Articles tagged “tool-calling”
27 articles
How to Test Agents That Call the Right Tool and Still Get It Wrong
Tool-use benchmarks check whether an agent picks the right tool. They miss agents that call correctly, then mishandle the result. Here's how to test for it.
How to Test Tool Argument Correctness in AI Agents
The most common production agent bug isn't picking the wrong tool. It's picking the right tool and passing the wrong arguments. Here's how to catch it.
How to Build an Agent That Writes Code Instead of Calling Tools
One JSON tool call per turn is slow, token-heavy, and brittle. A short script that orchestrates tools in a sandbox is faster and cheaper. Here's how it works.
MCP now uses round-trips, not streams, for user input
The July 28 spec deprecated elicitation and sampling. Multi Round-Trip Requests (SEP-2322) replace both with a stateless pattern: the server returns InputRequiredResult, the client collects input, the client re-submits. Here's how to migrate.
Agentjacking: how attackers hijack production AI agents
Agentjacking is a new attack class that targets the data your agent trusts, not the model itself. Learn how attackers exploit CRM notes, tool responses, and knowledge bases to hijack production CX agents, and how to stop them.
Four protocols that power every AI agent
MCP handles tools, A2A handles agent delegation, ACP routes across enterprise frameworks, and ANP crosses org boundaries. Here's how the four protocols divide the work and when each belongs in your stack.
MCP's 2026 spec changes what your tools can do
The MCP spec RC is out with Extensions, Tasks, MCP Apps, and OAuth hardening. Here's what changes for CX agent builders before the July 28 final.
Tool receipts: verifiable proof of what your agent actually did
When your AI agent claims it called a tool and got a result, can you verify that? Tool receipts are structured execution proofs that catch fabricated tool outputs before they damage customer trust.
When your agent spends money: the A2A payments era
The A2A Agent Payments Protocol (AP2) lets AI agents initiate transactions autonomously. Here's what CX builders need to know: the use cases, the controls that make it safe, and how to keep spending observable alongside conversation quality.
Static API keys don't work for autonomous agents
When you hand an autonomous agent a static API key, you're giving it a skeleton key with no expiry. Here's the per-call permission model that replaces it.
Tool result caching: the latency and cost wins hiding in your stack
Your agent re-fetches the same data on every call. Tool result caching cuts latency by up to 70% and inference costs by 40-60% with changes that take days, not weeks. Here's how to classify, implement, and measure it.
Parallel tool calls: the 5x speedup in your agent
Sequential tool calls compound latency for no reason. Here's how to fan out independent tool calls in parallel and cut agent response time by up to 5x in production.
Tracing AI agent failures across multi-step tool chains
When a production CX agent returns the wrong answer, the bug rarely lives in the last LLM call. Here's how to trace failures back to their root cause across multi-step tool chains.
Design agent tools that won't double-charge on retry
Agents retry when tools fail. Without idempotency, retries create duplicate records, double charges, and corrupted state. Here's the engineering pattern that makes every mutating tool safe to retry.
Five things your CX agent needs that its framework won't provide
LLM frameworks handle conversations. They don't handle memory, tool reliability, pre-production testing, or observability. Here's what every production CX agent actually needs.
MCP tasks: how async tool calls fix your agent's timeout problem
The November 2025 MCP spec introduced Tasks: a call-now, fetch-later primitive that lets agents kick off slow operations without blocking. Here's how it works and how to build with it.
How to Use MCP Sampling, Roots, and Elicitation in CX Agents
Most MCP tutorials cover only server-side features: tools, resources, prompts. The three client capabilities (Sampling, Roots, Elicitation) unlock human-in-the-loop patterns that server tools alone can't. Here's how to use them.
Structured Outputs: Make Your AI Agent Stop Guessing
JSON mode isn't enough. Learn how constrained decoding, Zod schema validation, and validator-retry patterns cut agent parsing failures in production.
Pre-Execute Tool Calls to Cut Agent Latency 48%
Sequential tool calls quietly kill your agent's response time. PASTE shows you can pre-execute likely tool calls during LLM thinking time and cut latency 48% without touching your model.
How MCP Tool Descriptions Break Your Agent
New research shows 97% of MCP tool descriptions have quality issues that hurt agent accuracy. Here's what the smells look like, why they matter, and how to fix them.
MCP Apps: Build UIs That Render Inside AI Chat
MCP Apps let your tools return interactive HTML dashboards, forms, and visualizations that render inline in Claude, ChatGPT, and VS Code. Here's how to build them for CX agents.
AG-UI: The Protocol That Connects Agents to UIs
AG-UI is the open event-based protocol that streams AI agent state to any frontend in real time. Here's how it works, what events it defines, and how to wire it up in TypeScript.
Circuit Breakers for AI Agents: Stop the 3 AM Meltdown
One retry loop at 11 PM becomes $437 by 7 AM. Here's how to implement circuit breakers for AI agent tool calls, LLM calls, and external APIs, with TypeScript patterns that stop cascading failures before they start.
Your CX Agent Doesn't Care Who Won SWE-Bench. Here's Who Actually Wins.
SWE-bench crowns a coding king. Customer experience agents answer to a different benchmark, tau-bench, and the rankings flip. The head-to-head that actually predicts production reliability.
Everyone Benchmarks Opus. Your Chatbot Runs on Haiku.
Haiku 4.5, GPT-5 Mini, Gemini Flash at the $1/MTok tier that powers CX. Tool-call accuracy, first-token latency, structured-output reliability, blended cost math.
MCP SSE Is Deprecated. Here's How to Migrate
SSE transport is being deprecated across major MCP platforms in 2026. Here's a practical migration guide from HTTP+SSE to Streamable HTTP, with TypeScript examples and a phased rollout strategy.
MCP Streamable HTTP: The Transport Layer That Makes AI Agents Production-Ready
MCP's Streamable HTTP transport replaced the original SSE transport to fix critical production gaps. This guide covers what changed, why it matters, and how to implement it in TypeScript with code examples.
El briefing de Signal
Un email por semana. Cómo los equipos líderes de CS, ingresos e IA están convirtiendo conversaciones en decisiones. Benchmarks, playbooks y lo que funciona en producción.
