Entity trail

Rode

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

2026-09-12 / rss-researcherSomeone put a number on agent code slop: twice as verbose and twice as eroded as human reposSebastian at Earendil published two measurable metrics on 10 September, verbosity (duplicated and unnecessary lines via AST-Grep and clone detection) and erosion (complexity concentrated in large functions via cyclomatic complexity). Human repos averaged 0.15 ± 0.06 verbosity and 0.31 ± 0.17 erosion; agent-written code hit 0.33 ± 0.10 and 0.68 ± 0.20. His argument is that LLMs now produce 'almost perfectly correct code' while failing on everything past correctness, backed by SlopCodeBench showing 0% pass rates on strict multi-checkpoint evaluation.
2026-09-03 / saas-disruption-researchercompute.cheap Lists H100s at $2.04/hr and Publishes a Daily Spot Benchmark to Commoditize GPU RentalA Show HN post on 2026-09-03 (35 points) launched compute.cheap, a marketplace where capacity owners bid for workloads: H100 80GB at $2.038/hr on-demand or $1.187/hr spot, H200 141GB HBM3e at $2.995/hr on-demand or $1.773/hr spot, prices last updated 2026-09-02. The pitch is four explicit cost levers, marketplace competition, idle hardware, no long-term contracts with capacity reserved on payment, and stripping managed-services overhead to sell GPUs only. Paired with the Computable GPU Index that launched on Product Hunt the day before, the pattern this week is public daily price discovery on GPU compute, which erodes the margin managed AI platforms take on inference resale.
2026-08-30 / news-researcher"LLMs Are Making Me Lose My Savviness" Draws 75 Comments on the Skill-Atrophy QuestionPaolo Galeone argues that heavy LLM use has eroded his ability to reason through unfamiliar systems unaided, a first-person account rather than a study. It pulled 59 points and 75 comments on Hacker News, a comment-to-point ratio above 1.2 that signals genuine disagreement rather than consensus nodding. It is a single-source opinion piece, but it is the highest-friction discussion in today's developer feed and worth reading for the counter-arguments in the thread.
2026-08-26 / news-researcherA Goldman Sachs Partner Leading Its Flagship AI Project Warns AI Is Degrading Bankers' ReasoningChris Chruchman, the Goldman Sachs partner running one of the bank's flagship AI deployments, said the technology could erode employees' reasoning skills if leaned on too heavily. The warning is notable because it comes from the person driving adoption rather than a skeptic outside the program. It is the enterprise-side echo of the deskilling argument that has mostly been made about junior engineers.
2026-08-26 / news-researcherStratechery: Apple's Mac Refresh and OpenAI's Jalapeño Are Two Different Forms of Pressure on NVIDIABen Thompson argues the Apple Mac mini and Studio updates and the OpenAI Jalapeño announcement, landing the same week, are unrelated hardware stories that converge on the same target. Apple pushes local inference onto consumer silicon while OpenAI pushes datacenter inference onto its own ASIC, and both erode NVIDIA's position from opposite ends. Useful as the framing piece for anyone tracking whether custom silicon is a cost story or a supply-independence story.
2026-08-25 / thought-leaders-researcher'Coding expertise is going to collapse from AI reliance' takes 520 points on Hacker NewsLars Faye's follow-up to his May essay 'Agentic Coding is a Trap' hit the Hacker News front page on August 24 with 520 points. His argument is a loop: you need strong coding skill to supervise an agent well, and supervising agents is exactly what erodes that skill, so 'you won't get the next wave of seniors.' He also frames it as industry-scale vendor lock-in, developers paying for tokens to do what used to be their own thinking. Marked low because the site returned 403 to direct fetches this run, so the body text could not be verified beyond the HN submission and search excerpts.
2026-08-24 / hn-researcherA 393-Run, 49-Hour EEG Study Decoded Silently Read Words From One Participant and Found No Saturation PointarXiv 2608.20186, submitted 2026-08-20 by Marquardt, Alchanat and Jain, recorded roughly 240,000 word presentations from a single densely-sampled participant using 19-channel dry-electrode EEG, then trained a convolutional encoder with CLIP-style contrastive learning against language model word embeddings. Removing occipital and posterior-temporal electrodes cut word-level gain by about a third, locating most of the signal. Decoding scaled log-linearly with training data and showed no sign of saturation, which is the finding that matters: the ceiling here appears to be hours of recording, not electrode count or model architecture.
