Field Notes on Production AI Agents

What an AI agent can take over in sales order processing (reading order emails and PDFs, matching customers and products, checking price and stock, creating the order, chasing missing details) and what stays with a person, including credit holds and price overrides.

What an AI agent can take over in accounts payable (invoice intake, three-way matching, coding suggestions, statement reconciliation) and what stays with a person, including payment release and bank detail changes.

Where AI agents handle real estate leads and scheduling well, and where a licensed agent needs to stay directly involved, with the fair-housing guardrails specific to this industry.

Where AI agents genuinely help in claims processing and underwriting support, and where regulatory and trust requirements keep a human as the final decision-maker.

LangGraph, Pydantic AI, and the major AI agent frameworks compared on what actually matters for a production build, not benchmark scores.

Most organizations aren't ready for autonomous agents. Here's the AI agent governance layer, access records, eval gates, observability, ownership, that separates the ones that scale from the ones that stall.

Where AI agents handle support end-to-end, where they draft for a human, and where they should never touch a ticket, with realistic scope and cost ranges for 2026.

What AI agent development services actually include: process, architecture, realistic costs, and how to tell a production-grade build from a demo. From a studio that operates agents daily.

A buyer's guide to AI agent development companies: the four provider types, five signals of a production-grade studio, red flags, and the questions to ask on the first call.

Real numbers for custom AI agent development: 3,000 to 5,000 USD for a single agent, 15,000 to 50,000 USD for a fleet, what drives the price up, and when an agent is not worth building.

AI agents are the units. Agentic AI is the property that makes them agents. What the distinction means in architecture, why vendors blur it, and how to tell which one you are being sold.

AI agents in manufacturing belong in the paperwork layer, not the control loop. What they handle, what integration really costs, why the OT boundary is non-negotiable, and how to scope a first deployment.

Where AI agents earn their cost in logistics: quote handling, track and trace, exception management, documentation, and carrier communication. Plus the integration realities nobody quotes for.

RPA is deterministic and repeats one path perfectly. Agents are probabilistic and handle inputs nobody enumerated. Which one your process needs, where each fails, and why the answer is often both.

Where AI agents work in hospitality: the phone that nobody answers during service, reservations, supplier ordering, and review responses. Plus the one thing an agent must never handle alone.