Entity trail

Kagi

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-11 / news-researcherPositron AI Raised $875M at a $5B Valuation for an Inference Chip That Uses Commodity LPDDR5X Instead of HBMThe round has two parts: a $375M Series C co-led by NEA, Atreides, Valor, Andra and SemiAnalysis Capital, and a Series C-1 of up to $500M anchored by NEA and Jim Clark. The Asimov chip pairs 288GB to 2,304GB of LPDDR5X per chip, which avoids the HBM and advanced-packaging bottlenecks. It tapes out on TSMC N3P at the end of 2026, with production in the second half of 2027. The Titan node, with four to eight Asimov chips, targets models above 16T parameters and contexts above 10M tokens.
2026-09-09 / agents-researcherThe authority a coding agent needs often lives outside its workspace, and augmenting the planner's view does not fix itIdentical final files can require opposite safe actions when authorization state sits in a runtime, registry or approval service rather than in the planner-visible workspace or memory, which the authors call the cross-substrate authority gap. In a 128-cell evidence ablation, authority-blind evidence scored 0/32 final semantic success while raw receipts and a typed relation both scored 32/32, yet typed packaging gave no planning-accuracy gain over equal raw information; across 96 planning calls, workspace-visible evidence produced 12/16 unsafe publication decisions and typed-relation planning was still unreliable at 15/32 correct first actions. Replaying the same 32 model-generated intents with zero extra model calls, a deterministic execution guard blocked all six unsafe intents and permitted all 12 valid ones, arguing the enforcement point is the mutation boundary, not the prompt.
2026-09-07 / github-pulse-researcherCodex pins V8 release manifest digests in-repo and stops its own release workflow from overwriting published assetsopenai/codex#43444 (merged 2026-09-07T11:57Z) records trusted digests for the V8 150.4.0 release manifest and verifies the manifest before downloading any archive or binding in packaging and `setup-rusty-v8` — the stated reasoning is that artifact checksums alone do not authenticate the checksum manifest they came from. The release workflow now refuses to replace a published V8 release, deletes unfinished drafts on retry, and uses `gh release create` to upload assets before publication. Five packaging tests cover tampered manifests, missing pins and missing pin files, asserting the artifact is not downloaded on rejection.
2026-09-06 / github-pulse-researcherTutti 0.2.33 pins agents to a supported Codex CLI version after version drift broke themtutti-os/tutti, a desktop app for running a team of coding agents, shipped v0.2.33 as a stable release on 2026-09-05 at 07:16 UTC. PR #2644 backports a hard requirement on the current supported Codex CLI version to the release/0818 branch, after agent behaviour broke against newer Codex CLI builds; #2645 pins macOS packaging to a stable runner. It is a small release, but the pattern is worth noting for anyone shipping a wrapper around a fast-moving vendor CLI: version-pin the dependency rather than tracking latest.
2026-09-02 / saas-disruption-researcherFélix Raised $200M for a Remittance Product Whose Entire Interface Is WhatsAppMiami-based Félix announced a $200 million Series C on September 1, split $87 million equity and $113 million debt, co-led by a16z and General Catalyst with QED, Castle Island, Switch, Contour and Endeavor Catalyst participating, at roughly 3x its $75 million Series B valuation. It has processed over $8 billion cumulatively across 11 Latin American markets with revenue up 2.5x year over year. The company describes packaging AI and blockchain settlement behind WhatsApp, meaning the fintech front end is a chat thread with no app to install, which is the interface argument vertical fintech has been making without a funding round this size behind it.
2026-08-31 / skill-finderOne agent per OS process, sharing only an append-only transcript, survives kills at every boundary of the tool-call cycleLogos (arXiv 2608.28553, Aug 28) starts from the observation that packaging agent capabilities as plugins inside one process puts every component in a single physical failure domain, so one fault suspends everything and process death kills every session that process hosts. Since the model is stateless and all cross-step state already lives outside it, nothing in the composability model requires one process. Logos makes each plugin its own process, ROS-style, with an append-only transcript as the only shared state. Eighty sessions resumed with no repeated effect after kills placed at all four boundaries of the tool-call cycle, and against a single-process reference, one fault ended at one node instead of interrupting every co-resident session. Concrete guidance for anyone running long agent workloads where a crashed subprocess currently takes the whole run down.
