Leaving them was always possible, just never worth it.

But yesterday, I spent four hours on DIY data exodus to cut my monthly subscriptions in ~half (“hey claude, can you figure out how to do a lossless export of everything I’m using this platform for, with no downtime and safe rollback?”).

If others are doing this sort of thing — and I have to imagine they are, since the ROI was splendid — then it’s a major hit to the biz model concepts of data lock-in and feature bundling.

For example: I’m no longer using Circle. Although this post is published via Circle as a final heads-up for anyone subscribed, it’ll land on a static Jekyll site containing all historical content at the same URLs, plus RSS. I used to benefit from Circle’s whole “bundle” and was happy to pay. But now I’m just using it out of inertia as a pricey CMS. Migrating would have cost me a week, which never felt worth it. But now it is. Not because they made it easy (it still required a data dump + API + scraping), but because Claude made it easy.

I’ve also noticed myself picking apart bundles. If 80% of a bundle is trivially re-creatable, I’m finding myself no longer valuing that 80% in my mental math. I’m still happy to pay for the killer features, but the nice-to-haves have stopped seeming so nice. (For example, I’m currently in the process of unbundling my accountants.)

Webapps are feeling increasingly like friction that I’m working around in an attempt to get at the one thing the service does uniquely well. More and more, I’d rather pay a premium for narrow tools with clean APIs and good integrations. Give me an itemized invoice of services I can mix and match, not a monolithic bundle.

Yeah, that’s supposed to be suboptimal from a business model perspective. But is it? Now that I’m off Circle, I’m not going back. Whereas when I went to do the same cleanup to my convoluted registrar/dns setup, DNSimple gave me a lovely itemized invoice, and I was able to cut my bill by 70% while remaining a happy customer. When I’m ready to scale my spend back up, it’ll be with them.

Unbundle yourself, or your customers will do it for you.

A buddy of mine who runs a ~20m/y bootstrapped software business said their customer feedback has started becoming a lot more demanding. It’s a shift in tone and expectation. Customers are waking up to the fact that they can now DIY a lot of what they used to have to accept as an annoyance of the platform. The baseline used to be that your product could do something valuable. Now it has to be able to do that thing better than Claude.

It’s not just webapps. I spent ages tweaking my OpenClaw to be safe and useful, so much so that I thought I was going to be locked into it forever. But I had its full state already sitting in a git repo, and it turned out to be a ten-second job to losslessly hibernate it and cancel the VPS.

The whole assumption behind subscriptions is that folks aren’t gonna unsubscribe on a whim. Subscriptions are the beneficiaries of inertia, friction, and convenience.

It’s always been far easier to keep paying than to stop.

What happens when it isn’t?

product feedback isn’t supposed to tell you what to do next. (although it sometimes can.) it’s supposed to help you see the truth of what you’ve already done.

even earlier than that, pre-product feedback isn’t supposed to tell what you to do either. it’s supposed to help you better understand the people you’re trying to build something for.

feedback isn’t about abdicating your creative decisions. it’s about grounding yourself in reality such that you are able to start making better ones.

you’ll still be making your creative leaps of faith. you’ll just be doing so from solid ground.

Vanity engagement is anything that shows up in your metrics dashboard without correlating to member success.

A simple example is shallow-but-active forum posts about tools and tactics. In the Nonfiction Authors’ OOC, despite actively muffling (or at least 🔒not incentivizing) these types of conversations, the three highest “engagement” posts right now (purple highlight in the image below) are all about exactly this sort of trivia. Folks are contributing to these threads not because they are important, but because they are easy:

If we blindly followed the metrics (believing that quantity of discussion equals quality of impact), then we’d end up doing what every content site ends up doing: lots of tactical, technical, ephemeral opinion-fests. We would become a place where people talk about writing, as opposed to a place where people write; the more serious authors would drift away and the death spiral would begin.

In an OOC, practically any metric can become vanity engagement: educational content; event attendance; weekly check-ins; newsletter opens; forum contributions; etc.

