Open warmly. "Welcome back. April was the conceptual frame, the six flaws, Green / Yellow / Red. Tonight is the hands-on lab. You're going to leave with a four-question rubric and a tool tier you can apply to any task in your business this week." Beat. "We're going to talk for about fifteen minutes, then you're going to use a lab, then we'll talk about what you saw. Twice. That's the shape of the night."

Set expectations. This is not a sit-and-listen session. You will be on your phone or laptop running real prompts at two points tonight. Pick one repetitive task in your business now, drafting a customer email, summarizing a meeting, writing a job description, naming an offer, and hold it in your head. We'll keep coming back to it.

Orient the room. Four questions everyone has had at some point in the last sixty days. By the end of the hour and a half, you should have an answer to all four. The "who this is for" line is for the new faces in the room: this is a small business AI talk, not a developer talk.

Sixty-second recap, do not relitigate. This is the trellis for tonight. Once you know whether a task is Green, Yellow, or Red, the next question is, "of the AI tools out there, which one do I actually use for this Green task?" That's tonight.

Reframe. Last month was "should I use AI for this at all." Tonight is "given that I should, which one." That's the question on every founder's screen right now and the answer is not "the most famous one." It's the one that matches the four things we're about to walk through.

This is the heart of the talk. The technical version of this list is "capability, latency, cost, reliability." We're using the plain-English version because that's how a small business owner actually thinks about it. The key insight, and the one that costs money when people miss it, is that no tool wins on all four. The ones that are "smart" are slow and expensive. The ones that are "fast and cheap" make more mistakes. You're always trading.

This is the Benjamin framework, made concrete. Three tiers. Real product names so people leave the room knowing what to actually open tomorrow. The pricing column is the part that surprises people. The difference between tier one and tier three is roughly 30 to 100 times in cost per task. That's not a rounding error. If you default to the fanciest tool for everything, you're spending 50 to 100 dollars where you should be spending one.

Land it. "Same task. Same prompt. The cheapest tool costs about a thirtieth of what the most expensive one costs. That doesn't mean cheap is always better, sometimes you need the expensive one. It means *defaulting* to the expensive one for every task is the single most common waste of money in small-business AI adoption."

This is James's note made literal. Side-by-side on a real task they all do every day. The point is not "the expensive one wins." The point is "the expensive one wins on this particular kind of task." On a different task, say cleaning up a list of email addresses, the cheap one wins because there's no nuance to add. Setting up the lab.

Walk them to the URL. Hold the slide up. "Open this on your phone or your laptop. The lab lets you pick up to three models from a list of about fifteen, paste a prompt once, and see all three answers next to each other along with what each one cost and how long each one took. Take fifteen minutes. Use a real task. We'll regroup at the bottom of the hour and you'll tell me what surprised you." Stay on this slide while they work. Mute yourself. Pop back on at the ten-minute mark to give a two-minute warning.

Open mic. Get three or four people to share what they tried and what they saw. Listen for the moment someone says "I was paying for X but the free version did the same thing." Underline it. That's the highest-leverage realization in this entire session.

This is Benjamin's rubric, in the form a founder can actually use. Four yes/no questions. If you can answer "yes, yes, yes, yes" to a tool on a specific task, you've got a green light to automate that task with that tool. If any of them is a "no," you either pick a different tier, change your workflow to add a human review step, or skip the automation for that task.

This is the moment Benjamin asked for. The four-axis framework, ground to a useful business example. Walk through each row slowly. The point is not "AI does this for me." The point is "the rubric tells me how much trust I extend, and where I add a human." That's what's missing in 90% of what they're being sold by vendors right now.

Reinforce the tiers with real departments. Founders see themselves in this list. The biggest payoff is the left column, the "workhorse" tasks where the cheap tier gets you most of the way and the only review needed is a quick read-through. That's where you'll get the most hours back per dollar.

Shorter lab. This one is more "click through and read" than "type a prompt." The reason it's worth eight minutes is that "agent" is the most overused word in this space right now and most of what's being sold as an agent is actually a workflow with one AI call in it. The lab gives them the vocabulary to push back when a vendor pitches them an "agent" for $5K a month that should cost $50.

The takeaway from the agent lab. Most of what your vendors are selling you as "AI agents" do not need to be agents. Workflows are cheaper, more predictable, and easier to operate. When someone pitches you an "agent" for a task that has a clear five-step structure, ask why a workflow isn't enough. Save yourself thousands.

The take-home. Read it slowly. This is the deliverable from tonight. Every person in the room should be able to run this list on a real task by the end of the week. The reassess-monthly step is the one most people skip and it's the one that compounds. The cheap tier in 2026 is doing things the expensive tier did in 2024.

Reserve eight to ten minutes for this. Three or four people share, two minutes each. The act of saying it out loud commits them to running it this week. The followup, even informally, is worth more than another slide of content.

Three takeaway cards. Read each. Beat between them. These are the lines you want them quoting on a sales call next month when a vendor tries to upsell them on something they don't need.

Last functional slide. Everything from tonight, in one place. Bookmark this URL, the talk page itself, and you can come back to all the labs and articles whenever you want.

Hold the contact slide while you take questions. Encourage people to grab the labs URL on their way out. The conversation that happens after the formal Q and A is usually the most useful one.