 jacket. What's the final price?

--- backbone: claude-haiku-4-5 ---
answer: The final price is $59.50.
tokens: 412 in / 18 out | time: 1.1s

--- backbone: claude-sonnet-5 ---
answer: The final price is $59.50, after a $25.50 discount.
tokens: 486 in / 27 out | time: 1.8s

Configuration

• BACKBONES — the list of model IDs to compare (default: ["claude-haiku-4-5", "claude-sonnet-5"])
• MAX_TOKENS, EFFORT, SYSTEM_PROMPT — see ../basics/README.md; these stay fixed across every backbone in the comparison, unlike model
• TOOLS — the single calculate tool available to every backbone under test

See also

• ../basics/README.md — where MODEL is a fixed constant instead of a swappable parameter
• ../token_tracking/README.md — a focused look at the usage object this file compares across backbones
• ../agent/README.md — the same “one harness, autonomous loop” shape, focused on goal autonomy rather than backbone comparison