--- title: "Designing for Our First Users: What We Expect (and Fear)" description: "What we design for before public launch — how we will find early travelers, what usually breaks, and the feedback we care about most." canonical: https://nowah.xyz/blog/our-first-hundred-users-unfiltered lastModified: "2026-08-07T08:32:24.355Z" --- # Designing for Our First Users: What We Expect (and Fear) What we design for before public launch — how we will find early travelers, what usually breaks, and the feedback we care about most. There is a particular kind of terror that comes with putting something you have built in front of strangers. You have tested it yourself hundreds of times. Then a real person uses it, and everything you thought you knew gets tested again. We have not run a public launch yet. This is how we are designing for the first wave of real travelers — who we will invite, what usually breaks, and what “good early access” looks like. ## How we will find them ## How we found them ![Illustration for this section](https://pics.nowah.xyz/website-media/founder-005-img-1.webp) No ads. No viral loop. No public [Product Hunt](/blog/product-hunt-launch-what-worked-differently). Just hustle. We started with people we knew who traveled frequently. Friends, former colleagues, people from online communities who had expressed frustration with existing travel tools. We sent direct messages. We offered nothing except early access to something we thought was genuinely better. The pitch was simple: "We built an AI that books your travel through a conversation. Want to try it?" About one in three people said yes. The ones who said no were mostly skeptical that AI could handle something as complex as travel booking. Fair enough. ## The first 48 hours The first bug surfaced within twenty minutes. A user asked about flights to "CDG" using the airport code, and the agent interpreted it as a destination name rather than a code. Small thing. Easy fix. But a reminder that real people in private testing do not behave like test scripts. In the first 48 hours, we discovered that people describe trips in ways we had not anticipated. One user asked: "What is the cheapest way to get from here to anywhere warm?" We had not considered the "from here" pattern, where the user assumes the agent knows their location. Another user started a conversation with "I am bored," which is technically a travel intent if you squint hard enough. The agent handled it gracefully, asking where they might want to go, but it was a conversation pattern we had not tested. The bugs were not catastrophic. The agent did not book wrong flights or charge incorrect amounts. The issues were at the edges: unusual phrasing, assumed context, edge cases in how people naturally talk about travel. ## What users expected versus what they got ![Supporting diagram](https://pics.nowah.xyz/website-media/founder-005-img-2.webp) The biggest surprise was the expectation gap. Many users expected a chatbot. They asked simple questions and expected simple answers. When the agent started actively searching flights, comparing prices, and presenting structured options with real prices, there was a visible shift. You could almost hear the "wait, it actually does this?" moment through the screen. Several users tested the agent deliberately, asking trick questions or making impossible requests. "Book me a flight to Mars." The agent handled it with humor and redirected. "Find me a hotel in Paris for one dollar a night." The agent explained realistic pricing and offered budget options. These tests were users [building trust](/blog/building-trust-ai-travel-booking), poking at the system to see if it was genuine or performative. ## The surprising things people said People are unexpectedly polite to AI agents. Within the first conversation, many users said "thank you" and "please." One user apologized for being indecisive about dates. Another said "sorry to bother you" before asking a follow-up question. This taught us something important about product design: users anthropomorphize AI within seconds. They treat it as a person, not a tool. This means the agent's tone, patience, and helpfulness are not just nice-to-haves. They are core to the experience. A brusque or unhelpful response feels like rudeness from a person, not a software error. ## Feedback that changed our roadmap Three pieces of feedback hit hard enough to change our priorities immediately. First, users wanted to see the agent working. Early versions returned results after a brief wait, but users had no idea what was happening during that pause. Were they being ignored? Had the app crashed? Adding [streaming responses](/blog/streaming-ai-responses-real-time-chat), where users could see the agent searching and comparing in real time, immediately improved trust and engagement. Second, users cared more about the post-booking experience than we expected. We had focused almost entirely on the search-to-book flow. But users started asking: "Can I see my full trip in one place?" and "What if my flight changes?" The [trip management](/blog/launching-proactive-trip-management-ai-acts-alone) feature jumped from backlog to top priority. Third, price transparency mattered enormously. The number two frustration travelers report is hidden fees. Users told us repeatedly that seeing a clear, honest price without surprise add-ons at checkout was the single most trust-building element of the experience. ## Month one metrics I will be honest with directional metrics rather than vanity numbers. Booking times were fast. Users were completing bookings in under two minutes from their first message, compared to the industry standard of hours across multiple sites. The booking success rate exceeded industry averages from week one, which validated the core thesis that conversation reduces friction. Users came back. The return rate in the first month was encouraging. People were not just trying the product once out of curiosity. They were coming back for subsequent trips, which is the metric that matters most for a travel app. Satisfaction trended positive from day one and improved as we fixed edge cases. The users who stuck around through the first rough week became our most loyal advocates. ## What we would do differently If I could rewind, three things would change. We would have built streaming responses from day one, not added them after public launch. Visible progress is not a polish feature. It is a trust feature. It should have been foundational. We would have talked to more users [before building](/blog/five-things-before-building-ai-product) certain features. Some of what we built in anticipation of user needs turned out to be wrong. The features users actually wanted were often simpler and more practical than what we imagined. We would have started the [feedback loop](/blog/ai-feedback-loop) earlier. Not just collecting feedback but systematically turning it into product changes on a daily cycle. The tighter the loop between user input and product improvement, the faster the product converges on something genuinely useful. ## The next 100 The first early travelers taught us more in a month than we had learned in the preceding months of building. They showed us what mattered, what did not, and where the real gaps were. Every AI product needs this humbling phase. The lab is not the real world. Test users are not real people in private testing. And the gap between what you think people want and what they actually need is always larger than you expect. If you want to be part of the next wave of users shaping how Nowah evolves, try it. Describe your next trip in a sentence. The agent is smarter than it was a month ago because of the hundred people who came before you. And it will be smarter still because of you. --- Nowah is an AI travel agent that searches and books real flights and hotels through conversation — no filters, no thirty open tabs. [Plan your next trip](https://app.nowah.xyz).