Projects
What's actually been built.
No hype, no vague "coming soon." Here is the real status of every project, in the order it currently matters. Anything parked says so, and anything unreleased says that too.
Huntloop
An autonomous outbound CRM for founder-led B2B teams that need pipeline but cannot justify an SDR. It turns a product brief into one measured loop (source, verify, score, enroll, send, handle replies, learn) while the founder keeps control of the offer, the audience, the sender, and how much autonomy the system gets.
- Specification approved 6 September 2026. Release hardening is the only work authorized so far: dependency remediation, SSRF hardening, and independent security review before anything else proceeds.
- First customer profile is locked: founder-led B2B software and technical-service teams, two to twenty people, no dedicated sales-development team.
- The validation path is deliberately small. Run it on our own outbound first, then five design partners, then ten only if the first five actually use it and want to continue.
- What it is not, yet: not launched, no customers, no published pricing. The first hundred real sends run with a human approving each one.
Huntloop is a separate product with its own brand. It is named here because it is where the work is going, not because this page is selling it.
The Org
The AI-native company behind The Nest. Fourteen agent seats cover the departments a one-person company still needs (engineering, security, product, revenue, marketing, legal and risk, research), each with a defined job, a review chain, and work that moves as cards on a board. Every project on this site was built through it, including this page.
The structure is not stable and is not presented as a finished system. It was rebuilt in September 2026, and a dispatch broker and a review seat were both retired for being redundant. What replaced them is simpler: a card, a coding seat, a test that has to pass, and a written decision.
The Null Result
A quantitative trading research system, built and run twice: once as historical backtests, once as a live shadow loop trading real market data in real time, paper-only, never a real dollar. The verdict from both layers: no large edge survives costs. One strategy cleared statistical significance for the first time ever, downward, at z = −6.27. A strategy proven to work, in the wrong direction.
The search did not stop at the null result. It changed shape. The current phase tests a much narrower question against licensed futures data, with every hypothesis written down and frozen before the data is bought, so a result cannot be quietly reinterpreted after the fact. The most recent independent review returned a hold.
INeedToKnowThis
Paste a saved reel or short, get a study doc you'll actually read: one thing worth keeping, up to five lessons with the exact quote each one came from, and questions you grade yourself on from memory. Every lesson and example must carry a verbatim quote from the source, checked before you ever see it. Anything that fails is deleted, and the doc says how many and why.
- Working end to end: transcript in, verified study doc out, flashcard export, a real usage meter. The test suite runs on every change.
- Not yet live at a public address. This section will carry a real link the day it is.
Three iOS games
Three native SwiftUI games, all built on 3 September 2026, in a single day, by directing an AI coding agent. They are on this page because the honest version of that story is more useful than the impressive version.
- Verdant Idle is a cozy idle farming RPG and the furthest along. Its economy lives in a separate, platform-neutral Swift package with a verification command, and GitHub Actions runs both that check and a full iOS build on every push. StoreKit purchases, a privacy manifest, and an original icon are in place.
- RiftRally is a push-your-luck dice game with Game Center matches for two to four players. Its one genuinely interesting piece of engineering: every player commits hidden randomness before anyone reveals it, so no client gets to choose the roll stream.
- TickerTown is a trading tycoon campaign played against a fictional, deterministic market. No brokerage or market-data connection, by design.
What is honest about this: the feature lists these projects generate are very long, because an AI writes a lot of code quickly. Two of the three have exactly one test file. None has been submitted to the App Store, and none has ever been played by anyone who did not build it. Verdant Idle is the only one whose build is mechanically verified, and even it still needs bundle identifiers, purchase records, and privacy URLs before it could reach TestFlight.
That gap, between what a feature list claims and what has actually been proven, is the reason these are worth writing about at all.
The AI Website Hustle
Full, multi-page websites for local businesses that don't have one, plumbers and HVAC first, built fast with AI and pitched already-live rather than as a mockup. The build recipe works and produced a real site. The selling motion never started.
- You are looking at client zero right now. This site is the first and only build under that recipe.
- Parked as of September 2026, with zero clients approached and zero revenue. The outbound engine that was being built to feed it became Huntloop instead, which is a larger opportunity aimed at a customer we understand better.
- Not deleted. The recipe and the templates still exist, and this is written down rather than quietly dropped, because a parked project is a result too.