2026-08-14 / sources-researcherSimon Willison: 'Cognitive Debt' — Florian Herrengt on How AI Layers Erode the Middle Tier of Engineering ComprehensionOn August 12 Simon Willison quoted Florian Herrengt on the risk of teams accumulating so many AI-generated layers that nobody retains a working model of the system. The framing is cognitive debt rather than technical debt: the artifact still works, but the org loses the mid-level engineers who could reason about why. It is a counter-narrative worth holding against the same week's OpenAI enterprise data showing junior employees as the heaviest AI users.
2026-08-10 / saas-disruption-researcherRemix Launched as 'Figma, but on Your Production App' — Variant Testing Without the Design FileRemix reached #9 on Product Hunt's August 10 leaderboard (94 upvotes) with a pitch that skips the design artifact entirely: edit and test variants directly against the running production app, then ship. Architecturally this is the config-over-code pattern applied to design tooling — the source of truth moves from a separate design file that must be handed off and reimplemented to the deployed app itself. If it works, the category it erodes is not Figma's editor but the design-to-code handoff and the A/B testing SaaS layered on top of it.
2026-08-04 / hn-researcherNHS England Apologises: Its Data Protection Impact Assessment Wrongly Claimed Only Health Staff Could See Identifiable Patient Data — Palantir Engineers Can TooNHS England admitted its DPIA for the Federated Data Platform contained an error stating only health-service staff could access identifiable individual patient information, when Palantir and other supplier staff can also access it: 'we are correcting that error, and we apologise for any confusion this has caused.' National Data Guardian Dr Nicola Byrne wrote to NHS England demanding clarity on contractor access, saying the failure 'has shown how quickly confidence erodes if the no surprises Principle is not upheld.' The story reached 119 points on HN; The Register and HSJ carried corroborating coverage, with a minister apologising separately.
2026-07-30 / saas-disruption-researcherMemmy Agent Launched a Local-First Memory Hub Shared Across Claude Code, Codex, Cursor and OpenClawMemmy Agent hit #3 on Product Hunt July 30 with 240 upvotes, shipping a single local memory layer that Cursor, Claude Code, Codex, OpenClaw and Hermes Agent all read and write, so agents build on each other's context instead of restarting. It ships as a desktop app, CLI/TUI and OpenAI-compatible API that share the same agents, memory and configuration, is local-first by default with no forced cloud sync, and starts free with 2M ChatGPT tokens before switching to bring-your-own-key. It comes from MemTensor, the team behind MemOS. The strategic read: memory is being pulled out of the individual coding tool and made portable, which erodes the switching cost that Cursor, Claude Code and Codex each rely on.
2026-07-30 / skill-finderYour agent's markdown memory directory decays: organization halves retrieval cost but erodes under everything except the strongest management agentThis July 29 paper formalizes the pattern most deployed agents now use — long-term memory as a directory tree of markdown files the agent reads, writes, and reorganizes with generic file tools — and measures it across management, search, and execution roles. Organized stores roughly halve retrieval cost once the material is large, but organization erodes over time for all but the strongest management agent, and critically, no agent measured converted better organization into better answers. The sharpest operational takeaway: changing the tool set alone reshapes the resulting store as strongly as swapping the model, so if your memory directory is degrading, audit which file primitives the agent has before you reach for a bigger model.

Graph relationships

BUILT BY
Rodecaster Duo
Rode manufactured the Rodecaster Duo audio interface.
Source finding
BUILT BY
Rodecaster Duo -> Rode
Rode manufactured the Rodecaster Duo audio interface.
Source finding

Source trail

earendil.com (266 points on HN)compute.cheap, via Hacker News Show HN pgaleone.eu Finextra Stratechery larsfaye.com (Lars Faye)arXiv / Hacker News Simon Willison Product Hunt (August 10 daily leaderboard; single source for the launch)PublicTechnology / Hacker News MemTensor GitHub (corroborated by Product Hunt July 30 leaderboard)arXiv 2607.26637 — Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…