A year in public, what we'd do again
Three decisions that compounded over a year of building PursuitAgent in public, one we'd reverse, and one we're still uncertain about. Honest operator notes on what worked and what didn't.
A year of building PursuitAgent in public. A few hundred blog posts, two State-of-Proposal-Tools research drops, a changelog that updates every week whether anyone is reading it or not, and an embarrassing number of “what we got wrong” posts that still embarrass me when I reread them.
Here is what I would do again, what I would reverse, and what I am still uncertain about.
Three decisions that compounded
Publishing the hallucination number. Twelve months ago we decided to put a hallucination rate on the grounded-AI page and update it weekly with the method attached. Every sales conversation with a regulated buyer in the year since has touched this number. Customers bring it up in their own internal reviews. Compliance officers at prospects have cited it in RFP responses for their customers. It is the single claim on our marketing site that has done the most unpaid distribution.
The decision compounded not because the number is low (it is, and that helps) but because the method is published. Anyone can reproduce it. Anyone can argue with it. The credibility comes from the willingness to be argued with.
Pricing in public. We published our pricing from week one and wrote about the decision at day 90. A year later the pricing is still public, still simple, still on one page. We have seen more than one prospect tell us the public pricing was the reason they shortlisted us — not because the price was cheapest, but because the transparency saved them three sales cycles of price-discovery games. The incumbents keep pricing private. The compounding for us has been that prospects arrive pre-qualified on budget.
Publishing our own proposal process. We wrote about how we respond to our own RFPs in month six. The decision looked minor at the time — the post was meant as a demo of the product. Over the year it turned into one of our most-linked posts, one of the posts sales engineers on prospect teams would read before their first call, and the thing that most clearly signals we operate the product we build. Dogfooding as a marketing strategy sounds strange until you realize that in a category full of vendors who don’t, dogfooding publicly is a differentiator.
One I would reverse
The first version of the changelog. We launched it as a quarterly digest. It was lovely, pretty, illustrated, and completely useless for the people who actually wanted to know what shipped. Three months later we replaced it with a running list, dated, one entry per ship, linked to a commit where possible. The engagement on the running version is roughly 10x the engagement on the digest version was, and customers use it as a reference — “what was the release where you fixed X?” is an answerable question now.
The reversal cost us three months of content design work that was pointed in the wrong direction. If I were starting again, I would launch the running version on day one and never pretend the digest was worth the effort.
One I am still uncertain about
How much to write about competitors. Our /compare pages are sourced, honest, and clear about where each competitor wins. The blog is mostly positive — we write about craft, about engineering, about the category. Competitor teardowns happen, but they are a minority of content and they always cite.
What I am uncertain about is whether this is the right mix. The most-read posts on the blog are the teardowns — the Loopio ten-year retrospective, the Responsive reviews synthesis, the Qvidian five-year review sentiment piece. Readers want the teardowns. Writing more of them would grow the audience faster. I am not sure writing more of them would grow the right audience, and I am not sure the brand we would build by writing more of them would be the brand I want to operate in for the next 10 years.
I have not reversed the mix, but I have kept the question open. The honest version is: the mix we picked is defensible; a different mix might have compounded faster with a different set of trade-offs. I will reread this post in another year and see whether the question has resolved.
The pattern across all three
The three decisions that compounded share one trait. Each one made it easier for a reader to trust us later. The hallucination number. The public pricing. The operating-in-public posts about our own process. None of the three were marketing bets in the traditional sense. They were reductions in asymmetric information between us and the reader. A year in, every sales conversation that closes quickly is a conversation where the reader has already verified most of what they needed to verify, on their own, on the blog.
The one I would reverse had the opposite trait. The quarterly digest was an information bottleneck — it aggregated updates so that readers would engage with one well-crafted artifact instead of 40 small ones. The logic was backward. Readers who care about a product want the 40 small things, because the 40 small things let them track velocity. One big thing every quarter hides velocity.
The one I am uncertain about is a harder case, because the competitor content would reduce information asymmetry, and I think that is the right end state. The uncertainty is whether reaching it by writing more teardowns is the path that leaves us in the right brand space, or whether the same end state is better reached by other means.
What year two looks like
More of the three that worked. Less of the one that didn’t. And a deliberate, slower attempt to resolve the uncertain one — probably by writing two or three more rigorous teardowns on specific architecture claims the incumbents make, rather than by volume-writing on competitors as category.
A year in public has been cheaper and more productive than I expected. The marginal post is not much work. The marginal reader who becomes a customer is, on this blog, still a significantly above-average customer in terms of fit. Year two will test whether the compounding continues after the category gets more competitive for attention. My bet is that it does, and that the three decisions that worked early will work harder later.
We will see. I will write the post again in April 2027 and tell you.