I run AI workshops for leadership teams. I have built AI products since 2004, starting with computer vision, so I have a practical sense of what each generation of this technology promised and what it delivered inside companies. This one delivers, but mostly to the companies that learn to build with it. The sessions start from that gap: a morning on where the technology stands and what it does to the economics of your work, then an afternoon of working prototypes built on your industry and scenarios drawn from your own business.
The goal is for your company to develop AI internally, the way we build companies at Cornell Tech's Startup Studio.
Three business questions frame every session: whether AI improves productivity in a way you can measure, what delivery costs once you count human review, corrections, data, software and support, and whether customers will buy, and buy again, a service that combines AI with your expertise. Faster model responses alone do not demonstrate savings or justify a staffing decision.
I rebuild the examples and the news items for each room. Several of the ones I used at the last session were days old.
Earlier shifts in work, from agriculture to offshoring, show what happens when one stage of producing an idea gets cheap. AI makes the middle stage cheap: turning an idea into something people can act on. You place your own revenue along that chain and see where customers still need you.
I cover what agents do, how they work together on longer assignments, and what they need to run: tools, memory, permissions and a budget. Smaller, specialized models get their own segment, since which model to use turned into the longest debate of the last session.
Customers will shop and move money through agents, which changes distribution and margins. Someone also has to answer for the harm an agent causes. Both questions are reaching the CEO's desk now.
AI performs unevenly, especially where an answer is hard to check. I show how to use it to review more evidence and explore more options while the people who know the customer keep responsibility for the decision.
We coach about 60 teams a semester to test assumptions with real users and bring back evidence, and the Studio itself now uses AI to support more teams. I show how the same structure runs inside a company.
When it fits, a founder from a Studio company joins to show how they built their business by learning from customers and how their team uses AI internally.
I run working systems live in the room, on public cases from your industry.
Before the session I build prototypes on public cases from your industry and run them live. Each one takes a day or two to build, and each one stops well short of a finished service. That is the point. They show what a small team inside your company could build if leadership backs it, and they start the argument about value and delivery cost that the afternoon is for. The patterns below repeat across industries, while the scenarios and the demos change for each room.
After the prototypes, teams work through scenario cards written for your industry. Each table takes a problem a customer or a CEO has to deal with and asks what AI could generate, automate, predict or personalize about it. We call that lens GAPP. The table develops a service or product idea, then names the evidence it would need to decide whether to pursue it. Every table leaves with an idea and a list of what it would take to test it, which is a better use of an afternoon than playing advisor to a fictional CEO.
This is how we run the Studio, and it is how I would run AI development inside your company.
Small teams work independently on a shared base of approved knowledge, tools, standards and access rules, with your experts deciding what others can reuse. Each team starts from a specific customer problem, which is the reverse of most corporate AI projects I have seen, and keeps its investigation narrow enough to learn before anyone commits to a large build.
Every team records its assumptions, prototypes, customer feedback, revisions and decisions in one project history, so leadership can follow the thinking and the evidence behind it. The rules about who can see which customer's data hold when a team reuses someone else's work. The structure is the same in a corporate function and in a customer-facing unit.
You leave with a short record of candidate services or products, the first test for each, the evidence each needs, and a proposed owner. Leadership keeps the prioritization and funding decisions. My job is done when each candidate has a first test and a name next to it.
The core is the same in every format. We set length and depth together.
The offsite covers everything above: the morning sessions, the prototypes and the scenario workshop. Your own operating partner or an outside practitioner can fill part of the morning.
I build the prototypes and the scenario cards on that unit's work, so the ideas that come out of the room belong to the people who will test them.
A board gets the morning's view of where the technology stands, one prototype, and the questions boards ask about risk and governance.
Your people build on your own scenarios with Studio coaching, so the capability stays in the company. Candidates that hold up can move into venture building.
The consulting work that follows a workshop has its own page.
The workshop ends with the record. For companies that want the strongest candidates built into businesses, that is venture building: the candidates, the gates, the teams and the economics are all described there.
How venture building worksSend inquiries to alberto@escarlate.com. Sessions run in English or Portuguese. I work from New York and keep a regular agenda in Brazil and Latin America.