Tech Meridian ← ENTITY INDEX
PROMY MERIDIAN RU

COMPANY · ENTITY #39

GitHub

Related event timeline, sources and context from the news index.

EVENT TIMELINE

76

COMPANIES · 1 SOURCE · TechCrunch AI

Y Combinator’s Garry Tan says he spends $100k–$200k/year on AI agent tokens and shared 'gstack' Claude Code skills

On Walter Isaacson’s podcast The Builders, Y Combinator CEO Garry Tan said he spends about $100,000–$200,000 a year on tokens for AI agents, uses agents for many work and personal tasks, and has published a set of 23 Claude Code skills called 'gstack' on GitHub; he also described an internal YC agent named 'Stray Light' that holds company-wide founder context. Tan said high costs are a barrier to mainstream personal agents and expressed concern about agents controlled by large tech platforms.

6.0

COMPANIES · 1 SOURCE · AWS Machine Learning

Circle K automates DevOps triage with Amazon Bedrock, cuts mean time to triage to under 15 minutes

Circle K built an automated, agentic triage system on Amazon Bedrock (initially with Bedrock Flows, later migrated to Bedrock AgentCore) that queries logs, services, and internal runbooks and posts structured summaries to Slack; the team reports mean triage time fell to under 15 minutes (a 67–75% reduction by their metrics). The implementation used managed Bedrock features (Agents/AgentCore and Bedrock Knowledge Bases for RAG) to avoid self-hosting models and orchestration code.

6.0

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA publishes NemoClaw blueprints to assemble secure, optimized agent stacks

NVIDIA introduced NemoClaw, a collection of open blueprints that assemble a full-stack agent platform by combining tuned agent harnesses, Nemotron models, NeMo optimization tools (including NeMo Relay), and the OpenShell secure runtime. The blueprints support OpenClaw, Hermes Agent, and LangChain Deep Agents, include a LangChain-tuned Nemotron 3 Ultra profile (near-frontier performance at about 10x lower cost per run), and provide GitHub examples, a quickstart, and a DLI course.

7.0

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

Google makes Data Agent Kit generally available, splits agent guardrails between skills and IAM

Google has made its Data Agent Kit generally available as an open-source, free integration that connects coding agents to over fifteen Google Data Cloud services and ships 37 versioned skills (Apache 2.0). The release emphasizes two control layers—advisory ‘‘skills’’ that teach best practices (e.g., accidental-data-loss-prevention) and enforceable access controls via IAM (including deny rules like blocking mcp.googleapis.com/tools.call)—while noting limits (blocked tools still appear in listings, controls cover only Google MCP servers, and some attributes are gcloud-only).

7.0

RESEARCH · 1 SOURCE · arXiv cs.AI

Agent4Sci testbed (arXiv:2610.28763v1) evaluates frontier coding agents on 20 ML‑for‑science benchmarks

The paper introduces Agent4Sci, a long‑horizon testbed that measures how well current coding agents can improve published best models across 20 scientific ML benchmarks in 11 domains. Empirically, frontier agents (e.g., Claude Opus) improve published baselines (Claude Opus: 10.4% mean relative improvement), but gains mainly come from ML‑engineering tricks; simple plug‑in skills for ideation and experiment management boost Claude Opus to 20.1% and Codex to 8.1%. Code and data are available on GitHub and Hugging Face.

7.0

RESEARCH · 1 SOURCE · arXiv cs.AI

RegretBench and BackUpper: measuring and automating backups for coding agents

An arXiv paper (arXiv:2610.28934v1) introduces RegretBench, a 205-task benchmark of real-world "regret" moments where users wished they'd preserved state, and BackUpper, a harness plug-in that checks tool calls and preserves affected state before modifications. Across six frontier models agents complete 86.1% of tasks but back up before regret moments only 15.2%; BackUpper covers all regret actions, restores states with 92.2% average similarity, costs about $0.014 per task, uses 7× fewer tokens than a self-backup baseline, and can match user backup preferences with an 84.5% F1 score.

7.0

RESEARCH · 1 SOURCE · Hugging Face

ThinkingBox-Training released on GitHub; RL fine-tune raises Qwen3.8-27B reliability using database-graded rewards

ThinkingBox-Training is now available on GitHub. The toolkit turns executable checks of terminal database state into RL rewards; fine-tuning Qwen3.8-27B with 50 GRPO steps raised average pass@1 from 51.70% to 60.78% and increased tasks solved on all 20 attempts from 38 to 107, using an offline policy-relative difficulty selection of training tasks and evaluation on the 507-task ThinkingBox-Bench.