It’s not just the 🔗bike-shedding posts either. Our most active forum users are not necessarily our most successful authors, and some of our zero-engagement members have gone on to successfully finish great books. So you could argue that basing any of our decisions on the pursuit of forum activity is an example of chasing vanity engagement. Even our Writing Accountability Groups, which do correlate with member success, shouldn’t be blindly maximized. (Non-attendance often just means that a member’s writing is going great, or that their kid is home sick, or any other number of reasons far more important than our own little metrics.)

You don’t want these metrics to be fully zero (and if they are zero, you probably have a problem that requires emergency intervention, since it means the 📘system is broken), but you also don’t want to be maximizing them, or really letting them influence your decision-making at all.

From 📝the early manuscript:

The appeal of chasing vanity engagement is that it feels weird doing nothing. But as we know from watching 🔗interventionistas in all fields, doing “something” often carries second-order consequences.

(Note: If you folks building “community” on twitter, and even on their own sites and forums, they’re frequently focused on creating exactly this sort of vanity engagement via polls, question of the day, ephemeral content, and other ways of making numbers to go up without helping members go forward.)

The “something” you should be doing, in an OOC sense, certainly isn’t to drive shallow engagement via easy questions 📘or loads of “just-in-case” content. Rather, it’s to grok the 🔒members’ goals and context, understand the workings of the system you are incrementally building, and figure out how to 🔒help with the hard part.

Different members will experience different aspects of the journey to be most “hard.” But your OOC isn’t supposed to help with every step of the journey for all possible people.

By planting a flag, picking a fight, and positioning yourself to attract a certain type of member, you render some parts of the journey “easy” (for your particular members) and other parts “hard” (for your particular members).

Which allows you to ignore the easy bits and focus on where your interventions and support are most valuable. This is why pinning down your member context is a precondition for the rest of OOC design.

*Example: In my authors community, members tend to know what they want to write about and how to write clearly, so we don’t spend any time “helping” with those areas (if we tried, it would just be busywork). 

They are, however, quite busy with jobs and kids, and they face practical challenges with making the time and doing the work, which we help with via live working sessions. *

But if we positioned ourselves to attract a different type of author, then their “hard part” would be totally different, which means that our community design would change as well.

If you want to help in a specific, pre-determined fashion (e.g., working sessions, courses, knowledgebase, group challenges, masterminds, etc.), then you’ll want to pick a journey and positioning to attract folks who stand to benefit from that particular type of help.

Conversely, if you care primarily about who you are helping, then it’s prudent to be a bit agnostic about how exactly you’re going to do so (since it may turn out that the deep knowledgebase doesn’t actually help with their particular “hard part”).

Imagine OOC design-space as having three dimensions: you’re helping 1) some type of person, 2) along some journey, 3) via some sort of support structure. (Who-where-how.) 

In order to give yourself enough flexibility to design something good, you’ll need to be able to relax at least one of those three constraints. 

Example (cont): With the authors group, we were strongly opinionated who and where, but agnostic about how, which gave us the flexibility to design/discover ways of helping that aligned with the flag we planted, the fights we were picking, and all other aspects of our positioning.

If you going into it with all three things (who-where-how) tightly defined, it’s very possible that you end up putting a ton of effort creating support that doesn’t actually help with anybody’s hard part.

start with the fewest possible spaces

When setting up a new community home (i.e., forum), the common mistake is to start with far too many discussion spaces divided into far too many little topics.

At launch, an OOC really doesn’t need (or want) too many forum spaces:

• One for discussion (two at a stretch)
• One for onboarding (“start here”; designed to drive them toward taking the first meaningful action)
• If relevant, a couple for organizing your own stuff (knowledgebase, events, etc.)

Early on, multiple discussion spaces must fight each other for critical mass, and it’ll end up feeling like a cluster of little ghost towns instead of a single healthy whole.

