HERNDON, Va. — Some of the hardest decisions a business owner faces don’t come with a spreadsheet attached. They come at 11 p.m., when there’s no one to call.

In Part 1 of this series, we looked at how AI can pull useful insights out of a business’s existing data. But SCORE mentors Charlie Morris and Matthew Krieger, presenting at a recent SCORE National Small Business Week webinar, devoted equal time to a different kind of use — not analysis, but conversation.

“The hardest problems aren’t in your data,” Morris says. “They’re in your head.”

Getting an outside perspective on a difficult business situation typically means scheduling an appointment, paying by the hour, and explaining all the context from scratch. By the time the meeting happens, the moment may have passed.

AI, Morris says, changes that equation. “It’s the first thinking partner available the moment the problem hits, not three weeks later.”

To demonstrate, he walked through a scenario familiar to anyone who has run a customer-facing business. A long-term jewelry store customer had brought in a vintage ring for resizing. When she picked it up, there was a crack in the filigree. The store’s goldsmith believed the metal had already been stressed and the resize simply surfaced an existing problem. The customer disagreed and was asking for the repair to be covered.

Morris gave the AI all the details in a written document: the history with the customer, the goldsmith’s assessment, the ambiguity about fault, the cost of the repair and what the ring meant to the woman emotionally — her mother had recently passed away. Then he asked it to help him think through what to do.

The AI didn’t give him a quick answer. It identified what the situation was really about: a grieving customer, an unresolvable factual question, and a long business relationship all pulling in different directions at once. It laid out the case for covering the repair fully, the case for not covering it, and the case for splitting the cost, with the honest trade-offs of each.

When Morris asked the AI to argue against his instinct to pay, it pushed back clearly. “The precedent you set here isn’t just about Carol,” it said in the demo. “It’s about every old piece you’ll ever touch.”

He then asked the AI to respond as the customer herself. What was she actually feeling? What did she need to hear? The exercise, Morris says, is something cleaners know well from their own work: making a situation right with a customer requires understanding what the customer actually needs, which isn’t always the same as what they’re asking for.

The demo closed with the AI producing a script for the phone call, including what to say and how to frame it.

“She didn’t need a consultant,” Morris says. “She needed someone to think with.”

From there, Krieger picked up the thread, moving from AI as a conversation partner to AI as an autonomous worker.

Krieger, who runs a manufacturing business in Connecticut and has spent significant time working on AI-related projects, introduced the concept of agents: AI that doesn’t just respond to questions but executes multi-step tasks on its own.

“Rather than defining a set of steps, you focus with AI on the objective,” he says. “The agent attempts to achieve the objective while working around any roadblocks it may hit.”

The distinction matters. Traditional automation follows a script and does what you tell it to do, step by step. An agent, Krieger says, pursues an outcome. If it hits an obstacle, it adapts.

“Your focus was on tasks,” he says. “Now it’s on outcomes.”

To demonstrate, Krieger gave an agent a business objective rather than a procedure: search Amazon for men’s raincoats in a specific size and rating range, find three different brands, build a Google Sheet with the product names, links and prices, add a column for the Connecticut sales tax equivalent and flag which product was the best value with a one-line reason.

The agent — running on Claude Cowork, Anthropic’s agent tool — went to Amazon, browsed through results, selected products meeting the criteria, built the spreadsheet, applied the tax calculation and identified a best-value pick. It handled the steps itself.

Krieger noted that when he’d tested a similar prompt earlier with a less specific request, the agent had asked a clarifying question before proceeding. When it couldn’t find enough results matching one set of criteria, it widened the search and explained what it had done.

“Your focus as a businessperson becomes managing that result, not the details of the process,” Krieger says.

Come back Tuesday for the conclusion of this series, where we’ll look at using AI to build simple business tools from scratch, and how to decide where AI actually belongs in your operation. For Part 1 of this series, click HERE.