Today you get a guest writer: Taylor.
If you’re expecting Laura’s beautiful prose and stunning pictures…fair warning, this isn’t that. But I promise that it will be interesting for other reasons. I hope you read this post and walk away thinking about three things:
Building a company or app from scratch has never been more accessible. All it takes is a lot of curiosity, patience, and a little helper (AI).
If you’re not using AI yet, start. We are all living through a transformation as big (or bigger) than the internet.
The impact of AI on society will be staggering — probably in years, but maybe in months.
I’ll start by providing a little context as to where/how/why I got started using AI for trip planning. Then I’ll move into a Technical section, which will be…well, more technical. As I’ve chatted with folks about building this web-based app, I’ve already gotten some great questions about the tooling I’ve used and how the app is built. I will then progressively climb onto my digital soapbox to talk about the broader implications of AI from a philosophical and political perspective (don’t worry, I’ll keep that short).
The Context
As you know from earlier blog posts, we used ChatGPT to help us plan our trip (spoiler alert, it was awesome). A year or two ago, we would have opened up Google Maps and started to think about our route, then performed about 1,000 Google searches to figure out what specifically we wanted to see in those areas, looked at Tripadvisor for activities, AllTrails for hikes, Recreation.org for camping, Kayak for hotels, GaiaGPS or onX OffRoad for dispersed camping and off-roading…and it would have taken weeks.
We do enjoy that process to some degree, but it eventually gets exhausting, especially when thinking about planning for a multi-month trip. The entire planning process is also much more complicated for overlanding. When you aren’t staying at established campsites, then you need to additionally look at MVUM (Motor Vehicle Use Maps) to find forest service roads and identify potential campsites…which is not easy.
You can see a few examples of what these MVUM maps look like, below. The solid lines on the map indicate more-developed U.S. Forest Service Roads, and the dotted lines are less-developed or maintained (i.e., requiring a high-clearance or 4x4 vehicle to access). You will notice that almost none of the roads are named/marked (this is true on the ground as well as in the app), so it is also challenging to follow the routes in real time, or to know which service roads are currently or only seasonably passable.
Fun fact: The U.S. Forest Service maintains roughly 380,000 miles of roadways, enough to encircle the earth 15 times. You can camp within ~200 feet of most of these roads. Where do you want to camp?
Enter ChatGPT! We used a fairly complicated prompt, but thirty seconds later had the spine of our trip. It produced activities that fit our interests, beautiful hikes that were manageable enough that they wouldn’t leave us immobilized for three days after, found both dispersed and traditional campsites, fun off-road routes that our vehicle/skills could handle, local restaurant gems, and more. The AI-produced itinerary that it came back with was also accompanied by a coordinates document that could be uploaded to our mapping apps.
Let me be clear: we changed and adapted this itinerary significantly before and during the trip, but it saved us hundreds of hours of work.
We felt this functionality helped resolve a real pain, but I spent not a small amount of time refining the prompt to really work. Essentially, you need to know what information is most pertinent and how to structure the output to be useful. I wanted to make this entire trip-planning process easier for folks and to incorporate my learnings, hence TrippieAI.com.
The App
TrippieAI.com is for overlanders, boondockers, road trippers, and other explorers. It builds custom itineraries to jumpstart your roadtrip planning.
Give it a try! It’s free (for now).
I decided to build this web-based app exclusively as a hands on learning process. I don’t expect to make any money, in fact I’m pretty much guaranteed to lose some, but I wanted to experiment with AI and some new technologies.
As many of you know, I’m an early adopter. I love playing with new tech, trying the beta version of software, buying the new gadget (Laura can confirm this last one in particular). However, I couldn’t code my way out of a paper bag. My career to this point has been technology adjacent, in that I had to know the trends and be able to talk the talk but never had to be truly technical. I don’t want to understate the value that my prior knowledge provided me in this process, but also don’t want to let that discourage you from pursuing a similar project. With some curiosity and YouTube, you can learn whatever you need to.