A members described the experience of this as:

Walking through a hallway full of closed doors, where you have to open each one to see what’s in there—and usually there’s nothing.

With lots of spaces, when members want to post something, they face a moment of friction and doubt, wondering whether they’re posting to the “right” place.

We indulge in creating loads of spaces, I think, as a reassuring sort of busywork: shuffling things about, organizing, and imagining what everything might eventually look like at scale, without dealing with the scary, all-important work of beginning to find and serve real members. (It’s like procrasti-cleaning, but for your OOC home instead of your home.)

Beyond the above issues (a cluster of ghost towns, hall of closed doors, friction to posting), there’s also the fact that you don’t even know what the right spaces are going to be—because your members haven’t shown you yet. 

desire paths / organic space design

Public spaces are designed with paths in very rational places, but people tend to find better routes. these unplanned, well-trod routes are called “desire paths.”

There’s a story of a clever university architect who didn’t put in any paths on a new campus, but instead laid grass everywhere, waiting to see where students walked, and then paving over the desire paths to formalize the emergent behavior. 

We can follow the same model for our earliest OOC discussion spaces:

1. Begin with as little structure as possible—a single discussion space (plus any non-discussion utility spaces)
2. As conversation levels grow above critical mass, “pave the desire paths” by creating a new space to support the strongest, most desirable emergent behavior you’ve observed
3. Give the new space an anti-ghost-town boost by going back through archives and moving all relevant discussions into their new, dedicated home
4. Observe and repeat from step 2

This leads to community homes that feel organic and progressively designed by the inhabitants (think medieval European cities) instead of artificial and designed all at once by some intelligentsia (think American suburbs). (This concept is via , with more detail in this comment and thread**.)

Example: fitness community begins with one discussion space. Before long, you notice members swapping detailed notes on extremely in-depth workout plans; that’s a desire path. If it’s something you want to embrace and encourage, and if you think enough people are into it to support critical mass in a new space, split it off and move over any historical conversations. 

other thoughts

• Muffling: sometimes, a members’ desire path is undesirable to the OOC’s goal at large. in that case, you can split it into a space with reduced notifications/visibility in order to muffle it from non-interested members without explicitly censoring or censuring. See: reframing moderation as signal vs. noise.
• Nouns vs. verbs: when we design spaces, top-down in advance, we tend to organize around nouns: topics, tags, etc. But in practice, some of the most powerful spaces are built around what people do there: the interaction, the activity, the verb. I don’t think this is possible in every scenario, but it seems strong when it happens, so worth staying open to. (Slightly more discussion in this comment.)
• Larger initial audience: if you’re starting with an existing audience rather than building up from scratch, then a single discussion space might be a little noisy, and you might decide for more than one. Still, I’d suggest erring on the side of too few rather than too many, since you can split spaces more easily (and productively) than you can collapse them.

What else?

E.g., last week in the authors’ group, while giving their new member intro, someone said something really killer about their launch plan, which really ought to be a high-value discussion on its own! But that discussion never happened since it was buried within an intro. Also here, there’s often amazing stuff deep in the discussion thread, whether a huge resource that deserves to be unpacked, or an interesting change of topic or new idea.

In my head, I really want something like a “quote tweet” behavior, where we can easily grab a piece of some other post, with credit, to bump it back up to a top level item discussion. Saying that, I guess retweets on twitter originally started as an emergent behavior (“RT @…”) rather than an explicitly supported one. 

I might start trying to do it that way in the two groups and see if anyone else follows along… If so, any ideas on the correct format/structure/indicators of it? Any other misc. ideas or suggestions?

Ralph Ammer’s notes on how to draw ideas align nicely with my own approach to writing/thinking: lots of messy, throwaway scribbles, doodles, and improvisations. May your waste paper basket overfloweth:

Now, let’s talk AI.

ChatGPT is really, really good. Try it and be amazed. It’s one of the reasons I’ve been writing less. I don’t want to write anything that could have been written by the AI, and it can now write a lot of things very well. Any summary of existing knowledge is absolutely worthless as human-created content: ChatGPT can do it better.

