Morning guys, happy Saturday.
I think we still talk about startups as if building the product is the main event. That made sense when software was expensive to make, engineering teams were difficult to assemble, and a decent first version could take a year. It makes less sense when one person with good tools can now get surprisingly far in a weekend.
The product still matters, obviously, but more of the difficult work seems to be moving around it. Which market deserves your next five years? How do you get the first thousand people to care when you have no audience? Which tiny choice in the interface quietly determines what users do? What happens when AI can implement almost anything you ask for and the real question becomes whether you should have asked for it at all?
That shift makes some older startup stories more interesting, not less. Paperform launched before generative AI, but the way it used AppSumo looks almost tailor-made for the current environment. SavvyCal spent months choosing a market before building. And Kieran Klaassen’s work at Every shows what happens at the other end, when the implementation itself starts disappearing into the background.
Sponsored by…
The fastest teams stopped typing.
Your day is Slack threads, emails, docs, and CRM updates. Every one of them starts at a keyboard, and the keyboard is the bottleneck.
Wispr Flow turns your voice into clean, ready-to-send text in any app. Speak a three-paragraph client update in 30 seconds. Dictate the proposal while it’s still fresh. Flowstrips filler, fixes grammar, and formats it so you just hit send.
4x faster than typing. Used by teams at 270+ Fortune 500 companies. Works on Mac, Windows, iPhone, and Android, with Flow for Teams for your org.
Try Wispr Flow Free →
1. Paperform didn’t really sell 3,000 lifetime deals. It bought 3,000 chances to learn.
When Paperform launched in December 2016, founders Dean McPherson and Diony McPherson did something that SaaS people often have mixed feelings about: they ran an AppSumo lifetime deal. In two weeks they sold close to 3,000 licenses, and the cash helped bridge the gap between leaving their jobs and building enough recurring revenue to support themselves. Dean later described the money as a kind of seed funding, except there was no investor and no equity involved. Dean McPherson’s full Paperform retrospective
The more interesting number is 1,000. At least that many buyers redeemed their accounts on the first day and started trying to use Paperform for actual work. Suddenly two founders who had almost no owned distribution had a thousand people poking at the product, asking for features, finding bugs and discovering use cases the founders might not have thought of themselves. That is a very different thing from having 1,000 people on a waitlist telling you the landing page looks nice.
AppSumo buyers have a reputation for being demanding and price sensitive, which sounds like a terrible customer segment if your goal is maximizing revenue per user. It sounds much better if your goal is finding every weak point in a young product. Dean’s point is basically that the qualities founders complain about made those buyers useful early users, because when something was broken or missing, they were very happy to tell Paperform about it. When thirty people independently asked for the same thing, prioritization became much easier than sitting in a room guessing at the roadmap.
I think the interesting way to look at the deal is that Paperform traded future revenue from one cohort for three things it badly needed at that moment: cash, distribution and product feedback. The lifetime price looks cheap only if you evaluate it as a normal subscription sale. If you compare it with raising seed capital, buying thousands of users through ads, or spending a year building without enough feedback, the economics look different.
That does not mean lifetime deals are secretly a brilliant business model. Paperform stopped running them as the company matured because repeated discounting can change how the market perceives the product, and every lifetime customer creates ongoing support, infrastructure and product costs. The same channel that makes sense when nobody knows you can become actively unhelpful once your problem changes from “please try this” to “this is a premium product worth paying for every year.”
There is a bigger distribution lesson here. Early-stage companies often act as though they need to build an audience before they can launch, but borrowing somebody else’s audience can be perfectly rational when you have nothing to distribute to yet. AppSumo, Product Hunt, marketplaces, communities and partnerships are all versions of the same trade: you give up some economics or control in exchange for access to people you could not cheaply reach yourself.
The important thing is knowing when the trade has done its job. Paperform eventually found that repeated Product Hunt launches produced smaller and smaller bumps, because by then the company had customers, word of mouth and its own distribution. Borrowed distribution is useful precisely because it helps you survive long enough to build distribution you actually own.
2. SavvyCal chose the market before it chose the product
Derrick Reimer’s path to SavvyCal is almost the opposite of the usual “I had an idea in the shower and started coding that night” founder story. After selling Drip and later watching his Slack alternative Level fail, he spent months deliberately looking at markets before writing the first line of SavvyCal. His criteria were fairly boring: proven demand, recurring revenue and enough room for a meaningful point of differentiation. Derrick Reimer’s account of choosing the SavvyCal market
Scheduling passed the test because people were already paying for scheduling software, subscription economics were obvious, and Calendly had proven the market was real. Reimer was not trying to convince the world that scheduling software should exist. His insight was much smaller: sending someone your booking link can feel weirdly one-sided, so SavvyCal made the recipient feel more involved by letting them overlay their own calendar and compare availability more naturally.
That is a less romantic way to start a company, but maybe a healthier one. Founders spend enormous amounts of time thinking about product risk because product risk is fun. You can design around it, prototype it, argue about features and use new tools. Market risk is less satisfying because sometimes the answer is simply that people do not care enough, the budget is too small, or the incumbents already solve the problem well enough.
AI makes avoiding that question even easier because building the wrong thing has become much less painful. You can have a working prototype quickly enough that it feels wasteful not to keep going. A month later you are polishing onboarding for a market you never really established was worth entering.
Reimer’s story is useful because the market was part of the product decision. He was also choosing what kind of company he wanted to run: bootstrapped, small, profitable and durable rather than a giant venture bet. That changed which markets were attractive before he ever got to features, and six years later SavvyCal is still a small team with a five-figure monthly recurring revenue business.
