Hi friend,
Two weeks before I launched, I wrote down the number I actually cared about. I wrote it down early on purpose, so that I could not quietly move it later when the result came in looking like something else. The number was conversations. Not visitors, not signups, not ranking.
Orinyx went on Product Hunt on August 1. This issue reports that number, along with the two underneath it, and none of the three are good.
Last issue I asked what in your setup would be genuinely hard to rebuild if your laptop died tonight. That question is still open, and it turns out to pair with this one better than I planned, because both of them are about the difference between what you feel is true and what you can actually check. If you launched something this summer and the number underneath it confused you more than it disappointed you, this one is for you. Hit reply. I read all of them.
🧠 Big Idea
A launch does not tell you whether the product is good. It tells you how many people already knew you existed.
Here is what came back, and I am giving you the shape rather than the digits, because the shape is the part that transfers to your situation. Across the full launch week: double-digit visitors to the site. Single-digit referrals from the launch itself. Zero signups on the waitlist. Zero conversations with anyone who could actually buy.
I want to be careful about what that means, because the temptation is to explain a number before I have understood it. A launch figure is a lagging indicator with an unusually long lag, and it measures the twelve weeks before the launch rather than the launch day itself. Reading it as a verdict on the product is a category error. Publishing it while I still cannot fully account for it is the only version of this worth your time.
The frame I use now is three numbers and two gaps. Traffic, list, conversations. Almost every launch recap reports traffic, because it is the biggest one, and then the person who wrote it feels vaguely bad for a month without being able to name which part went wrong. The totals are not the report. The gaps between them are.
There are two ordinary diagnoses. If traffic is large and your list is small, your page is not asking for anything, and that is an afternoon of work. If your list is large and conversations are near zero, you collected people who like the idea and have no urgency, and that one is slower to fix and much more important to know, because no headline edit touches it.
Mine was neither, because traffic was small too. That is the third diagnosis and it is the one nobody writes the recap about. When all three numbers are small, the gaps have nothing to say, and the answer sits outside the launch entirely. On July 1 there were not thirty clinical AI founders who knew my name. Nothing I could have done to a page on August 2 reaches backward into July and fixes that.
The strange part is what I wanted to do about it. Not talk to anyone. Change something.
Key takeaways:
A launch is a withdrawal, not a deposit. It pays out whatever audience you built in the weeks before it, and it cannot pay out more.
Write down all three numbers and read the gaps between them, not the totals. The widest gap picks your next month.
If all three are small, stop diagnosing the launch. The problem is upstream of it, and it is a distribution problem wearing a product problem's clothes.
⚡ Quick Wins
1. Freeze the surface for seven days after anything you launch. No homepage rewrite, no new landing page, nothing that changes what a stranger sees. The freeze is the whole point, because it removes the most available way to feel busy and pushes the week's effort somewhere it can be counted. If seven days is too many, do three and pay attention to what you reach for on day four.
2. Write the one question only a real buyer can answer. Put it on your intake form, in your replies, at the top of your DMs. Make it specific enough that someone who has not lived the problem answers it vaguely and someone who has answers it in detail and probably at length. That length is your signal, and it beats a click-through rate, because a person had to spend something to produce it.
3. Separate your three numbers and stop averaging them. Traffic, list, conversations, written down as three lines rather than one summary. Do it for a launch from six months ago if you do not have a recent one. The exercise takes fifteen minutes and it is the entire post-launch review.
None of these need a tool, and I notice that is true of most of what I write here.
🛠 Tool of the Week
Two this week, on purpose. One of them builds the thing. The other one finds the people, and this issue is entirely about the distance between those two jobs.
Lovable
What it is: an AI app builder that turns a plain-language description into a working React app, database and deploy included. The Orinyx site runs on it, which means the blog ships as a code edit rather than a CMS entry. That setup is genuinely good for what I use it for.
How to use it: the free plan gives you 5 daily build credits, up to 30 a month, plus 20 monthly Cloud credits, which is enough to find out whether the thing is for you. The pricing page is unusually honest about what a change costs: "make the button gray" runs about 0.50 credits, and building a landing page with images runs about 1.70. lovable.dev
Why it matters: that price is the entire problem I am describing. Fourteen hours after the launch I opened it and started describing a new homepage, and I got two sentences in before I noticed I was treating a demand problem as a copy problem, because the copy problem has a button and the demand problem has a phone call. So I made one edit instead, and it took twenty minutes: a required question on the intake form asking what their team does today when the model outputs a medication assertion nobody verified. A browser answers that in four words. Someone who has lived it answers in a paragraph and keeps going. The length is the filter, and the rebuild would have eaten the whole day and produced a page I could not prove was better.