So what still works? And what will work into the future?

I think Ralph Ammer’s approach, above, is a good one: longer form, more detailed, and multi-media. Raise the production value. He uses animations, doodles, diagrams. His stuff is great.

Another approach, related to the topic of creativity, is to have fresh ideas. Paul Graham is probably the paragon of this, with lots of interesting, new ideas. The other day, I spent six hours working through a single one of his essays (this one), because it was prompting so much thinking in me.

ChatGPT is incredible at answering any sort of “best practice” question (or to generate cookie-cutter “ideas” like this), but I wasn’t able to get it to be truly novel. It’s certainly going to replace a lot of “answer to this question” SEO articles, where a smart language model, similar to ChatGPT, will simply replace 90%+ of search results without ever sending you anywhere. (And that sort of content was pretty meaningless and humanity-destroying to “write,” anyway, so no great loss.) So writing returns, hopefully, to being about real new ideas.

A third approach might be to go the opposite direction with style and aesthetics. As AI’s prose approaches formal perfection, we can go avant garde. What’s the Dada or Cubist take on content marketing? Might be time for a re-read of The Work of Art in the Age of Mechanical Reproduction (1935).

Or maybe wrap more of your personality into it. More video. More vulnerability. More humanity. More of you. I don’t necessarily think that any one of these is enough, on its own, since any can be fairly accurately modelled. But maybe in combination, being less polished and more you across the board, might be viable. (Also: humor.)

Lastly, and I think maybe the most practical for most situations, is to niche down real hard. Language models are, by definition, a bit generic. They give a good answer for the average person, but most people aren’t average. If there’s a way to speak just to them, directly enough that it stands head and shoulders above other advice for that paticular type of person, then maybe that works.

The talk’s opening is largely technical and tactical. But the last three slides just bowled me over, an absolute masterclass in side-project philosophy (images below are from Simon’s talk notes):

Guilt guilt guilt. How terrible to begin a project in earnest excitement and then feel trapped by your ongoing obligations toward it. And what is Simon’s answer?

Slight tangent: I was recently chatting with a couple founders about the the experience of someone signing up for your stuff and then immediately demanding more features. They suggested that people get what they’ve bought, at that particular moment in time, with no entitlement to any future improvements.

I didn’t fully agree, but it took a few days to figure out why.

The reason is switching costs + downside risk.

If I walk into a restaurant that doesn’t make me happy, then I can leave, no harm no foul. If I go so far as to order something, but I don’t like it, that’s still on me. No lock-in, low downside, no obligation. But if I hire a caterer for a large event and then realize, late in the day, that they refuse to serve anything vegetarian… Well, I think that’s different. Given that it’s extremely costly (in terms of finances, time, attention, stress, etc.) for me to switch caterers last minute, and that the downside is very large, I believe that they have do have some obligation – maybe not 100%, but certainly not 0% either – to accommodate my needs.

So rather than saying we don’t owe our customers anything beyond what they’ve currently bought, I think a fairer statement is to say that it depends on the idea: some ideas do carry future obligations; some ideas don’t. And user accounts – i.e., a container for storing your stuff – certainly pushes it toward the “yes obligation” direction.

We can run the heuristic in reverse: if you’re NOT up for ongoing obligations (which is true of side projects, passive income, and experiments), then you should avoid picking ideas that involve high switching costs and/or high downside risk for your users. Don’t collect their data. Don’t lock them in. Don’t even let them make an account!

A lot of founders will do it anyway, because switching costs boost retention, and downside risk (i.e., “being mission critical”) boosts pricing power. But that comes at the cost of either guilt (acknowledging that you screwed up and can’t keep up) or denial (pretending you didn’t even do anything wrong).

In either case, far better to pick ideas that fit your future constraints. The final piece:

Or in other words: wrap it up and tie a bow on it. Documentation is part of the product. So is customer success, onboarding, everything. 