7.0

RESEARCH · 1 SOURCE · InfoQ AI, ML & Data Engineering

Harvard–Jellyfish study: AI coding tools raise code volume but not completed work due to review bottlenecks

A Harvard study using Jellyfish data (≈300 million work events across ~700 firms, Jan 2021–Mar 2026) finds AI coding assistants modestly increase code output (lines +12%, commits +9%, PRs +5%; only commits statistically significant) while AI agents produce larger increases (lines +30%, commits +20%, PRs +23%). Despite more code, assistants did not increase resolved Jira issues and agents raised resolved issues by only 12%, with significant downstream impacts: average PR review time +49%, comments per PR +53%, share of PRs requiring changes doubled, and share of engineers performing reviews +14%—suggesting review and other complementary stages are bottlenecks.

7.0

RESEARCH · 1 SOURCE · The Decoder

OpenAI agents using GPT‑Rosalind analyze mouse whole‑brain CRISPR atlas in 32 minutes

OpenAI, with PerturbAI and Scripps Research, deployed thousands of AI agents running on GPT‑Rosalind to analyze a whole‑brain CRISPR atlas covering millions of mouse cells; the distributed analysis finished in 32 minutes, produced a public "Reasoning Atlas" on GitHub, and is described in a bioRxiv paper. The agents identified unexpected differences between two disease-linked genes and a possible link between cholesterol regulation and neurodegeneration, and in blinded tests experts preferred distributed-agent reports over single-agent reports 75% of the time.

8.0

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

Developer: AI-built PhotoCraft Adobe clones ‘not anywhere close to ready’

Brandon Thomas released PhotoCraft — an open-source suite of seven Adobe-like apps generated with Claude Opus 5.5 (Rust + WebAssembly) — but the project’s ROADMAP.md, issues and user reports show it is still alpha. The repo (MIT/Apache) claims ~76% feature breadth but only ~45% readiness for real work, estimates beta at 1,100–1,800 Opus agent hours and full parity at 2,300–4,100 hours, and records heavy activity (2,152 commits, 1,004 merged PRs, ~539k Rust lines, 317 contributors; Claude appears as a commit author 277 times).

6.0

CODING · 1 SOURCE · Anthropic

Agent Skills standard donated to Agentic AI Foundation (AAIF)

The maintainers of the Agent Skills open standard announced they are donating the Agent Skills name, logo, GitHub organization, and agentskills.io domain to the Agentic AI Foundation (AAIF) under the Linux Foundation. Agent Skills — an open standard for packaging procedural knowledge for AI agents — has been integrated into over 70 platforms (including ChatGPT, Cursor, GitHub Copilot, and VS Code), with about 1 million published skills and nearly 280 million installs to date; David Soria Parra and Mahesh Murag are named as Lead Maintainers and a roadmap, proposal process, and security guidance were proposed.

8.0

CODING · 1 SOURCE · AWS Machine Learning

AWS open-sources Deep Insight, a production multi-agent system using Amazon Bedrock AgentCore

The AWS Solutions Architecture team published Deep Insight, an open-source, production-grade multi-agent system and a three-part blog series detailing its architecture, Context Engineering practices, and deployment. Deep Insight (GitHub) includes self-hosted, web, and a managed AgentCore deployment that runs agents in isolated per-user environments inside a private VPC with a Code Interpreter on AWS Fargate.

6.0

MODELS · 2 SOURCES · NVIDIA Developer · Hugging Face

NVIDIA BioNeMo releases RFD4-Proteina: reward-guided protein design model and NIM microservice

RFD4-Proteina, an open-model co-developed by NVIDIA BioNeMo and the Institute for Protein Design (UW), co-designs protein sequences and all-atom structures and uses reward-guided inference-time search to focus compute on promising candidates. The method uses a residue-level latent representation and engineered diffusion to cut sampling from ~400 to 60 steps, yielding ~3.8× faster generation and ~84% lower peak memory than RFD3 at 200 residues; the project provides an open GitHub model for development and a NIM microservice with an HTTP interface for rapid inference and automated campaigns (examples include PD-L1 binder design with hotspot constraints).

