openwiki-wiki/themes.md

# Themes

Recurring topics, signals, and trends observed across Robin's sources. Detailed evidence and item lists live on the relevant `/sources/<connector>.md` or topic page; this file is the index.

| Topic key | Theme / Signal | First seen | Last seen | Confidence | Sources | Evidence count | Status | Evidence |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `open-weight-ai` | Open-weights / open-source model releases and policy debate (Inkling, Grok Build OSS, "invest in open source AI" PDF) | 2026-07-15 | 2026-07-16 | source-backed | hackernews | 4+ | active | [HN: 48924912, 48926590, 48927095, 48931225](/sources/hackernews.md#open-weight-ai-source-backed) |
| `computer-use-agents` | API/startup activity around agents that drive UIs without APIs (Coasty YC launch, AI-agent databases, GPT-Red) | 2026-07-15 | 2026-07-16 | source-backed | hackernews | 3 | active | [HN: 48922706, 48931392, 48931204](/sources/hackernews.md#computer-use-agents-source-backed) |
| `codex-openai` | OpenAI Codex-related product/safety coverage (Codex Micro, follow-up video, GPT-Red) | 2026-07-15 | 2026-07-16 | source-backed | hackernews | 3 | active | [HN: 48923079, 48931078, 48931204](/sources/hackernews.md#codex-openai-source-backed) |
| `ai-safety-debate` | Public debate on AI alignment, open-weight risk, and "AI is just a tool" framing (DeepMind departure essay, alignment-faking probes, Torvalds, Haskell defector, Karen Hao, LLM-escape fiction) | 2026-07-15 | 2026-07-26 | source-backed | hackernews | 8 | active | [HN: 48925271, 48931234, 48930971, 48930929, 48930363, 48931225, 49059995, 49059660](/sources/hackernews.md#ai-safety-debate-source-backed--refreshed) |
| `llm-on-old-hardware` | Small / tiny LLMs running on cheap or legacy hardware (Gemma 4 26B on 13-yr Xeon, 28.9M-param LLM on $8 ESP32, Inflect-Micro-v2 9.36M voice model) — promoted from watchlist after recurrence | 2026-07-15 | 2026-07-26 | source-backed | hackernews | 3 | active | [HN: 48922434, 49050512, 49053375](/sources/hackernews.md#llm-on-old-hardware-source-backed--promoted) |
| `context-engineering-claude5` | Anthropic's 427-pt "new rules of context engineering for Claude 5" post plus Claude-Code-context-deletion, Opus-5-hardcoded-no-subagents, prompt-cutting experiments, and an Opus 5 status incident — dominant AI-Agent-workflow signal this window | 2026-07-26 | 2026-07-26 | source-backed | hackernews | 5 | active | [HN: 49051361, 49056194, 49056022, 49056689, 49055752](/sources/hackernews.md#context-engineering-claude5-source-backed) |
| `class-organoid-3d-bioprinting` | Recurring organoid & 3D bioprinting coverage from MIT, NPR, MIT Tech Review, Nature, MSU, McGill — academic / bio-Mech-E stream, not vendor launches | 2026-07-16 | 2026-07-16 | source-backed | hackernews | 12+ (across 4 queries) | active | [HN: organoid / organoids / bioprinting / 3D bioprinter clusters](/sources/hackernews.md#class-organoid-3d-bioprinting-source-backed) |
| `MiniMax-m3` | MiniMax-M3 (1M context multimodal) HuggingFace model card + Twitter launch; cleanest signal for the bare `MiniMax` query which is otherwise substring noise | 2026-07-16 | 2026-07-16 | watchlist | hackernews | 2 | watchlist | [HN: M3 model cluster (huggingface + twitter)](/sources/hackernews.md#MiniMax-m3-watchlist) |
| `openwiki-brand` | `langchain-ai/openwiki` (96 pts) — same product family as Robin's OpenWiki; high-engagement on HN | 2026-07-16 | 2026-07-16 | source-backed | hackernews | 1 (+ 2 adjacent wrappers) | watchlist | [HN: openwiki cluster](/sources/hackernews.md#openwiki-brand-signal) |
| `coding-agent-tooling` | Claude-Code / Codex-CLI wrapper and harness ecosystem (Ruflo, Termic, terminai, Argus, Hallmark, data-loss write-ups, blind taste tests) — derivative tooling but dense recurrence | 2026-07-26 | 2026-07-26 | source-backed | hackernews | 11 | active | [HN: 49051072, 49050011, 49055227, 49056316, 49054070, 49058547, 49058441, 49058217, 49049978, 49053523, 49050042](/sources/hackernews.md#coding-agent-tooling-source-backed) |
| `local-LLM-stream` | Adjacent to "本地大模型部署" focus — Debian LLM policy vote, Cloudflare AI traffic options, DeepSeek-fundraise leaked transcript, Inflect-Micro-v2 model card | 2026-07-26 | 2026-07-26 | watchlist | hackernews | 4 | watchlist | [HN: 49050859, 49052564, 49052912, 49053375](/sources/hackernews.md#local-llm-stream-watchlist--refreshed) |
| `agent-memory-architecture` | Show HN items on structured memory / trust layers for agents (Worklog SQLite, Open Knowledge format v0.2, Assumption Ledger) — directly adjacent to OpenWiki's role | 2026-07-26 | 2026-07-26 | watchlist | hackernews | 3 | watchlist | [HN: 49059959, 49059889, 49060215](/sources/hackernews.md#agent-memory-architecture-watchlist--new) |

## Notes

- Promote a watchlist signal to a full theme row only after it recurs or comes from a high-quality source.
- Keep prose here to 1-2 sentences per theme; link out for detail.
- HN evidence rules: low-engagement items are watchlist by default; promotion requires recurrence, strong engagement, or corroboration.
- Vendor brand signal: CelVivo, 3Brain, Cellink, ClinoStar, and ClinoReactor still have no material HN footprint (see [/sources/hackernews.md](/sources/hackernews.md) for the per-query hit table). Watch the next run — recurrence would change the picture.
- Query-set rotation: the 2026-07-26 run focused on LLM-agent topics and dropped the life-science / organoid / vendor queries, so `class-organoid-3d-bioprinting` and the vendor brands were not refreshed. Next run should re-add those queries alongside the LLM-agent set.