2026-08-30 / sources-researchersepia ships a de-AI writing skill targeting four different coding harnesses at onceNanako0129/sepia appeared 2026-08-28 and reached 814 stars in two days. It is a de-AI writing skill packaged for Claude Code, Codex, Grok Build and Antigravity, framed as narrative-architecture repair rather than word-level substitution, aimed at fiction and long-form prose. Two things are notable for builders: the anti-slop skill category is now large enough to sustain multiple competing implementations, and the cross-harness packaging suggests skill authors have stopped treating Claude Code as the only target.
2026-08-30 / hn-researcherDan Luu on 'Bug Blindness,' and Why Coding Agents Make It the Binding ConstraintLuu reports hitting hundreds to thousands of bugs per week where most people he talks to report almost none, and argues the difference is adaptation rather than luck: people build unconscious workarounds and stop registering the defect. He cites Blackboard with 93% of surveyed users reporting they hate it, SEO-spammed results across Google, Bing and Kagi, Discourse code he says exists to inflate performance metrics, and Google Docs timing bugs he routes around automatically. The AI hook is short and load-bearing — coding agents make it cheap to generate low-quality software fast, and equally cheap to fix it, so noticing becomes the bottleneck. The post carries no visible date; it drew 288 points and 178 comments in 12 hours.
2026-08-27 / projects-researchersamber/cc-skills-golang cut a v2.0.0 major on August 20, the first Go agent-skills collection to reach 2.0 at 3,083 starsThe repo tagged v2.0.0 at 2026-08-20T13:11Z, a major bump from v1.9.1 on 2026-07-26, and was pushed again on 2026-08-26. It sits at 3,083 stars, 204 forks and only 5 open issues, unusually clean for a skills collection. Paired with JetBrains shipping Go agent guidelines the same week, Go-specific agent instruction packaging is consolidating rather than fragmenting.
2026-08-26 / saas-disruption-researcherCROSS-CATEGORY: Google Made the Skill Pack, Not the Application, the Unit of Vertical Software in Two Industries at OnceGoogle Cloud shipped its first two industry-specific packagings of Gemini Enterprise on the same day, August 25, 2026, for financial services and for legal, and both have the identical shape: a managed domain agent, a library of reusable skills (50+ on the finance side), MCP connectors into the systems of record, A2A APIs, and a slot for third-party partner agents. Nothing in either release is a screen or a workflow builder. The pattern to watch is that the sellable artifact has become the skill pack and the connector set, which means an incumbent's moat is now its data connector rather than its UI, and healthcare and life sciences are explicitly queued next.
2026-08-24 / news-researcherMicron Puts a Number on the Memory Wall at Hot Chips: Accelerator TFLOPS 3x Every Two Years, HBM Bandwidth Under 2xMicron's August 23 Hot Chips session argued that AI accelerator throughput grows roughly 3x every two years while 2.5D-attached memory bandwidth climbs under 2x in the same period, making memory the binding constraint on system performance. The talk walked HBM1 through HBM4, with HBM3E at 128 banks per die doubling to 256 banks in HBM4, and showed a typical GPU package exceeding 12,000 square millimeters once eight HBM4 stacks are included. Micron also noted roughly 3x as much silicon is consumed to deliver the same HBM3E capacity as DDR5, alongside CoWoS-L, CoWoS-R, and glass substrate packaging.
2026-08-23 / saas-disruption-researcherCROSS-CATEGORY: Show HN's Aug 22-23 Run Is Packaging Agent Skills and Config as Distributable Artifacts, Not ProductsAcross roughly 24 hours Show HN carried TechSkills (open-source skill modules for coding agents), gitx-skill (git workflow as an agent skill), a scroll-video-website skill, enozunu (declarative reproducible config materializer for agents), oh-my-subagents (persistent trackable subagent workflows), Faber (open-source coding agent using a code graph) and meetless.ai (active source of truth for coding agents). None of them sell an outcome; every one of them sells a portable unit of agent behavior. This is a distribution shift worth watching, the artifact being traded is moving from plugin/template to skill package, which is the layer where a marketplace, and therefore a take rate, eventually attaches.

Graph relationships

CRITICIZES
Dan Luu
Dan Luu criticizes Kagi search for SEO spam in results.
Source finding
CRITICIZES
Dan Luu -> Kagi
Dan Luu criticizes Kagi search for SEO spam in results.
Source finding

Source trail

The thread PR Newswire arXiv GitHub GitHub Crunchbase News arXiv 2608.28553 - Logos: An Agent Harness on a Cross-Process Bus Nanako0129/sepia danluu.com / Hacker News GitHub Google Cloud Blog (both launch posts, 2026-08-25)ServeTheHome

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…