There is probably more value in that constraint than founders admit. “What can I build?” now has almost infinite answers. “What market fits the company I actually want to spend ten years running?” cuts the list down very quickly.
3. One extra click can matter more than another page of persuasion
A field experiment by researchers from Harvard Business School and Boston University looked at cookie-consent interfaces and found something that anyone who has designed software probably suspects already: tiny changes in choice architecture can move behavior a lot. When cookie options were placed behind extra clicks, users shifted toward whatever choice was easier to reach, and many people later showed substantial confusion about what the defaults had actually been. Read the NBER paper on cookie consent and choice architecture
What I like about this study is that nobody’s underlying opinion about tracking needed to change. The interface changed, so the behavior changed. That seems obvious, but product teams still spend a lot of time treating every user action as evidence of preference rather than asking how much of it was produced by the path we put in front of them.
This is why defaults are such an underrated part of product strategy. The default onboarding path, notification setting, pricing option, dashboard view or sharing permission quietly becomes the way most people experience the product. If one route requires three decisions and another requires none, you should expect the second route to win even when users would tell you they prefer the first in a survey.
There is an ethical line here too, especially when the easier path benefits the company more than the user. The lesson from the research is not “hide undesirable choices behind extra clicks.” It is that interface design is doing more persuasion than we like to admit, which means defaults deserve the same scrutiny as copy, pricing and feature decisions.
For a good product, the useful question is probably whether the easiest path is also the path most users would choose if they fully understood the tradeoff. When those align, reducing friction is genuinely helpful. When they do not, you may have found a growth lever, but you have also found the sort of growth lever customers eventually resent.
4. Gumloop’s ad worked because the ad was a story about not buying ads
Gumloop CEO Max Brodeur once paid a trumpet player to perform outside the company’s office holding a Gumloop sign, then expanded the idea to more musicians around San Francisco during a conference. The post that travelled was not a detailed explanation of Gumloop’s workflow automation product. It was the much more retellable question: what if we spent our marketing budget on musicians instead? Tom Orbach’s breakdown of the anti-marketing idea
Tom Orbach calls this “anti-marketing,” and his examples include Sticker Mule spending a portion of its ad budget handing cash to strangers and Reddit buying only five seconds of Super Bowl airtime because it could not afford the conventional version. The common mechanic is not really that these companies stopped advertising. They spent money creating something people would voluntarily explain to somebody else, which gave the spend a second distribution layer.
That distinction matters because buying reach and creating a story are different jobs. A normal ad gets the impressions you purchased and hopefully converts some percentage of them. A strange enough marketing act can get the initial impressions and then produce screenshots, conversations, press, reposts and people like me writing about it later.
The obvious danger is that once every company starts “spending its marketing budget” on quirky stunts, the mechanism gets tired too. Anti-marketing only works while it feels like somebody broke the normal pattern, and copying the visible stunt without understanding why it travelled will produce the same problem as copying any other marketing tactic.
I think the more durable idea is to ask whether the spend itself can contain the story. Paying for an ad normally buys distribution after the creative is finished. Gumloop made the allocation of the budget part of the creative, which meant the media plan and the message were basically the same object.
That is a useful constraint for small companies because they are rarely going to outspend incumbents. They can sometimes make a small amount of money unusually interesting, and interesting money tends to travel further than ordinary money.
5. When AI builds the software, somebody still has to decide whether it feels good
Kieran Klaassen at Every has been writing about a software workflow where agents handle an increasingly large part of the engineering loop. His article on “polish” describes a surprisingly manual final step: put the working product in front of yourself, use it, notice what feels wrong, tell the agent, then keep repeating until the experience feels right. Kieran Klaassen on polishing software built by agents
What changes here is not that humans disappear from software creation. The human moves to a different part of the loop. If an agent can write the implementation, review a lot of the code and iterate quickly, spending your scarce attention inspecting every line becomes less useful than spending it deciding whether the finished thing behaves the way you actually want.
That sounds almost trivial until you try to automate taste. A test can tell you that a button works, but not necessarily that the animation feels cheap, that the information hierarchy is confusing, or that a technically correct interaction makes the whole product feel slightly worse. Those judgments are difficult because they are often built from hundreds of previous products you have used and a fuzzy internal model of what “right” feels like.
Klaassen’s interesting move is turning repeated feedback into reusable rules. If he keeps telling the agent that scrollbars should behave a certain way or animations should resolve toward the click, that preference can be codified so the next feature begins closer to what he wants. The human still judges the output, but the system slowly learns the boring parts of the judgment and stops asking for the same correction every time.
I suspect this is going to become a fairly normal way of working. A lot of knowledge work currently contains two jobs mixed together: producing the first version and deciding whether the first version is any good. AI is attacking the production part much faster than the judgment part, which makes the second job more visible.
That loops back to Paperform and SavvyCal in a way I like. Paperform’s scarce resource was not code, it was enough real users to tell the founders what mattered. SavvyCal’s scarce resource was not the ability to build scheduling software, it was choosing a market worth spending years inside. Gumloop’s scarce resource was not access to another ad unit, it was a story people wanted to repeat.
The product still matters, but it is increasingly surrounded by things that matter just as much: market selection, distribution, feedback, choice architecture and taste. As the cost of making software keeps falling, I think those surrounding systems will explain more of why one product wins and another perfectly functional one disappears.
If you’ve been enjoying Token Economy, subscribe and share it with someone who might like it too. Each week, we look at what’s changing in AI, tech, startups, and business, why it matters, and where things might be heading next.