Portfolio Log #004
This week was not especially disciplined but also awfully packed with things eating my time that had nothing to do with computers.
I had a clear idea of things to do, then spent large parts avoiding them or not having time to do them. Leadrack found a sharper way to look for customers, and I got nerd-sniped into starting another product. More on that later.
So, not focused. But productive.
Leadrack can now test competing ICPs
I mentioned this idea last week, but only as a direction: instead of assuming a company already knows its ideal customer, Leadrack could help work out which ICP is actually worth pursuing.
That is now built.
A user describes the business and Leadrack proposes three competing theses, each with a different customer segment, trigger and angle. It then runs them as parallel lead pipelines and tracks what each one produces: useful leads at first, then replies once enough outreach has gone out.
The experiment view puts the results next to each other and produces a “what we learned” report. There is also an optional auto-drive mode that can pause weak theses and give more capacity to the strongest one, while keeping a record of every decision and making it reversible.
It still needs real usage before I know whether the scoring works. But the vague idea from last week is now something I can actually run across the portfolio instead of just thinking about it.
Leadrack starts watching the competition
I also pushed further in the same direction with competitor watch.
A team can now give Leadrack its competitors, or let Leadrack infer a starting list from its website. A new agent then looks across Hacker News, Bluesky and Reddit for people discussing those competitors: complaints, pricing gripes, requests for alternatives, and other moments where someone may genuinely be open to switching.
The same competitor list is available to the rest of the fleet too, so it can influence how other agents search rather than living as an isolated filter.
I also added a search-ads agent. It finds companies advertising against the keywords a team cares about. Someone paying to appear for a problem is often a much stronger signal than someone who merely resembles an ICP on paper.
Both additions move Leadrack away from “find more companies like this” and toward “find evidence that this company is worth contacting now.” That distinction is becoming the product.
A weekend detour into trading bots
I also built a personal trading bot for Hyperliquid.
It discovers traders with enough history to evaluate, mirrors their positions in proportion to account size, and applies hard limits around leverage, exposure and daily loss. It has a local dashboard for positions, trades, performance and managing the traders being followed.
The important word is personal. It runs in paper mode by default, and I have no immediate plan to turn it into another product. I mostly wanted to understand the mechanics properly and build something where conservative defaults and failure handling matter more than the interface.
It was a pretty cool distraction. It was also... a distraction.
A new product appeared
The biggest new thing this week is a new product I started working on: a small marketplace for newsletter sponsorships.
The idea was originally shared by Thomas, who runs the Uneed newsletter and could use something like it.
The basic idea is that neither side really needs another directory. Newsletter writers do not want to chase sponsors, negotiate dates, collect creative and follow up on invoices. Advertisers do not want to search a giant list, contact publishers one by one, and manage a separate booking and payment for every placement.
So the product is meant to act more like a concierge. An advertiser describes the audience they want to reach. I shortlist relevant newsletters by hand. Both sides approve the match, the advertiser pays, the creative is handed over in one shared room, and the newsletter gets paid after the placement runs.
Over two days, that went from a landing page to a surprisingly complete first application: onboarding for both sides, human-reviewed matching, booking states, Stripe Checkout and payouts, creative formats and approvals, a shared conversation, tracked links, and proof that the issue ran.
That is far too much product to build before validating the marketplace, which I am aware of.
The next step is therefore not another feature. It is finding the first few newsletters and advertisers, making the matches manually, and seeing whether either side actually wants this handled for them.
I'm sitting on first lists of 100 newsletters and sponsors respectively, so next week will be about reaching out to them.
I shut down ingestlayer
I also removed ingestlayer from the portfolio this week.
When I started these logs, I wrote that I was worried about it. In my mind, tt was an SEO-heavy product, but it never showed the early search traction I wanted to see, and I did not find another signal strong enough to justify continuing to invest in it.
I added 120+ programmatic SEO pages over the last months, but they barely got picked up by Google or geenrated any traffic (especially compared to some other products with a similar timeframe).
Keeping a product around still takes attention, maintenance and a little bit of guilt every time I choose to work on something else. At some point, “giving it more time” just becomes avoiding the decision.
So I shut it down.
And it feels good. Also sort of justifies starting another product in the same week. Keep trying things, but do not keep all of them forever.
The cost of all this motion
Contentcron, Grepture and Modsignal were mostly quiet this week.
Next week should be much less about starting and much more about putting things in front of people.