Continuous launch strategy: the launch is a cadence, not a day

You should launch something every four to six weeks, and the reason you don't is that you think a launch has to be big. One launch day a year is a strange way to run distribution for a product you work on every week. You ship continuously and you announce discontinuously, so the announcement has to carry twelve months of work in a single afternoon — and if that afternoon goes badly, you wait another twelve months. Nobody would design that on purpose. We inherited it from companies that had a press cycle and a launch budget, then kept the ritual after dropping the budget.

The alternative isn't launching harder. It's launching smaller and more often, on a cadence you can hold with one pair of hands, rotating which room you walk into so no room hears from you twice in a quarter. This post is the system I run: what counts as a launchable moment, how to scope one to fit six weeks, the rotation that keeps it from being spam, and a thirteen-week calendar you can copy.

How often should you launch?

Every four to six weeks, with a substance gate on each slot. That window is not arbitrary. It is short enough that no single launch has to be decisive, and long enough that something real can be built between them — which is the whole test, because a cadence with nothing new in it is just a posting schedule.

The clearest public example of a cadence run deliberately is Supabase. In their own writeup of how they launch, they describe running Launch Week three times a year on a three-month cycle, shipping one major feature or announcement every day of that week — and, the part founders miss, they say plainly that they rarely hold features back for it. Features ship when they're ready. Launch Week is the amplification layer, an arbitrary deadline they set for themselves because a deadline organizes everything else. They credit that rhythm with 47% month-on-month database growth over eighteen months.

A solo founder cannot ship five announcements in five days, three times a year. But the mechanism worth stealing was never the volume. It is this: the calendar comes first, and the work fills it. A date you committed to publicly generates scope discipline that no amount of intending to ship ever produces.

What does single-launch thinking cost you?

It costs you most of what you build, because everything that isn't the one big thing gets shipped silently. That's the tax, and it's paid in four ways.

You under-use finished work.The integration you spent a week on, the API you exposed for yourself, the free tool you built to test something — each of those is a first impression for someone who has never heard of you, and each went out as a changelog line because it didn't feel like a launch. It doesn't have to feel like one. It has to be new to a stranger.

You put a year of variance on one day. Launch outcomes carry enormous randomness: what else posted that morning, whether someone with reach happened to be awake, whether the feed liked your thumbnail. One launch a year means your distribution is a single sample from a very noisy distribution. Eight means you get the average, and the average is knowable.

You let the crowd go cold. The people who show up for a launch are the people you have been useful to recently. Twelve months of silence between asks means starting from nothing every time, and rebuilding a crowd costs far more than maintaining one.

You stay invisible in the layer that's replacing the click. AI referral traffic to top websites reached 1.13 billion visits in June 2025, up 357% year over year on Similarweb data reported by TechCrunch, with ChatGPT accounting for more than 80% of it — and yet AI still sends under 1% of publisher referral traffic even when it cites the source. The click is shrinking; the mention is the asset. Mentions accumulate from being written about repeatedly, in different places, over time. One annual launch produces a single cluster of coverage. A quarterly cadence produces four.

What counts as a launchable moment?

A launchable moment is any change a stranger could try. That is the whole definition, and it is deliberately broader than “major feature” and much narrower than “anything we shipped.” Run your backlog and your last three months of commits through this inventory once and you will usually find two or three moments you already built and never announced.

The right-hand column does the real work. Without it, this is a license to announce everything — the failure mode people correctly associate with “always be launching.” The gate is not effort, and it is not novelty to you. It is whether a person who has never used your product could do something today they couldn't do last month. A redesign fails that test. A public API passes it easily. Whether a change is big enough to justify a full second launch rather than a moment is a different question, and the eligibility test for it is in how to relaunch your product.

How do you scope a moment to fit six weeks?

Scope backward from the announcement, not forward from the idea. Write the one sentence you would put on the launch page first. If you can't write it, the moment isn't defined yet, and building for three weeks will not define it. If you can, that sentence tells you exactly what has to exist — and, more usefully, what doesn't.