User Interviews
What it is: a research recruitment platform. You describe the person you need to talk to, they source and screen them from a panel they say runs to 7.6 million people, and they handle the scheduling and the incentive payments. There is a second product for building your own panel of people you want to keep talking to over time.
How to use it: signup is free and you pay per participant, with the pricing behind a plans page rather than on the homepage, so treat the first search as your quote. Start narrower than feels comfortable. The screener is the useful part, and it is the same discipline as the intake question above: write a screening question that a person who has not lived the problem answers wrong.
Why it matters: I put this next to Lovable deliberately. Everything I could do to my product in the week after the launch was available to me in about twenty minutes, and none of it was the problem. The problem was that I had not spoken to enough of the right people in the twelve weeks before. This is the least glamorous category of tool there is and it points at the actual gap, which is that finding people to talk to is a solvable logistics problem that most of us quietly treat as a personality flaw.
🧭 Product Hunt Radar
Three launches from today, read by someone who was on the other side of this page nine days ago and has not fully recovered.
Remix "Figma, but on your production app." Anyone on the team can spin up a sandboxed variant of the real app by describing the change, preview it, then merge it into a pull request. Every prompt is recorded, and changes get checked against your design system and compliance rules before they deploy. Why it matters: this is the tool trap sold as a product, and I mean that as a compliment to the execution. If you are going to redesign your way out of a hard week, at least do it somewhere that keeps a record of what you changed and why.
oqoqo Build evals and custom benchmarks for real-world tasks. You define the jobs your users actually do, set what success looks like, then run agents against them in isolated environments and watch every step, including what it cost in tokens. Why it matters: it is the same move as writing the one question only a real buyer can answer, pointed at software instead of people. Deciding in advance what counts as working is the part everyone skips.
Paritok A compression gateway for coding agents that squeezes tool schemas, file reads, and stale conversation history before any of it reaches the API. The claim is up to 85% less spend and roughly three times longer sessions, with a recall step so the agent can ask for the original text when it needs it. Why it matters: building keeps getting cheaper, which is exactly why choosing what to build keeps getting more expensive to get wrong.
💻 Prompt for Productivity
Prompt: Help me write the one question that only a real buyer of my product could answer well. My product is: [describe it in one sentence]. The person I want is: [describe them by their situation, not their job title].
Do not write the question yet. Ask me one question at a time until you can answer all four of these, and do not fill any of them in for me:
1. What has this person already tried, and what specifically went wrong with it?
2. What happens in their week when the problem shows up? Who notices, and what do they do right now?
3. What would they have to admit is true about their own setup in order to need me?
4. What could someone say that would sound informed but prove they have only read about this problem?
Then write me three candidate questions. For each one, show me the four-word answer a browser would give and the paragraph a real buyer would give. Tell me which of the three has the widest gap between those two answers, and say why. That gap is the whole point of the question.
WHEN TO USE IT: right after any launch, campaign, or post that sends you a burst of people you cannot tell apart.
WHAT IT DOES: it stops you writing a qualifying question that everyone can answer. Most intake forms ask things a polite stranger will happily fill in, which is why they filter nothing. Optimizing for the gap between the lazy answer and the lived answer is what turns a form field into a signal.
TIP: make it required, and never make it multiple choice. The length of the answer is half the information.
🌎 In the News
Rippling built an AI spending tool after its own bill got away from it my POV: Rippling was on track to spend 40% of its R&D headcount budget on AI tokens, one engineer spent $50,000 in a month, and 10 to 15% of staff drove about 60% of the total. Their CPO's line about the finance presentation was "We were incredulous." What I keep sitting with is that this is a company with a CFO, and they still did not know until someone put it on a slide. A number nobody is looking at keeps growing at whatever rate it likes.
