Designing Tools a Model Calls Well
Function calling as API design: JSON Schema definitions, names and descriptions, granularity, actionable errors, idempotency, parallel calls, result size and pagination, the tool-use loop, and selection evals.
Overview
Why this topic
When an agent fails, the model is rarely the first thing to fix. The usual cause is a tool: a vague description, two tools that overlap, an error message that says only 500, or a result so large it pushes the task out of context. Tool design is API design for a caller that reads every word of your docs on every request and cannot ask a follow-up question.
Interview questions
- Write the tool-use loop for an agent. What does each step guard against?
- Your agent has 40 tools and keeps choosing the wrong one. How do you fix it?
- Design the error contract for your tools. What goes in an error result?
- How do you make tools with side effects safe to call from an agent?
- When should a model make parallel tool calls, and what can go wrong?
- A tool returns a 2 MB JSON document. What do you do?
Sign up free to read the full topic
More Agentic AI topics
- AI Gateways: Routing, Fallback, Caching & Metering
- Async Long-Running Tool Calls
- Autoscaling LLM Inference on Kubernetes
- Browser & Computer-use Agents
- Context Engineering & Prompt Design
- Context Window Management for Persistent Agents
- Distillation & Small Models
- Embeddings & Vector Search
- Evaluating Agents: Task Suites, Trajectories & Regression Gates
- Evaluating RAG & Catching Hallucinations
- Fine-tuning LLMs: SFT, LoRA/QLoRA & Preference Tuning
- Guardrails & Prompt Injection Defense
Start free on Cariara · Coding problems · System design studio