Three rules keep a moment inside the window:

• Sixty seconds, or it's two moments.If you can't demo it in a minute without saying “and we also,” you have bundled. Split it. Two clear moments six weeks apart beat one muddled one, and you get two calendar slots out of the same work.
• Two weeks of build, four of everything else. Building is the part that expands to fill available time. Cap it at a third of the cycle and the rest goes to what actually produces the launch: lining up people, writing, the artifact, and the follow-up afterward.
• Nothing on the critical path you don't control. A moment that depends on another company approving your listing, a partner co-announcing, or an app store review is not a six-week moment. Those are real, and they belong in the slot after next, with slack built in.

One more thing that sounds soft and isn't: put the date in public before you start. Announce it on the launch calendar, tell the three people whose opinion you care about, and let the date do the scoping. The reason the Supabase mechanism works is that the deadline is external and the work bends to it. Working alone, the date is the only external thing available.

How do you rotate surfaces so nobody fatigues?

Send each moment to a different audience. This is what separates a cadence from spam, and it is almost entirely mechanical: every launch surface has a refractory period, some written down and some purely social. A rotation that respects all of them means no single group hears from you more than once or twice a year, even though you are launching eight times.

The six-month figure is not folklore. Product Hunt's help center on relaunching asks makers to wait at least six months between posts for the same product or company, requires a significant update alongside the wait, applies the same gap to products sharing a root domain, and routes early requests to a human reviewer. Read that as scheduling information: Product Hunt can absorb at most two of your moments a year, so the other six need somewhere else to go. The full map of where — which directories take what, and in what order — is the startup launch directories guide, and the working list lives on the launch directories hub.

The last two rows decide whether any of this works. Communities and your own list have no policy, which founders read as permission and should read as exposure — there is no error message when you overdraw goodwill. The founders for whom the fourth announcement of the year still lands are the ones running a build-in-public practice in between, because the announcements are a small fraction of what those people have seen from them.

Why does each launch make the next one bigger?

Because launches leave residue, and the residue outlasts the traffic spike. Four things survive a launch day: the people who tried it, the pages that now link to you, the search queries you started ranking for, and the founders who showed up because you showed up for them. All four are inputs to the next launch, which is why the eighth one is not eight times the work of the first.

The link and citation layer compounds hardest right now, and that's where the timing advantage sits. GoodFirms surveyed over a hundred digital marketing professionals across twenty-plus countries in early 2026 and found that 65% call adapting to AI-driven search their single biggest SEO challenge, 43% are actively implementing generative-engine optimization, and only 14% track AI visibility at all. Most of the field knows the ground moved and is not yet measuring where it landed. A founder producing four fresh, linkable, quotable events a year accumulates exactly the material those systems pick up, at a moment when most competitors are producing one.

The fastest-compounding part, though, is the least technical. Every launch is an ask, and asks are funded by what you did for other people beforehand. That is the constraint that actually binds a solo founder's cadence — not platform policy, not build speed, but whether there is anyone to call. It's why I built Favors.dev as a founder marketing co-op with a points economy: you earn points by doing verified marketing favors for other founders — real testing, honest feedback, reviews from people who actually used the thing — and you spend those points when your own moment arrives. You can't spend what you haven't earned, which turns the quiet weeks between launches into the part of the cycle that does the work. Sorting out which moments actually produced anything is its own discipline, covered in the launch metrics that matter.

What does a quarterly launch calendar look like?

Thirteen weeks, three moments, three different surfaces, one retro. Here is the template I run. Copy the shape, not the specifics — the only structural rules are that build blocks are shorter than they feel like they should be, and that no two consecutive launches go to the same audience.

Two things about this calendar surprise people. The first is how little of it is building — roughly a third. That ratio is not a concession to marketing; it is what a launch actually costs once you stop counting only the code. The second is that the reciprocity block sits inside the build weeks rather than in some separate marketing time. Helping other founders launch is not an interruption to the cadence. It is the part that makes week three work.

Frequently asked questions