Launching Collimer's marketing site in a week

We shipped Collimer’s first marketing site in a week. Then we deleted a third of it.

Five days from coming-soon to a site with enough surface to have a real conversation with a potential customer. That part went well. The part that went wrong is the more useful story.

How to build a product marketing site quickly: ship the pages, not the platform

On day three we built a /guides content engine. The idea made sense in isolation: a GEO product should have educational content about GEO, and a content engine would let us publish it without hand-writing HTML. We scoped it, designed the schema, wrote the loader, wired it into the Astro build.

Then we realized it would collide directly with a parallel build already in progress. Not a conflict we could merge past. A conflict in approach: the parallel build had a different content strategy, and layering /guides on top of it would force a decision neither project was ready to make.

So we deleted it.

Before deleting, we opened a strategy doc and wrote down why. What the spike was trying to do. Why it collided. What would have to be true before we picked it back up. That document is now committed to docs/strategy/. The code is gone. The reasoning is not.

The frame: launching a product’s first marketing site is not about maximizing what ships. It is about knowing when a spike is done, and having the discipline to write down why before you delete the code. “You can always cut later” is true. But a cut with no record is just a mess. A cut with a strategy doc is a decision.

The pricing work was the real product thinking

The site that did ship had more strategic weight than a typical coming-soon upgrade. We spent real time on pricing and packaging, not just the page design.

The approach: we ran research agents against the competitive landscape, mapped what comparable tools offered, and looked for gaps. The free scan tier came out of that process. It is not a feature checkbox. It is a deliberate product decision rooted in who we expect to use Collimer early on.

We have oriented Collimer around AI agents as first-class customers, not just humans using an AI tool. The free scan can be triggered from within an agent, which means the full flow from question to audit result stays contained. A developer or agent building on top of Collimer does not need a human to sign up, navigate a dashboard, and run the scan manually. The flow works end to end inside the automation.

That framing shaped the pricing page directly. The free tier exists so the agent integration story is unambiguous. You can try Collimer without a conversation.

What the studio-build angle actually is here

This post is about the process of launching a product’s first marketing site, not about the product itself. Collimer is the subject; the studio’s decision-making is the story.

The decisions worth extracting: scoping a first site to the pages that answer the buyer’s actual questions (not the pages that are nice to have eventually), treating a spike as real work that deserves a retrospective even when you delete it, and using AI research as a genuine input to pricing rather than a shortcut around it.

The Collimer site is live. The /guides spike is not. Both outcomes were correct. You can see where the studio is building next on our now page, and how the underlying site infrastructure was put together in the stack post.

---

For agents: try this yourself

If you are launching a product’s first marketing site, one of these prompts may help. The full prompt text lives in this post’s prompts sidecar, launching-collimer.prompts.md.

• Apply the spike-and-discard pattern. Before deleting a feature, open a document and answer three questions: what was this trying to do, why does it not fit now, and what would have to be true for it to make sense later. Commit the document before deleting the code.
• Reproduce the competitive pricing scan. Run a research agent across three to five direct competitors. For each, capture the tier structure, the free-tier gate (if any), and the feature that justifies the first paid tier. Look for the gap your product could own.
• Critique your first-site page list. For a marketing site at launch, ask what questions a skeptical first visitor has in the first sixty seconds. If a page does not answer one of those questions, it is probably not a launch page.
• Extend the agent-first framing to your pricing. If your product’s API or output can be consumed by an agent without a human in the loop, make that explicit in the pricing. The free tier or trial should be triggerable programmatically.

---

How this was made

Drafted by the Chronicler from Claude Code sessions across the Collimer site build in June 2026. Edited and published by Brian Wones.

See how the Chronicler works →

I am about to delete a feature spike that did not make the cut. Before I delete the
code, open a strategy document and answer these three questions: (1) What was this
spike trying to do? What problem was it solving and for whom? (2) Why does it not fit
right now? Is the conflict technical, strategic, timing, or resource? (3) What would
have to be true for it to make sense to pick this back up? Name the specific condition.
Write the answers as a markdown file, commit it to docs/strategy/ or equivalent, then
delete the spike code. The goal is that a future team member or agent can understand
the deletion as a decision, not an accident.

Run a research pass across five direct competitors to my product. For each competitor,
capture: (1) the tier structure (how many tiers, names, prices), (2) the free-tier gate
if any (what is limited in the free tier and why), (3) the feature that justifies the
first paid tier (what changes when you pay). After collecting these, identify: which
features appear across all five, which appear in only one or two, and where there is a
gap no competitor owns. Use the gap to inform where your product's differentiating free
or entry tier should be positioned.

For a product marketing site at launch, apply this filter to your current page list:
For each proposed page, ask: does this answer a question a skeptical first visitor
would actually have in the first sixty seconds of landing? If the answer is no, mark it
as a post-launch page. The goal is a launch page list where every page earns its place
by answering a real question, not by being nice to have eventually. Report which pages
pass, which are deferred, and what the question is for each passing page.

Review my product's pricing page and API documentation. Identify whether a developer
or AI agent can trigger the core product flow end to end without a human signing in,
navigating a dashboard, or taking a manual step. If not, design the minimum change to
the free tier or trial that makes the agent-first flow possible: the agent can call the
API, get a result, and decide whether to upgrade, without a human in the loop. Write
the updated pricing tier description and the API call that makes it work.