Sam Nadler built an AI estimator for Offsiteio that cut offsite proposals from a week and a half to 3-4 days, plus when Jordan reaches for Google AI Studio.
An AI estimator can cut offsite proposal time from about a week and a half to three or four days. That is what Sam Nadler reported on Episode 38 of Built This Week after building an estimator and client presentation tool for Offsiteio, the corporate offsite company in our portfolio. Jordan Metzner then walked through Google AI Studio and how he picks between AI coding tools.
Offsiteio plans offsites and corporate events for groups of about 20 to 200 people, sometimes more. Sam says there are a lot of moving pieces: food and beverage minimums, AV setup, conference room fees and making sure everyone has their own room. And one of Offsiteio's promises is that once it quotes a price, that price never changes. So every cost has to be identified before the quote goes out.
Sam used a simple example. Quote $100,000 for a 100 person offsite in Austin, and then the real bill comes in higher:
If that off-site ends up costing us a 110,000 or a 105,000 or a 120,000, we're upside down on that event, which would be a big problem.
— Sam Nadler, Episode 38
Before AI, the team handled this in spreadsheets that kept getting more elaborate. The estimator replaces that Google Sheet. You enter everything from a hotel's bid, such as meals, activities, meeting space and AV, in a structured way. The key feature, Sam says, is that it flags pitfalls like food and beverage minimums and AV costs, the places where hotels can "sink their claws" into a planner who isn't used to negotiating with them. Those costs get locked in with the hotel before the client ever sees a number.
Watch Sam Nadler demo the Offsiteio estimator in Episode 38 (starts at 1:50) · Episode 38 notes and transcript
Once locations and prices are confirmed, the options still have to be presented. Sam says that used to mean corralling all the information into a Canva or Figma presentation, which could take days to perfect.
The second tool pulls straight from the saved estimates. You upload the client's logo, choose which properties to include (maybe seven out of ten estimates), and it creates a website you can share right away. The client sees a map, what is included, the all in per person rate, the dates, the number of nights and the number of guests. They can also shortlist favorites, so the next call can focus on, say, two options in Miami and one in New York City.
Sam calls it "not that crazy of a tool," but the time savings are real: seven or eight estimates used to take about a week and a half to prepare, and now take three to four days.
When you have a client who's potentially interested in off-site or really interested in purchasing anything, speed matters.
— Sam Nadler, Episode 38
Sam started in Claude Code. After some bugs and updates, he moved to Codex. The back end runs on Supabase, and the front end is a React web app hosted on Amplify. The tool is live in production. We have a full breakdown on the Offsiteio estimator build page.
Jordan asked whether customers had noticed. Sam says the first client to see it, early this year, was a repeat customer who had used Offsiteio for years. Moving from a PDF built in Canva to a well designed website they could click through was an upgrade, and that customer went on to book their biggest offsite with Offsiteio yet. The team now uses the tool every day.
Sam says version two has two goals. First, Offsiteio uses Cvent to generate hotel bids, and he wants those bids flowing into the tool so the bid process is as automated as possible.
Second, he wants it to become a tool for the client, not just an internal tool and a presentation tool. Clients make a lot of choices after they confirm an offsite: the rooming list, dietary choices, the exact itinerary. Much of that isn't decided until days before the trip. The plan is to give them one place to confirm choices, see where things stand and request adjustments.
In the second segment Jordan showed Google AI Studio, which you reach at ai.studio. He says setup can be a little complicated because you may need an API key to pay for usage. What he likes is that many models and tools sit in one standardized interface.
Sam's rule: if it's a throwaway or a quick demo, he builds it in Google AI Studio. Otherwise he starts in Claude Code and moves to Codex, or the other way around.
Jordan says each tool lives in its own vertical. Big pieces of software go to Cursor, Claude Code or Codex. Something that needs a quick database, lives for a while and then gets thrown away fits the Replit and Lovable world. A demo to show someone a gimmick goes to Google AI Studio, Lovable or Bolt. He still loves AI Studio for building little web apps because it is fast and clean and gives him good inspiration, and sometimes he just has Claude build something in the browser.
In the news, Sam covered Anthropic's announcement of a new model it says is too powerful to release yet. Instead, he says, dozens of corporate partners are testing security first, and some of its scores were well above Opus 4.6. Jordan says he read that Anthropic formed a consortium of 40 companies, including AWS, Microsoft and Apple, to address the vulnerabilities it found. He is not surprised that a very large coding focused model can find security holes humans missed, and expects future models to get better at both finding and exploiting them.
It does make me feel weird that, something this powerful according to them is in the hands of, like, just Anthropic.
— Jordan Metzner, Episode 38
He was quick to add that we like Anthropic, use it a lot and pay it a lot. He also wants to start coding with Mythos himself. He doesn't expect that to happen anytime soon.
Sam's screen share shows the estimator and the client proposal site step by step, and Jordan's AI Studio tour includes the Nano Banana image and the Lyria song. Watch Episode 38 of Built This Week, where the full transcript is also on the page. For another internal tool built by a non-engineer, see Episode 39.
