Software that does nothing
The feature I defend most often in product meetings is the one that does nothing.
Every tool built for founders wants to act on their behalf. Send the sequence. Post the reply. Book the meeting. It demos beautifully, because a demo is a room full of people watching software do things. Then an account gets banned for automated posting, the tool gets blamed, and the founder's name is the one on the thread that everyone screenshots.
I built the opposite, deliberately. This is the reasoning, written down so I stop having the same argument with myself.
What it does
You paste a product URL. It reads the page the way an analyst would: what the product is, who buys it, where it sits against competitors, and how well the page can be read by search engines and by the AI assistants that increasingly answer buyers' first questions.
Then it goes looking for the people already asking. Hacker News, GitHub issues and discussions, Stack Overflow, YouTube comments, Bluesky, Dev.to, and the wider web through search. It reads those threads, decides which ones are a buyer, an evaluator, a builder or noise, and drafts a reply in your voice for the ones that matter.
Then it hands you the draft and stops.
You post it. From your account. Under your name. With your judgement between the draft and the send button.
The five refusals
It will not post or reply on your behalf
This is the one that costs the most in meetings, because it is the one everybody asks for. The answer is always the same. The thread is public. The name on the reply is yours. The account that gets banned is yours. The reputation that follows you into your next company is yours. Software that puts your name on words you did not read is not saving you time. It is spending your name.
There is a quieter reason. The moment a tool posts for you, it has to lie a little to work: about who wrote the reply, about how fast a human could have found the thread. Every platform's rules are written against exactly that lie. A tool that never posts never has to tell it.
It will not paste a draft into forums that ban generated text
Hacker News asks that you write your own words, and its moderators say so plainly. Stack Overflow removes generated answers. On those sites the tool hands over talking points instead of a paste: here is the thread, here is why it matters, here are the three things you know that the asker does not. You write the sentence.
This is not a limitation I apologise for. Those communities are valuable precisely because they enforce that rule. A tool that helps you break it is helping you burn the one place your buyers still talk honestly.
It will not show a rank without evidence
Early on, my own dashboard showed "market position: 3rd of 5". Underneath, in small print: default rank, no evidence. I found it walking through the product as a new user, and I was embarrassed. A disclaimer is not honesty. It is a number in a costume.
Now every score, every draft, every recommendation carries a written reason, and a value with no reason does not render. Where the software cannot know, it says "not ranked yet" and tells you what would produce a rank. People forgive a wrong number with a visible reason. They do not forgive a confident number with none.
It will not scrape a platform that offers no official route
"Find every conversation about my product" sounds like one feature. It is a dozen legal questions, one per platform. Some offer an API for public content and publish the terms. Some offer one for commercial use only with written approval. Some offer nothing and forbid bots in their robots file.
The rule is simple. Official routes only. Where there is no official route, the platform is skipped, and the product says so on the pricing page rather than pretending. This loses some coverage. It keeps the product something a buyer's compliance team can approve, which is the only kind of coverage that matters to a company with customers.
It will not send more than the caps allow
Where the tool does send anything, the emails that tell you what it found, there are caps, and the caps are constraints, not settings. Deliverability is a shared resource. A tool that lets one user burn it burns it for everyone.
Why refusing is the product
Each refusal cost me an argument with myself, and most of them cost a feature request from someone who would have paid for the opposite. I kept them for one reason. The value of the tool is not the clicks it saves. It is the judgement it preserves.
A founder who posts a reply they read, in a thread the tool found, under their own name, has done something no automation can do: shown up as a person where a buyer was asking. The tool made that possible by finding the thread yesterday instead of three weeks from now. It did not make it cheap by removing the person from it.
There is a boundary idea underneath this, the same one that runs through everything I have written this month. The line between the tool and the world is a boundary. What crosses it is your name. Verify at the line. Do not let software cross it for you.
The honest cost
Doing nothing is slower. A founder using this tool posts fewer replies than a founder using something that blasts. The bet is that they post better ones, on platforms that still tolerate them, and that the outcome ledger inside the tool, which records what converted, proves it over a quarter rather than over a demo.
If the bet is wrong, the ledger will say so, in numbers with reasons attached. That is the standard I hold the product to, and the standard I would rather be held to than believed.
This is sheet 038, the last essay in Series 01 of a drawing set on software, security and AI. Sheet 031 was the short version. Sheets 032 and 034 are two of the refusals told as their own stories. The tool is called Trackshon, and it launched this week.