After about 30-40 hours of work and $30 of investment, I launched TrippieAI.com. Honestly, about 80% of that time was spent creating fun icons and fixing about three persistent bugs. Let that sink in. Someone with zero coding background and $30 in their pocket was able to build and launch an application that uses cutting-edge AI to produce a custom roadtrip itinerary.
The app can take into account your camping setup, vehicle type, off-road capabilities, whatever activities you can think of, and gives you usable itinerary/routes/coordinates to all of it. When a new itinerary request comes in through the app, I do nothing, everything is automated. I just have to make sure shit don’t break (or more specifically, fix it when it does). Admittedly, it’s not perfect and it’s still fairly simple, but it’s functional.
What would you build if you knew you couldn’t fail — or if failure only cost $30 and a weekend?
The Technical
[If you don’t care about technical stuff, skip down to The Philosophical & Political]
If you’re not a technical person and still want to read this section, don’t let it discourage you in trying to build a project of your own. The language below can be intimidating, but I only knew 30%-40% of it a week ago. You will figure it out, just get curious and be patient. Use YouTube and ChatGPT to help you. The beautiful part of ChatGPT is that you can ask it the dumb questions, like thousands of them, with zero judgement and an infinitely patient response.
Below is the tech stack, the costs, and some considerations.
Front-End
Framer ($20/month) – Web hosting, drag-and-drop UI design. I started with Typeform for gathering inputs and then moved to free Google Forms, but ended up using the vibe-coded custom code developed with ChatGPT. If I did a project like this again, I’d just vibe code the entire thing and host it for free on Wordpress.
React (with Framer Motion) – This was all written with ChatGPT and brought into Framer. It chose React, it wrote the code.
Tailwind CSS (indirectly via Framer) – Used for styling within Framer components. Again, all ChatGPT.
Back-End / Logic
Make.com (currently free) – Used for automation workflows triggered by form submissions or sheet changes. Maps and feeds user inputs to ChatGPT API with prompt, parses ChatGPT output to create .kml coordinate file, creates attachment, sends to Google Workspace for email compilation and send. This tool is the spine of the entire app and a super powerful missing word. Their free tier is also very generous and fairly easy to learn.
ChatGPT API (~5 to 20 cents per run) – Used the 4.1 model (cheap, fast, and thorough enough) to take user inputs from Google Sheets via the app script/Make custom webhook and produce itinerary and .kml content. Model selection makes a big difference and token limits gave me issues, so just set the limit high.
Google Apps Script – For handling form submissions, watching changes in Google Sheets, and acting as a webhook endpoint. Again, all written by ChatGPT. The terminal/console for this took some time to learn in order to start for debugging and running tests.
Google Workspace ($14/month) – Provides Google Sheets, Google Drive, and email client. Does a lot more, too. If I were to run this stack again I might try Gemini, given the integrations with Google Workspace.
Storage & Data
Google Sheets – Used as the primary data store for form responses and app data, serving as both a database and trigger source.
Google Drive – Houses .kml coordinate attachments and produces links and email attachments.
APIs & Webhooks
Webhook-based architecture – Passes form data to Apps Script endpoints, which then forwards or transforms it using Make.com scenarios. I tried some of the pre-built Make modules for this and they sucked. Just have ChatGPT create the app script and do a webhook.
ChatGPT API – see above.
Dev & Debug Tools
Safari Developer Tools – Used for inspecting payloads and debugging API requests/responses.
Console logging in Apps Script – For lightweight ‘server-side’ debugging.
ChatGPT Code Editor – So good. Easy to use, integrates well with ChatGPT. However, it has a tendency to rewrite the entire code base WAY too often. I want to try the new ChatGPT codex tool and a Git repo. I played a little with Windsurfer (like Cursor and Replit), and those seem really good. There were times where I hit a wall with ChatGPT on debugging things, so I tried Anthropic’s Claude and Claude solved the problems on the first try; I see why Claude gets such acclaim from developers. However, the context window wasn’t great.
