AI made me faster. AmongFolk changed how I think about building products.
The code for a private social app arrived at extraordinary speed. The harder work was deciding what deserved to exist, what needed protecting, and what to remove.
On the 19 August, the AmongFolk repository contained an initial backend scaffold.
Less than an hour later, the next commit was titled: “Deploy beta: auth, groups, feed, photos, digests.”
Eighty-two minutes separated the framework from something recognisably like a product.
The commit history accelerated from there. The following day brought voice notes, passwordless authentication, iOS push notifications, video uploads, billing, moderation, App Store metadata and a help desk. Within 18 days of the first commit, the repository had accumulated 258 commits.
Commit count is not a measure of value. That number includes merges, generated builds, refactoring and housekeeping. But the shape of the history still tells a story.
AI has made building software dramatically faster. It has also forced me to reconsider what product building actually is.
A quiet place for people who already know each other
AmongFolk began with new parents and grandparents living at the other end of the country.
There is no shortage of ways to share things online, but most of them seem to push us towards one of two extremes.
There is the group chat, where photos, dates, voice notes and conversations disappear into an increasingly noisy stream. Then there is the public social network, where sharing becomes publishing. You have a profile, an audience and an algorithm deciding what gets seen.
I wanted something quieter.
AmongFolk is a private, invitation-only space for family and friends. People can share photos, video, voice notes, little updates and plans. Everything appears chronologically. There are no public profiles, no discovery, no adverts and no algorithmic feed.
No audience to build. Just your folk.
The proposition was simple enough to write down. Turning it into a real product meant authentication, private media, invitations, notifications, native apps, billing, accessibility, moderation, store submissions and a long tail of details that only become visible once something works.
That is where AI changed the pace.
The speed is real
AmongFolk was not my first experience of this new pace.
The git history for dandoing.dev shows an earlier, smaller version of the same pattern. Its first commit landed at 11:34am on 15 July, co-authored by Cursor. It already contained the main site, services pages, contact routes and a Medium feed integration. Over the next three hours came mail delivery, copy changes, contact form improvements, background fixes and feed repairs. SEO polish followed the next morning.
A website and a native social product are very different things, but both histories show what happens when the cost of turning an intention into working software collapses.
AI lets me move from an idea to something inspectable incredibly quickly.
That matters because an abstract feature is easy to like. A real screen is much easier to judge.
“Shared lists” sounds useful. A shared list sitting inside the wrong tab, competing with the feed and making the navigation feel crowded, is a product problem.
“A calendar” sounds sensible. A calendar component that adds more interface than value can be removed.
“A buy button” sounds commercially necessary. A buy button at the wrong point in the experience can go too.
All three of those decisions appear in the AmongFolk history. The Updates tab became a Group tab with search and shared lists. An event calendar component was removed. Later, there is a wonderfully direct commit titled “remove buy button.”
This is one of the biggest changes in how I now think about product development.
Code has become a sketching material.
From specification to conversation
In a traditional delivery environment, implementation is expensive enough that teams try to resolve uncertainty before development starts.
We write requirements. We estimate. We debate edge cases in meetings. We ask whether the work deserves a place on the roadmap. All of that is rational when building the wrong thing could consume several people for several weeks.
AI changes the economics.
For a certain class of decision, it can now be cheaper to make a working version than to hold another meeting about the hypothetical one.
That does not mean planning has become useless. It means the useful level of planning has changed.
I still need to be clear about the problem, the constraints and what good looks like. In fact, that clarity matters more because AI will happily produce a large amount of plausible implementation around a vague instruction.
But I no longer need to pretend I can reason my way to the perfect interface in advance.
I can describe the intent, build a version, put it in my hand and react to it. I can discover that a modal feels wrong, that a label is too technical or that a feature belongs somewhere else. Then I can change it while the context is still fresh.
The product process becomes a conversation between intention and reality.
The loop is no longer just think, specify, build and release. It is think, make concrete, inspect, adjust and repeat.
That is not carelessness. Done well, it is a faster route to care.
When code arrives faster than judgment
The uncomfortable part is that AI does not only accelerate good ideas.
It accelerates everything.
The AmongFolk log contains commits fixing migration order, authentication state, privacy controls, media handling, notification behaviour and mobile build problems. One commit reconciles two independent fixes for the same cache issue. Another fixes “the Android defects the iOS build had hidden.”
There is even a commit titled “Removed bad word added by AI.”
That may be my favourite one. It captures the entire relationship in six words.
AI can generate useful code at astonishing speed. It can also introduce the wrong word, make an assumption that does not fit the product or produce something that looks finished before it has been properly understood.
