One post / Trust boundary, first protocol, and what remains unproven
Nereid Build Thread
An evidence-led record of building a delivery-control plane for coding agents.
Open threadBuild threads, product notes, and technical writing about trying to make useful software instead of just interesting prototypes.
Latest launch note
A background-removal tool becomes a much more serious product once you treat public use, abuse handling, moderation, and operator accountability as first-class requirements.
Featured build threads
These series are where I actually work through product shape, architecture, trade-offs, and commercial questions — not a highlight reel, the real back-and-forth.
One post / Trust boundary, first protocol, and what remains unproven
An evidence-led record of building a delivery-control plane for coding agents.
Open threadThree posts / Positioning, commercial path, and hosted-product reality
A build thread on turning CampaignForge AI from a rough prototype into a smaller product that could actually be sold.
CampaignForge AI takes one campaign brief and turns it into a stronger starting point across strategy, messaging, and creative direction.
The best next step for CampaignForge AI is not a big platform play. It is to pick the smallest shape that makes sense and see if people want it.
Four posts / Product strategy, architecture, roadmap, and release readiness
A build thread on building a city-first travel product across iOS, backend, and a future web planning layer.
CityScout is a city-first travel companion built to help people feel oriented before a trip starts, without forcing AI into every corner of the product.
CityScout keeps the iOS app light, the backend authoritative, and OpenAI behind a server boundary so the product stays easier to secure and evolve.
Seven posts / Product, design, architecture, and process
A running record of how Hoxa is taking shape, from the product idea through the training system and architecture.
Hoxa starts from a simple observation: a lot of fitness products ask for confidence before they do enough to help someone build it.
The visual and interaction design of Hoxa has to lower friction without flattening ambition. Clear priorities matter more than sanding the product down until it feels generic.
One post / Product strategy and agent architecture
A build thread on building CreatorOS, from the product thesis through the agent system and evaluation approach.
Open threadThree posts / Launch, evaluation strategy, and engine architecture
A build thread on turning EvalKit into a real product, from the evaluation argument through the FastAPI engine and SaaS plumbing.
A launch note for EvalKit covering the problem, the product model, the current build, and where it goes next.
A first-principles case for regression testing LLM features, focused on the failure modes spot checks miss and the workflow that keeps them visible.
All writing
How Cutout Studio moved from a simple background-removal tool to a safer public beta with verified login, moderation, audit logging, and no image retention by default.
How a small internal review habit turned into a reusable Codex workflow for updates, follow-on actions, and all the admin that usually gets missed.
A small editorial product for turning a current London weekend into a useful shortlist, with the publishing and newsletter loop kept deliberately simple.
How smaller, clearer evaluation loops can create more product velocity than broad benchmark suites.
Designing tools for research and analysis means optimising for orientation, not just speed.
How staged execution, fallback rules, and operator visibility make agent workflows much easier to trust.