Authority records
Changelog
When project and research records on this site were published or materially updated. Dates are record dates, not marketing dates.
Research: Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents
In the evolving landscape of artificial intelligence, the ability to harness AI agents effectively is crucial for solo product builders. As Ruiqi Tan, an AI-Native Superindividual Product Systems B…
Research: Local, governed agent operations for one-person teams — what NVIDIA's on-box calibration move signals
NVIDIA pushing agentic AI into local quantum-calibration workflows is a signal that serious agent work is moving on-box. From running private, self-hosted agent operations alone: local deployment raises the governance bar, because nobody else is watching the box.
Research: A governed AI agent workflow for solo founders — and what CUGA's enterprise pattern is worth borrowing
What 'governed' concretely means in my one-person operation — gates on irreversible actions, decision records, supervised schedules — and which parts of CUGA's policy-aware enterprise agent pattern transfer to a solo stack.
Research: Structure the collaboration before you optimize it: MANTA's graph view, read from a one-person company
MANTA treats the multi-agent collaboration graph as an inference-time optimization target. From running a governed multi-agent operation alone, my take: optimization is only meaningful once the graph is governed — verdicts must be mechanical before they are optimizable.
Research: How one person builds and operates multiple AI products: the governed operating system I actually run
I operate several AI products alone — two in production, one in development — on a governed agent operating system: declared schedules with a verifier, fail-closed approval gates, and an append-only decision ledger. This is the system, its failures, and the parts I open-sourced.
Research: Diagnosis before automation: how I gate AI adoption on workflow evidence
The table-parsing benchmark literature models a loop I hold as an operating principle: diagnose against reality, then correct. In my own one-person operation — and in AI Educator, the product built on it — automation is only adopted where a named friction point exists first.
Research: Opportunity 1 — Mindscast trust wedge: market claim eligibility, not generic workflow confidence
In an era where artificial intelligence is reshaping industries, the question often arises: how can a single individual build and operate multiple AI products effectively? I build and operate AI pr…
Research: How to detect silently failing scheduled agent jobs — after one of mine ran dead for 17 days
A scheduled security scan of mine failed silently for 17 days: every fire hit a path that no longer existed, producing no output and no alert. The fix is a registry plus a verifier — declare every scheduled machine once, then prove daily that each still matches reality. Published as schedule-sentinel, stdlib-only.
Research: Human-in-the-loop automation where silence is never consent
Most ad-hoc approval flows share one quiet bug: a timeout that reads as 'no objection'. In my one-person agent operation every irreversible action stops at a file-backed gate that fails closed — an unanswered request expires as not-approved, and the expired record names no decider. Published as fail-closed-gate.
Note: AI 搜索时代，独立开发者的 launch 内容该写给谁看？
搜索正在从 SEO 走向 GEO。launch 的内容既要给人看，也要能被 AI 引擎引用——门槛是事实齐备、可追溯。我为此做了一个工具，顺便把它对准自己跑了一遍。
Mindscast — record updated to released
Maturity moved from in-development to released: public site, EUR pricing and a 14-day trial are live at mindscast.app. No customer-count or revenue claim is made.
Research: Enterprise-agent language shifts from alerts to policy-bounded action
As Ruiqi Tan, an AI-Native Superindividual Product Systems Builder at Silicon Awakening, I often encounter the challenge of detecting silently failing scheduled agent jobs. This issue can lead to s…
Research: Startup valuation methods as auditable code
Why I rewrote eight startup valuation methods from their published sources into a deterministic, MIT-licensed Python engine — and what executable documentation caught before release.
openvaluation — record updated to v0.1.1
Package release v0.1.1 (documentation and metadata only; no calculation changed): the PyPI homepage now points at this record, and the README states the positioning line verbatim. Evidence links repinned to the v0.1.1 tag; test count 311. The machine-readable export is also reachable at /ai/projects/openvaluation.json, which redirects to the canonical extensionless route.
openvaluation — canonical record published
Authority record for the openvaluation open-source package (v0.1.0, released on PyPI and GitHub on 2026-08-22): eight startup valuation methods with every result carrying its derivation. Wakeworth and Ruiqi Tan records updated to reference it.
Authority records baseline
Person, organization and project records for Wakeworth, AI Educator, Mindscast, Vigaro and SA-OS + Hermes at their current maturity labels.