Get Dan’s stories in your inbox
Join Medium for free to get updates from this writer.
The faster implementation becomes, the easier it is to confuse movement with progress.
On dandoing.dev, I describe one of the questions I am exploring as where AI-assisted development earns its keep and where it merely makes plausible-looking mess faster. AmongFolk has made that question very practical.
Speed moves the bottleneck.
The constraint used to be my capacity to implement. Now it is often my capacity to review, integrate and decide. The limiting factor is less frequently “Can this be built?” and more frequently “Should this exist, does it belong here, and am I prepared to be accountable for it?”
AI has not removed the need for engineering judgment. It has increased the amount of judgment required per hour.
Constraints are part of the product
AmongFolk deals with private family photos, voice notes and conversations. Privacy cannot be a layer added after the interesting features are complete.
So the project has a written privacy contract.
Media uses temporary private links. There is no discovery system. There are no public profiles. Non-members cannot inspect whether private content exists. There is deliberately no general administrator screen for browsing people’s posts and photos.
Those are technical constraints, but they are also product decisions.
The absence of a feature can be one of the most important things about a product.
AI is at its best when those decisions are made explicit. It needs rails: conventions, tests, privacy promises, architectural rules and clear definitions of done. Otherwise every new task is an invitation to make a locally reasonable choice that gradually changes the nature of the product.
This is where my experience leading engineering organisations keeps resurfacing.
Most engineering problems are clarity problems. Teams struggle when nobody has said what good looks like, who owns the decision or which trade-offs are unacceptable. AI has exactly the same problem, only at much higher speed.
If the instruction is vague, the output may still look confident.
Product direction therefore cannot live only in my head. It has to be expressed in ways that shape the work. Some of it becomes tests. Some becomes project guidance. Some becomes deliberately unavailable architecture.
The product is not just the collection of things we have built. It is also the set of promises the system is not allowed to break.
Building more makes editing more important
The surprising effect of faster building is that I have become more interested in subtraction.
When implementation was expensive, getting something built felt like the main achievement. Once implementation becomes abundant, selection becomes more valuable.
The AmongFolk history is full of additions: birthdays, recurring plans, shared albums, check-ins, daily prompts, reactions, mentions, shared lists and shared notes.
But the more important product work is often visible in the verbs around them: remove, replace, refactor, clarify, reconcile.
A feature is rarely right because it exists. It becomes right through placement, language, timing and restraint.
A birthday feature in a family app makes intuitive sense. But that still leaves questions. Is sharing a birthday opt-in? Does the year need to be stored? Which group can see it? What should the notification say? Can someone turn it off? Does it feel warm or automated?
AI can help implement every answer. It cannot decide what kind of relationship the product should have with the people using it.
That remains my job.
The founder’s role changes
I do not think AI means one person should blindly attempt to replace an entire product and engineering organisation.
I do think it lets a founder operate across a much wider surface before organisational scale is justified.
In AmongFolk, I have been able to move between backend behaviour, mobile interaction, privacy language, store preparation, support tooling and product copy without waiting for each discipline to become a separate queue.
That changes the early stages of a product.
The advantage is not simply that I can do more. It is that I can keep the idea coherent for longer. A privacy promise made on the homepage can be traced into the storage model. A product phrase can influence the navigation. A support problem can become a product change while the context is still present.
The risk is that coherence exists only in my head, or that the pace outruns my ability to verify what has been produced.
The answer is not to slow everything back down. It is to become more deliberate about where speed is useful.
I want AI to make exploration cheap, repetitive implementation quick and verification thorough. I do not want it making unexamined product decisions simply because it can produce an answer before I have formed one.
What I think product building is now
I used to think of product building largely as turning good decisions into shipped software.
I now think it is closer to creating a sequence of increasingly concrete questions.
- What problem are we solving?
- What promise are we making?
- What is the smallest real version that lets us judge the idea?
- What did using it reveal that the specification could not?
- What should be removed?
- Which decisions need to become permanent constraints?
- What evidence would make us change our minds?
AI dramatically shortens the distance between those questions. It does not answer them for me.
AmongFolk exists at the intersection of two ideas that might appear contradictory: moving very quickly and building something intentionally quiet.
I do not want to use AI speed to create more noise, more engagement mechanics or more reasons for people to perform online. I want to use it to build a small, private corner of the internet for people who already matter to one another.
That feels like a worthwhile use of the acceleration. The code arrived faster than I would once have believed possible. The product is taking shape through the slower work of choosing what all that speed is for.
