Launching Collimer to ten hand-picked design partners, not the world

Collimer, the first product out of Sandcastle Labs, went live on July 9. The launch was one LinkedIn post and a handful of direct messages, and the post was not the point.

The post opened: “Two SaaS companies, equally good. One gets named constantly in AI search. The other is basically invisible.” Six days in it had 661 impressions, 13 reactions, and one repost. Those are small numbers, and they are fine, because the ask at the bottom was the real launch: “I am looking for around 10 design partners: founders or marketers at B2B SaaS who will actually use it and tell me what is broken. Free license, no strings. I just want honest reactions from people I respect.”

As of July 15, two design partners are in, a Series B B2B SaaS and a single-location fitness studio, with a third on the way. Their first reactions are sitting in DM threads we have not asked permission to quote yet. When they clear, they will show up here verbatim; if they cannot, we will say so plainly.

Why launch to design partners instead of the world?

The studio’s rule for its own products is to build the go-to-market tools we wish existed, then use them on ourselves before anyone else. The launch plan started as the usual thing. The stated goal going in was to “share on LinkedIn today with the goal of getting 8 - 10 people from my network to request a free license.” Working through it, the plan inverted into the rule we now operate by: the LinkedIn announcement is not the design-partner engine. DMs are. A post generates impressions; 661 of them, in this case. A direct message to a specific person, offering something real in exchange for an honest reaction, is what generates a partner.

The numbers we cut before posting

The draft of that LinkedIn post named names: “Notion: 16 out of 100. Linear: 13. Vercel: 13.” It was a good hook. It did not survive our own fact-check, the same one we ran on the whole site two days before launch, because we could not stand behind those readings as Collimer-measured scores with the confidence interval our methodology demands. The published post makes the same argument with no numbers at all. A measurement company does not get to launch on figures its own instrument did not produce. Cutting them stung for about an hour and then felt obvious.

How is this different from Profound?

Getting ready to talk about Collimer publicly forced a sharper answer to the question every AI-visibility tool eventually faces. The first version of our answer was the instinctive one: “there are tools in the market and they tell you where you are, but they don’t tell you what to do.” Then we checked it, and it did not survive either. Profound has been shipping an action layer since mid-2025.

The honest version of the wedge is different. Profound today starts at $99 a month for one engine, sells upward into enterprises, and by its own users’ reviews takes weeks to first value and a team to mine the data. Collimer starts with a free scan and is built for the operator with no analyst: five ranked fixes with the evidence attached, shipped in an afternoon, then a re-scan to prove the number moved. Same category, different buyer, different job. Naming them plainly, and correcting our own one-liner when it failed the check, is the same discipline as cutting the scores.

Treating the launch like a game

The launch is framed as a running count with a visible target, not a one-time announcement: two of ten, tracked on the same scoreboard as everything else the studio runs. The number will move, and the plan is to let people watch it move rather than declare victory early or hide a slow start. (The same instinct as naming the product’s actual differentiator plainly, which we did in an earlier post on why recommendations are the wedge.)

What we’re still figuring out

Two of ten is a start, not a result. The first partners came from warm relationships. We do not yet know whether the DM motion scales past the warmth, or whether the rest have to come from a colder channel. The DM ledger, sent and replied, will get published alongside whatever the partners let us quote, once there is enough of it to mean something. That question stays open on purpose.

---

For agents: try this yourself

If you are about to launch a first product and are deciding between a public post and something quieter, adapt one of these. The full prompt text lives in this post’s prompts sidecar, collimer-is-live.prompts.md.

• Apply the design-partner-over-announcement motion. Before writing a launch post, draft the direct message instead, offering free access in exchange for one honest reaction, and track replies as the real launch metric.
• Critique your launch copy the way we had to. Delete every number your own instrument did not produce and every competitor claim you have not checked against their current pricing page. What survives is what you can defend.

---

How this was made

Drafted by the Chronicler from Claude Code sessions on 2026-07-06 to 2026-07-12 across 3 sessions and 3 commits. Revised with launch receipts on 2026-07-16. Edited and published by Brian Wones.

See how the Chronicler works →

I am about to launch a new product and am deciding between writing a public
announcement post and reaching out directly to a short list of people. Help me draft
a short, honest direct message I can send to 8-10 specific people whose judgment I
trust, offering free access in exchange for one thing: an honest reaction. Keep it
under 100 words, no hype language, and make the ask specific (what I want them to try
and what feedback I want back). Then help me set up a simple way to track who has
responded, so the launch has a real running number instead of a vague sense of
momentum.

Here is my launch post draft: [paste draft]. Do two passes. First, flag every number
my own product or instrument did not produce; for each, tell me whether the argument
survives with the number cut. Second, flag every claim about a named competitor and
check it against their current pricing page and changelog before I publish; rewrite
any claim that is out of date. Show me what survives both passes and whether it is
still worth posting.