8.0

CODING · 1 SOURCE · GitHub AI & ML

Java Champions report using GitHub Copilot to accelerate coding and prototyping

Three experienced Java Champions describe using GitHub Copilot (in IntelliJ and the Copilot app) to speed analysis, design and implementation—examples include building a web UI for Hypersistence Optimizer in two days and migrating complex teaching slides. They outline practices for reliable use: reviewing AI output, creating custom “skills” to match their coding voice, selecting models (e.g., Opus with a 1M context window) by task, and integrating Copilot into team coaching and maintenance workflows.

4.0

CODING · 1 SOURCE · GitHub AI & ML

Open GitHub Copilot app sessions directly in VS Code

GitHub documents how to open a Copilot app session in Visual Studio Code while preserving the session’s task, plan, conversation, diff, and isolated worktree. This lets developers review agent-made changes, make manual edits, run local validation, and push updates to the pull request tracked by the session without recreating the session history.

4.0

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

Reverse Engineer Anything (REA) links AI coding agents to Ghidra, Hopper and IDA

Reverse Engineer Anything (REA) is an open-source project (200K+ GitHub stars) that exposes reverse‑engineering tools such as Ghidra, Hopper and IDA to coding agents like Claude Code and OpenAI Codex via MCP servers, enabling agents to query disassemblers, trace functions, and iteratively inspect binaries locally; it also supports static analysis for JavaScript/Electron and .NET and is published under the MIT license.

7.0

STARTUPS · 1 SOURCE · TechCrunch AI

Replit CEO Amjad Masad says AI agents are creating a 'self-driving company'

At TechCrunch Disrupt, Replit CEO and co-founder Amjad Masad described a "self-driving company" model in which AI agents handle much of the routine work (e.g., triaging support tickets by inspecting GitHub issues, the codebase, and Linear tickets) while humans retain judgment and taste. He said Replit's operations have shifted to an agent-driven loop that lets people focus more on architecture, product decisions and creative work.

5.0

MODELS · 1 SOURCE · GitHub AI & ML

GitHub deploys ModernBERT-based context-aware classifier to reduce secret scanning noise

GitHub replaced an LLM-heavy secret-scanning pipeline with a two-stage system: fast regex candidate generation followed by a fine-tuned ModernBERT classifier that evaluates each candidate plus surrounding context. The new pipeline yields an 86% faster median post-push scan execution time, over 98% lower serving cost versus the previous LLM-based approach, and an 11.5% improvement in user acceptance by reducing false positives.

7.0

MODELS · 1 SOURCE · The Decoder

OpenAI releases 700+ claimed mathematical proofs; mathematicians react with shock

On October 6, 2026 OpenAI uploaded more than 700 manuscripts to GitHub claiming solutions to hundreds of open mathematics problems; the release prompted over 100 responses collated by the blog Proofs and Prompts. Experts are divided — some say the results contain striking ideas, others warn many papers are hard to read, some withdrawn, and the broader consequences for research practice, careers and even cryptography are being debated.

8.0

RESEARCH · 2 SOURCES · TechCrunch AI · The Verge AI

OpenAI releases 719 AI-generated math manuscripts; researchers say sorting truth from 'slop' could take years

OpenAI published a sprawling GitHub repository containing 719 manuscripts with nearly 400 AI-generated mathematical results across many subfields; the company said about 300 top-line results (≈42%) have formalizations, and it plans to add more. Mathematicians told The Verge the volume and uneven verification quality—only some proofs are formalized in Lean and formalizations vary in quality—will likely require years of community work to assess and verify.

9.0

COMPANIES · 3 SOURCES · Anthropic · The Verge AI · The Decoder

Anthropic launches Cyber Mission with CIDP and OSS Scanner to support defenders

Anthropic announced the Cyber Mission, a long-term effort to help secure critical infrastructure and open-source software by providing frontier Claude models, on-site engineers, threat research, and funding. Initial components include the Critical Infrastructure Defense Program (CIDP), launched with founding partners such as Accenture, Booz Allen, CrowdStrike, Deloitte and others, and OSS Scanner, a free model-powered security scanning service for open-source projects.

