Field Notes

Look closely at how Dots is set up. The most important feature is the settings page where a person decides what the agent is allowed to do. That's where the value of an always-on assistant gets made or lost, and it's a business decision more than a technical one. Before anyone on your team turns one of these on, sit down and write out the boundaries in plain language: which tasks it can finish alone, which ones need a person's sign-off, and which ones it should never touch. The companies that do that thinking up front will get real help from these tools. The ones that click through the setup will get surprises.

If you're trying to pick the one "best" AI model for your business, this week is a good reminder of why that's a losing game. The top spot changes hands every few weeks, and whatever you choose today will be matched or passed by spring. The more useful question is how your team checks the work. A 15 percent hallucination rate is a big improvement, and it's still well above zero. The models keep getting better. Someone on your team still needs to know what right looks like.

The agent did exactly what its settings technically allowed, and the person using it understood those settings differently. That gap between what a button says and what an agent actually does is going to show up inside a lot of businesses this year. Picture an agent with "Allow Always" access to your company inbox, your scheduling system, or your quoting tool. The fix is mostly human: train people on what each permission really means, start agents with narrow access, and widen it only after someone has watched how they behave.

The part worth watching is where the FTC thinks responsibility lands. FTC Chair Andrew Ferguson has said developers could face responsibility when they direct agents to run exercises that end in real-world breaches. That's aimed at the big labs for now, but the principle that whoever gives the order owns the outcome travels. If your company tells an agent to send emails, issue refunds, or book jobs, it's reasonable to assume your company owns the results. Keep a simple written record of what each AI tool is allowed to do and who approved it. It's good practice today, and it may turn out to be good protection later.

Set aside the headline risk language for a moment and look at the business side. The tools your team uses every day come from companies that are growing at a staggering rate and spending even faster. That's a reason to enjoy today's low prices while planning for change: pricing, plans, and features will keep shifting as these companies chase profitability. The businesses that handle those shifts well are the ones whose know-how lives in their people and their written processes, so moving to a different tool is an inconvenience instead of a crisis.

Every AI provider has bad days, and the right response is a plan. If a piece of your daily operation now depends on one AI tool, whether that's drafting quotes, answering customer messages, or scheduling, ask what your team does when it's down for an hour. The answer is usually simple: a manual fallback everyone knows, a second tool that's already set up, and a person who decides when to switch. That kind of plan costs almost nothing to write and saves a lot of scrambling.

Big numbers like this can feel abstract, so here's how we'd translate it for a business owner. We'd expect most of that change to happen inside existing companies, as roles slowly shift around new tools. That gives employers a real choice. You can treat it as a reason to replace people, or you can treat it as a reason to grow the people you already have into the jobs that are coming. Your current team already knows your customers, your systems, and your standards. Teaching them new tools is almost always faster and cheaper than finding and training someone new.

This one's for the trades and field service owners who read Field Notes. The technician crawling through an attic or rewiring a panel isn't about to be replaced by a robot, and this research suggests that's true for a long time. The AI changes in your business are going to show up first in the office: dispatch, estimating, scheduling, invoicing, and customer communication. That's where the near-term opportunity is, and it's also where your office team will need the most support as their work changes.

We love this, and we think it's the most telling story of the week. The company with 1.2 billion weekly users, which could easily put a tutorial video on its website, chose local advisors and in-person workshops to reach small businesses. We read that as an admission from the top of the industry that access to the tool was never the bottleneck. People learn new ways of working from other people they trust, in their own context, with someone around to answer questions. If you're a small business owner, it's worth checking whether your local SBDC is part of the pilot.