Lots of the above that I need to get better at. Guilt is my primary emotion; it’s the water I swim in. Hearing Simon’s take, I’m optimistic. Yeah, I’m still going to have to deal with the shitty decisions I’ve already made. But I’m already dealing with that, so no biggie there. But now I’ve got a clearer heuristic for what I take on next, and a way to evaluate whether or not I’m going to be able to live up to it like I’d like to. And that’s exciting.

It’s also a stark reminder that the customer journey of community has a ton of touchpoints, ranging from social media and email through to the community tool itself.

If a community is its people moreso than its tools, then anywhere you and those people can interact is technically part of it, no matter how far outside the “official” software it happens to be. This is all doubly important at emotionally critical moments, whether positive or negative, excited or frustrated. (Which is exactly when folks are most likely to email!!!)

Emails are such a delicate thing and so many companies/products seem to treat them with such casual disregard… I think one of the reasons I hold community tools to such an unreasonably high standard is that they are always attempting to send emails on my behalf but without my explicit permission, as if there’s no downside for getting it wrong. Sending emails to a customer’s customers feel similar in importance to handling financial information, deserving of its own set of checks and balances, reassurances and transparency. For example, there’s no way for me to know what Circle is even sending! It’s literally impossible for me to audit, much less to improve.

There’s also some sort of parallel to the overly rational/reductionist worldview where someone says, “oh, only X% of members are asking a question, so it isn’t super important that we reply.” But once you stop replying, it sends a signal to everyone else that questions don’t get a reply, which destroys the perception that a safety net exists. It’s another case where the data would seem to mislead, and where efficiency isn’t. Automations appear to be super efficient, but in a set of critical situations, they’re fiddly and prone to fail – not in a technical, noticeable way, but in an emotional, “evangelist into detractor” sort of way.

For example, when a high school student “churns,” that could represent:

• Graduation (good churn)
• Dropout (bad churn), or;
• The student’s family has moved (neutral churn).

Just as churn doesn’t always signifify something bad, retention doesn’t always signify something good. when someone “retains” for another year of high school, it can be:

• Advancement to the next year (good retention)
• Flunking and repeating a year (bad retention), or;
• Necessary accommodation for a learning disability, illness, or injury (neutral retention)

While a typical SaaS subscriber is supposed to be “forever,” an outcome-oriented user is supposed to be temporary. The idea is to get them to success, as quickly as reliably as possible, and then trust that they’re going to say good things about you. (As your business evolves, you might daisy-chain multiple outcomes together to extend a member’s time with you; but you’d never want to intentionally delay a single outcome.)

For these sorts of outcome-oriented products (as opposed to engagement-oriented products), the key metrics might be:

• Graduation Rate (% of users who reach successful outcome)
• Throughput Speed (how quickly they can do so)
• Referral Rate (expected # of referrals per successful graduate)
• Repetition Rate (harder to measure and not applicable to every outcome-oriented offering, but represents the % chance that, should they reencounter the problem, they’ll return to use your thing again)

If the above is broadly accurate, it would suggest that the growth engine of outcome-oriented products behaves less like the long slow SaaS ramp of death and more like a modified viral loop. (As a reminder, the viral loop is basically the expected # of referrals per user, multiplied by your conversion rate of a qualified lead, accelerated or delayed by the time until referral.)

The delay before referral is important and often overlooked. One of the reasons social games like Farmville could grow so fast was that the referral happened within the first hour or so.

In the case of offering an inherently lengthy process (like a long-form course or community), the time until referral is already going to be quite slow. This means you wouldn’t want to try to “extend retention” by delaying value in any way. If you can front-load the value, if you can give it to them faster, then even if they do unsubscribe (as successful graduates, i.e., good churn), that faster “loss” in subscriber will be immediately compensating by the faster gain of an extra referrer. 

So I think it’s possible, in this category of businesses/product, to find yourself in a situation where “lower” retention (via faster throughput and high graduation rate) equals higher growth.