A Few Other Learnings
Use Git and Github. I didn’t think I would be vibe coding at the start, so I didn’t use it. So dumb, such a mess.
Use an IDE. Windsurfer, Cursor, or Replit will help you get past some of the buggy stuff with ChatGPT. I had it crash multiple times, hit context window limits, rewrite or erase code way too often just using the native code editor.
At this scale, ChatGPTs API is cheap. I thought this would be one of the more expensive parts of the build but I’ve spent less about $3 running a decent model and a high token limit. With testing and actual users, I’ve had about 100 responses.
Use high token limits at the start, since it’s so cheap at a small scale.
Make.com is awesome helping you connect a bunch of stuff together, it’s free/cheap and real easy to use.
Try vibe coding. It was fun and surprisingly easy. I hit the limits of flexibility on prepackaged tools (e.g. Typeform) really quickly. Those types of tools can be easy at the start but as soon as you want it to do something that isn’t preprogrammed, you are SOL.
The Philosophical & Political
[Gets on soapbox]
I hope my TrippieAI.com example illustrates just how powerful AI technology can be. A year or so ago, this app would not have been possible for me to build. The project would have required a designer, a mediocre full-stack developer, multiple weeks of development/refinement, the cost would be a couple of orders of magnitude more, and the output would have been much less valuable and dynamic.
It wasn’t since 2008 — when the first iPhone came out — that I got this excited about a technology, and never has a new technology made such an impact on my daily behaviors.
A few quick use cases of other ways that I’ve already used AI: understanding nature (animals, rock formations, natural history, etc.), creating images, thinking through business ideas, as a career/life coach, help refining recipes, as a debate sparring partner, summarizing books, interpreting medical test results, valuing a used car, troubleshooting bad internet…I use it a lot. But these are all meaningless use cases in comparison with the true power of AI, for projects like new protein development, utilities optimization, large scale automation, discerning patterns from big data, digital twins…the list goes on. These are the things that hold much more promise than my entrepreneurial aspirations and curiosity about lizards.
We came across this little gal on a hike during our trip. While Laura was (successfully) trying to pet its tiny head, I took a picture and uploaded it to ChatGPT. Turns out that the greater short-horned lizard gives live birth and can shoot blood out of its eyes!
With all that promise, there is also some dread. What happens if people are no longer the means to production and become effectively worthless to the economy? What jobs will go away, and across how many industries? (The answer is a lot and a lot). Will it be the “hard-takeoff” or the “soft-takeoff”? Will the labor market fully recover? Who is going to dominate the market, USA or China, and do we have the infrastructure or political constitution to win? These questions don’t even begin to touch on concerns around misinformation, governmental surveillance, AGI threats, and other nefarious uses.
These questions and the impacts of policy are easy to ignore because the space is hard to understand and moving at an insane speed. What we DO know is that the AI cat is out of the bag and there is no putting it back in. The bubble might pop tomorrow, but AI is here to stay.
AI will have profound impacts on how we consume information, the means of production, how/what personal information/data is shared, ideas of privacy, labor markets, energy needs, environmental impacts, infrastructure investment, national compute capacity, the focus and importance of education, healthcare delivery, economic inequality, and more. Do a little reading on this and you will quickly see that we need very different governmental focus and overall policy when it comes to investment in education, infrastructure, and energy production to have a fighting chance against China.
I’m probably about as much of an oracle as I am a developer (not great at either). But after watching this space go from theoretical to paradigm shifting in a couple of years, I am convinced that ignorance or ambivalence toward AI isn’t a choice.
So start by playing around with it; it’s fun and powerful. Bring your app idea to life or learn about that lizard. Do that for 3-6 months and you won’t just see how fast it’s moving, you’ll feel it. Then do the exercise of thinking about where we will be 3-6 years down the road.
The story of AI is being written, and it’s your choice whether you want to be an author or a reader. But either way, you’re a character.
[Gets off soapbox]