OpenAI says it slowed Astra's development over security concerns my POV: a different and later story than the eval incidents I wrote about two weeks ago. This time the finding came before release: the model performed well enough on agentic cybersecurity that OpenAI said it "cannot rule out Critical capability level at this time," and paused work rather than shipping and watching. Publishing a result that costs you a launch date is the version of this I want to see more of, and I notice that is exactly what I am asking of myself in this issue.
OpenAI's smart speaker is reportedly a $300 to $400 device my POV: Bloomberg reports a donut-shaped thing with moving parts, designed with Jony Ive's studio, expected in 2027. Reported, not announced, so hold it loosely. The part worth noticing is the strategy underneath it. The company with the largest distribution advantage in the industry is spending it on getting into your kitchen, while the rest of us argue about headlines. Distribution is the whole game, and it is bought early or not at all.
🔧🎓 Workshops, Events & Happenings
One of these is mine, and one of them is open right now and is the most founder-shaped thing on the internet this month.
Claude Loves Lovable. Five-part class, starting in September. Sign-up dates coming. This is the thing I have been building all summer and have not talked about here yet. Five live 90-minute working sessions, one a week, in the order a real product actually comes together: brand, then pitch, then MVP, then website, then code. Every session pairs Claude for the thinking and the copy with Lovable for the building, because the handoff between the two is the part nobody teaches and the part everyone gets wrong. You build alongside the demo rather than watching one, and you leave each session with a working Claude skill you keep and reuse on every project afterward. That last part is the bit that outlasts the room. No coding experience needed. Dates and the sign-up link land soon, so keep an eye on this section, or reply to this email and I will make sure you hear first.
RevenueCat Shipaton 2026. August 1 to September 30, 2026. Online. Over $1 million in total prizes, $700,000 of it cash, with a $100,000 grand prize. The rule that makes it interesting: your app has to be released for the first time during the hackathon window, and it has to integrate payments. It is a two-month hackathon about shipping something that actually charges money, which is a more useful constraint than any of us apply to ourselves voluntarily. Seven weeks left. Enter here
Arm Create: AI Optimization Challenge. Closes August 14, 2026. Online, $8,000 in prizes. Four days left and over two thousand people already in it. Narrow and technical, which makes it a better weekend than a broad one. Register here
CUTC: Transform Hackathon. August 8 to 15, 2026. Online. Tagged low-code and AI, run by the Canadian Undergraduate Tech Conference. Small, five days left, and a soft landing if you have never entered one. Details here
TechCrunch Disrupt 2026. October 13 to 15, 2026. Moscone West, San Francisco. Far enough out to plan for. The Builders stage is the one aimed at people at the stage this issue is about. The founder's guide is here
✍️ Personal Note
I froze the site the day after the launch. No homepage edits, no new landing page, nothing a stranger could see, for seven days. I would like to tell you it was discipline. It was closer to removing the car keys, because I did not trust myself to sit with a bad number without doing something expensive about it.
What surprised me was how quickly the urge arrived. Not an hour of grief and then a plan. Fourteen hours, and I was already typing a prompt to rebuild the thing. The page was never the reason nobody came. On July 1, nobody knew who I was in that world, and I have known that for months and preferred the version of the problem that has a button.
So I am doing the slow one now, which is talking to people one at a time and being findable in the places clinical AI people already are. It will not produce a screenshot. It is the only thing that changes the number.
Also, my newsletter list is eleven people. You are one of them, and I mean that warmly and slightly apologetically.
📬 Closing CTA
If you launched something this quarter and the number underneath it is confusing rather than bad, that is the conversation I like having most. Book a Vibe Check. It is twenty minutes, there is nothing to buy on it, and I will tell you plainly if I am not the right person.
And one direct ask, which I have earned the right to make only because I just showed you the numbers. Orinyx is the reason this issue exists. It sits between a clinical AI tool and the patient record and checks every medication recommendation against FDA labeling and federal safety guidance before a doctor signs off, because a system that grades its own homework is not a safety check. It is still early and it is still small. The Product Hunt page is here. Follow it, and you will see the build as it happens, including the parts that do not work. If you know someone inside a hospital or a clinical AI company who should be arguing with me about this, forwarding them this email would do more than a hundred visitors would.
💌 One Ask
What is the widest gap in your three numbers right now? Traffic to list, or list to conversations. Reply with just that, one line. I am collecting them, and the answers are already teaching me more than my own launch did. 🌿