Most AI demos are built to be watched. You leave with nothing on your machine.
Somebody competent drives, the example was chosen three days earlier, the runs that failed got edited out, and forty minutes later a room agrees it was impressive. Then everyone goes back to their inbox and nothing has changed.
This preview is built the other way around. The system runs on that morning's actual sources while you watch, so the result is dated and neither of us has seen it before. You spend six of the thirty minutes writing a file of your own. The file works on its own, in any chat window you already use, with no software and no spending.
Thirty minutes is enough to tell whether the two-hour build is for you. It is also enough to be worth your time if the answer turns out to be no.
What happens in the thirty minutes
One public document, read two ways.
A filing from the last seven days, open on screen. What it means to somebody running a job search, and what the same page means to somebody selling into that market. Every word of it is public and almost nobody read it.
The system runs, live, on that morning's news.
One typed command. The run happens on screen while it happens, sources opening and items getting scored and discarded. What comes back is dated this week and unrehearsed.
You write the file. Six lines. Yours to keep.
Memory is the first of the five patterns and the only one that needs no software at all. Six lines in a plain text file, written in the room, pasted at the top of every AI conversation you have afterward. This segment is the reason to come even if you never buy anything.
The machine marks its own guess as a guess.
The system works out a contact from a naming pattern, labels its own answer Low confidence, declines to draft an email, and routes the person to LinkedIn instead. Nobody prompts it to do that. The rule lives in a plain text file, written in plain English. This is the part people repeat to a colleague afterward.
What the paid version is, in ninety seconds.
A plain description of the two-hour build session, the requirement, and the price. No urgency, no seat counters, no closing question. You decide later and on your own.
Questions.
Chat is open the whole time and gets read at minute 13 and again at minute 25. The live output stays on screen while the questions get answered.
What you leave with, including if you never buy anything
A working memory file
Six lines you wrote yourself. It stops you re-explaining your situation to a chat window every morning, and it works in whatever tool you already pay for.
One dated result
Something specific the system surfaced that morning, with the source and the date named out loud. Proof that the category does something other than summarize documents.
An honest read on Friday
Send six companies you would take a call from tomorrow and you get a straight answer on whether a build session would produce anything for you. Some people get told no.
What this preview is not
- Not a build session. Nobody builds a system in thirty minutes. You watch one run and you write one file.
- Not recorded and sent afterward. The value is that the system reads news none of us has seen. A recording of that is a different object and I would rather you came live.
- Not a sales call with a demo attached. The paid sessions get ninety seconds near the end. Nothing gets asked of anyone in the room.
- Not a prompt-engineering class. No phrasing tips. The five patterns are software architecture explained in plain language, and you see one of them today.
Thirty minutes, and I end on time.
Save a seat
One email to confirm, one reminder the morning of, and no sequence after that.
What the system produced on a run like this one
The Savannah run, July 2026
In a demo market, the system surfaced a $398M convention hotel groundbreaking the week it happened. It identified the operator, found the executive responsible for the opening, labeled its email guess honestly, and left this in the drafts folder:
Subject: Your 444-room groundbreaking on the riverfront
Congratulations on breaking ground on the Signia. A convention hotel that size changes how the market competes for citywide events. Staffing it will start long before 2028...
Five sentences. Their news, one line of proof, one small ask. The contact line above it carried a Low confidence label and got routed to LinkedIn rather than emailed.
Wednesday's run is a different market and a different morning. Nobody knows what it returns, which is why it is worth thirty live minutes.
And if you want the whole thing
The paid version is two hours on Friday, July 31. You build the complete system on your own laptop while I build mine on screen.
Seventeen numbered pastes. No code, and nothing typed into a terminal. Your first real search runs at minute 75 on your market. All five patterns get named out loud as you build them: Memory, Specialists, Recipes, Conductor, Ledger.
Three tracks. Find Your Next Job, Find Hidden Leads, Find Your Next Story.
Two things stated plainly
It requires a paid Claude plan, twenty dollars a month, in place before the session starts. The free plan runs out of capacity partway through a build, which would leave you watching instead of building.
The session price is {PRICE}.
Both are named here rather than at checkoutNone of that is the reason to come Wednesday. Come Wednesday for the thirty minutes, and decide about Friday afterward.
Save a free seat for Wednesday Read about the paid sessionsQuestions people ask about the preview
Is this actually free?
Yes. No card, no trial, no upgrade prompt at the end. The paid sessions get about ninety seconds of plain description near minute 22 and nobody in the room is asked for a decision.
Do I need a paid Claude plan to attend?
No. Nobody builds anything on Wednesday, so nothing needs installing and no subscription is involved. The paid plan is a requirement for the two-hour build session on Friday, at twenty dollars a month, and it gets named plainly there too.
Will I get a recording?
No, and the reason is the whole design. The system reads that morning's news and none of us knows what it will return. A recording of an unrehearsed run is a different object and it teaches something else. Come live or come to the next one.
What if the live run fails?
Then you watch me handle a failed run. I do the entire demo ninety minutes before every session and keep that output saved, so the session continues from the earlier run and I say so out loud. Watching somebody deal honestly with a broken run tells you more about the system than a clean one does.
Do I have to be in hospitality?
No. Hospitality is where these systems were built and where the examples come from. The architecture cares about signals and sources, and neither is industry-specific. Bring six companies from your own market and the exercise works the same way.
Should I prepare anything?
Two minutes of work. Write down six companies you would take a call from tomorrow, using real names. Open a blank text file and leave it behind your browser. People who arrive with real names leave with a real file, and people who arrive with a category leave with a template.
Will you email me forever afterward?
One confirmation, one reminder the morning of, one follow-up the same day, and one message two days later. Then nothing further. Reply with the word stop at any point and it ends sooner.
Who is running it
Michael J. Goldrich is the founder of Vivander Advisors and teaches agentic workflows to industry leadership groups, including the AHLA Global Technology 100. He built the systems shown in the preview and runs the same patterns every morning on his own business.
Save a free seat, Wednesday July 29