Guides

What it takes to run an AI agent in production after the demo works: evals, tracing, safe changes, context, memory, tool integrations, model choice, cost limits, and infrastructure.

Test an AI agent with evals built from your real work: fixed inputs, expected outcomes, and automatic checks on actions as well as answers, run on every change.

Record a trace of every agent run: each model call, tool call, result, decision, and cost, linked in order, so any bad result can be traced to the step that caused it.

Handle every prompt, model, and tool change like a code change: version it, review it, and run the full eval suite before it ships, with past failures kept as tests.

Agents lose track when context is too thin, too long, or stale, and when task progress lives only in the conversation. Manage context per call and keep state outside the model.

Build agent memory as a data system with rules: facts stored with source and date, separate memory per person, summarized history, and a rule for which fact wins.

Usually yes. If a system has an API, webhooks, file exports, email, or a web page a person can use, an agent can work with it through a connector built for that system.

Use the model that passes your own tests at the lowest cost your data rules allow, and expect to use more than one: a large model for judgment, a smaller one for routine steps.

Measure cost per run and per task, cache repeated context, route simple steps to cheaper models, and put hard spending limits on every agent and tool.

An agent harness is the code around a model that makes it an agent: the loop, tools, context, limits, and records. When a framework is enough and when to build your own.

An agent graph breaks a job into steps with defined paths, branches, and approval points. When you need one, how to design it, and where graphs fail in production.

Where AI saves time in an independent or captive insurance agency: re-keying for quotes, renewals, certificate requests, service email, and lead follow-up, with licensed decisions kept with agents.

Where AI saves time in a wealth management firm: meeting notes into the CRM, account paperwork, review prep, and document-heavy operations across CRM, Microsoft 365, e-signature, and custodians.

Where AI saves time for electrical, solar, and construction contractors: sorting bid invitations, reading specs, moving takeoff data into estimating software, submittals, change orders, and back-office work.

Where AI saves time for property managers and landlords: maintenance request intake and triage across texts, email, and calls, tenant and leasing questions, vendor scheduling, and owner reports.

Where AI saves time in accounting firms, law firms, and other professional services: client intake, document collection, summaries, first drafts, and categorization, with a professional approving before anything reaches a client or the books.