Week of 12 October 2026: the launch week, a name, and 94 commits

94 commits, 3 signups, 38 invitations sent, 20 accepted, and 0 replies. The replies are zero because an invitation sent on Thursday is accepted on Saturday and messaged on Monday; that is the ramp doing what the ramp does, and the next week's post has the number. This is the week the product got a name, a domain under fifteen dollars, three friends' LinkedIn accounts to run on, and most of the surface a customer sees today.

Monday and Tuesday: the name, and the people already talking

The money-back window replaced a free trial: 14 days, refunded in full on request, because a connected LinkedIn account costs real money at the provider from the first day and the ICP is already the free part. Today got the one number that matters, replies this week, at the top. A false claim on the comparison pages, that nobody else publishes limits, was removed, because one competitor does, in a blog post, and a claim that is checkable and wrong is worse than no claim. And twenty-eight days of social posts were written, three a day, with screenshots, most of which I did not post.

Tuesday the product learned to find the people already talking about the problem: a watch list of profiles whose posts are read daily, and everyone who commented under those posts arrives in the queue with the comment as the reason they are there and the post as the thing the message may open on. That is still the warmest signal the product has after a person's own posts. The same day: a smoke test that proves every LinkedIn read against the live provider, a rule that the daily run stands down while the approval queue is already full, the model switched to run natively with an effort setting, and the name. Hellofew. The reasoning is in the repository; the short version is that it says what the product does to a stranger in one word.

Wednesday: sixty-four commits

I had help that day, several branches worked in parallel and merged in the evening, which is why one Wednesday has more commits than most weeks. I am not going to list sixty-four things. These are the ones a customer can see, grouped.

• Today was redesigned around three questions: what happened since you last looked, what is about to happen, what do you need to do. Every action is acknowledged on the screen it happened on, a reply reads as the thread it belongs to, the LinkedIn page reads as a status with the ramp in view, and the first queue appears without a reload.
• The numbers page: what our own accounts did this week, published live, and a ledger of what the product refused to do, across every customer. The page no competitor can publish without admitting their numbers.
• The inbox: every reply gets a label for what the person meant, an email to the owner, and a signed webhook. You can answer from the inbox, with a drafted reply on request, and the draft is never sent without you.
• A public REST API behind per-workspace keys, an MCP server exposing the queue, the replies, the ICP and the stats as tools, an integrations page to manage both, and a developers page that explains them.
• Lists: your own never-contact rows honoured by every check beside the global list, pasted profile URLs that go through the same read-and-write stages into today's queue, and every lead and reply as a CSV download.
• Three ways to pay for one product: monthly, yearly, and an agency plan with client workspaces that are entitled while the agency pays. Every plan runs the same limits; the agency plan is more workspaces, never more volume.
• Referrals, a worst-message clinic, attribution of where each signup first heard of us, a public label so a friend's account can appear on the numbers page under their first name, dogfood presets so a friend's ICP is one click, and analytics named in the privacy policy.
• Five more comparison pages, the citable limits page of what LinkedIn has and has not said, and this blog, as a registry of blocks that read the constants rather than markdown files that would go stale.

The bug that mattered

The first commit of the Wednesday is a fix, and it is the most important one in the series. The inbox had been recording every message in the connected account's LinkedIn inbox as a reply, including from people the product had never written to. The unread badge counted them. The weekly number counted them. And the complaint classifier ran on them, so a stranger saying "not interested" to the customer's personal inbox landed on the global do-not-contact list and paged me. For two weeks. The fix is one early return: a message from somebody who is not a lead of that account is not recorded at all. The test for it is in the suite. The principle it taught, that the customer's inbox is theirs and only conversations we started are ours, is now in the design document and in the help page.

Thursday and Saturday: the site gets one frame

A pricing page with every plan side by side. The home page opened on the outcome, replies, with the week's numbers from our own accounts as live proof under the headline, a weekly numbers card and a badge, a free restriction check, and a skill file so a model can be taught the product. The what-you-get page, in the order it happens. Google sign-in. Three emails between the welcome and the goodbye. And a rule, with the code to enforce it: one header and one width on every page, because the site had grown five headers in a week. The limits those pages publish are on the limits page, and this week's numbers, zero replies and all, are the first row on the numbers page.