What AI agents can actually do for your business (and what they cannot)

Every week a business owner asks us some version of the same question: "Should we be doing something with AI?" The honest answer is usually yes, but almost never in the way the marketing suggests. This article is the summary we wish we could hand out at every intake.

What an AI agent actually is

Strip away the branding and an AI agent is a program that reads incoming information, makes a decision within boundaries you define, and takes an action: sends an email, updates a record, schedules a meeting, or asks a human for help. The language model inside it is good at one thing above all: understanding messy, unstructured text the way a person would.

That single capability is why agents work now where automation failed before. Classic automation broke the moment input stopped fitting the template. An agent reads the email a customer actually wrote, not the form you hoped they would fill in.

Tasks agents handle well today

• First-line email: sorting, extracting what matters, drafting a reply for approval or sending it within strict rules.
• Lead qualification: reading an inquiry, checking it against your criteria, asking follow-up questions and scheduling the ones worth your time.
• Document processing: pulling structured data out of invoices, contracts and application forms, with a human reviewing edge cases.
• Internal lookups: answering "what is our policy on X" or "where is the file for client Y" from your own documentation.

Where they still fail

Agents fail where the cost of one mistake is high and verification is hard: final pricing decisions, legal commitments, anything that touches money without review. They also struggle with tasks that need context that lives only in someone’s head. If your process is undocumented chaos, an agent automates the chaos.

How to pick a first project

Look for a task that is frequent (daily, not monthly), text-based, rule-describable, and annoying enough that your team celebrates losing it. Then automate that one task end to end, measure the hours saved, and only then move to the second. Companies that start with one boring, high-volume process see payback in months. Companies that start with a moonshot usually see a demo and nothing else.

If you want a second opinion on which of your processes qualifies, that is exactly what our intake call is for.