8.0

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

GitHub migrates Copilot runtime from Node.js/TypeScript to Rust via AI-assisted rewrite

GitHub replaced the runtime powering GitHub Copilot CLI, the Copilot app, and the Copilot SDK by migrating from a TypeScript/Node.js implementation to a native Rust runtime through an AI-assisted rewrite that touched more than 800,000 lines of production code. The incremental 14.5‑week migration (128 pull requests, 135 releases) cut measured client startup/session latency from 5.25 seconds to 292 milliseconds, enables embedding the runtime via a C ABI (reducing ~100 MB per client vs Node/V8), and combined AI-generated implementation with compilation, testing, and human review.

8.0

COMPANIES · 1 SOURCE · TechCrunch AI

Meta’s Muse launches dedicated iPad app one month after iPhone debut

Meta has released a dedicated Muse app for iPad, roughly one month after Muse’s iPhone launch. The assistant, which has an estimated 6.6 million installs across iOS and Android, adds new connectors (e.g., Asana, Canva, Dropbox, Figma, QuickBooks, GitHub, Klaviyo, Zoom, Notion, Granola) and retail partners (Best Buy, Gap, Sephora, Walmart, Wayfair) to enable account linking and shopping/transaction capabilities.

6.0

COMPANIES · 1 SOURCE · GitHub AI & ML

GitHub adds fine‑tuned classifier to extend push protection for unstructured secrets

GitHub reports that AI agents now appear in roughly one in three pull requests and says secret exposures are doubling yearly; to scale prevention it introduced a fine‑tuned classifier (built with Microsoft Applied Sciences) that evaluates candidate secrets in under two milliseconds to extend push protection to unstructured secrets. GitHub also published operational metrics: a public secret appears about once every two seconds, push protection prevents roughly 30% of newly detected secrets before they enter repo history, screened pushes grew 2.84× while pushes carrying credentials grew 2.59× between Q2 2024 and Q2 2026, and mean manual revocation time remains around 40 days.

7.0

COMPANIES · 1 SOURCE · The Verge AI

Meta’s Muse adds native iPad support

Meta updated its Muse AI agent app to add native iPad support, optimizing the interface for larger screens and iPadOS multitasking. The release notes also list new third‑party connectors (e.g., Canva, Dropbox, Figma, QuickBooks, GitHub, Zoom, Asana, Klaviyo, Notion) and a business‑goal feature for small‑business workflows; a separate Mac version exists and Meta has announced a standalone Muse Charm device.

5.0

COMPANIES · 1 SOURCE · WIRED AI

Mathematicians criticize OpenAI over planned release of AI-generated math solutions

WIRED reports that OpenAI told about 40 mathematicians its internal model has resolved the Navier–Stokes Millennium Prize problem and more than 100 other long‑standing open math problems, and that the company planned to publish hundreds of results (reportedly via GitHub). Attendees and other mathematicians say OpenAI and Anthropic’s release practices — blog posts and quick dumps rather than peer‑reviewed papers and proper attribution — risk undermining academic norms and have led to disputes over credit, including an allegation by Tristan Buckmaster about front‑running related work with an Anthropic collaborator.

8.0

RESEARCH · 1 SOURCE · GitHub AI & ML

ReviewBench: an open benchmark for AI code review

ReviewBench is a new publicly available offline benchmark for AI code review agents that contains 219 public pull requests across 19 languages, modeled on GitHub distributions and validated by senior engineers. It builds a multi-source golden set (human reviewers, follow-up commits, static tools, and multiple LLMs), uses Claude Sonnet 5 as a consistent grader, and provides metrics and labeled findings (severity and category) to compare reviewers and track likely production impact.

6.0

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

InfoQ: Platform engineering playbook for production LLMs

InfoQ publishes a platform-engineering playbook based on a retail inventory recommendation system that reduced hallucination rate from ~15% to ~1.5% without changing the foundation model by applying platform-level controls (automated retry loops, intent-validation gates, prompt registries, tool authorization, and custom instrumentation). The article links a companion GitHub repo with runnable examples that use a local llama3.1 via Ollama connected through LiteLLM and describes routing through foundation models such as Gemini and GPT under enterprise traffic.

7.0

RESEARCH · 1 SOURCE · The Decoder

Google researchers propose RRSI to stop agent harnesses memorizing tests

Google Cloud AI Research and academic collaborators introduce RRSI (Regularized Recursive Self-Improvement), a method that constrains automated harness rewrites so agentic systems avoid overfitting to benchmark tasks. Tested with a frozen Claude Opus 4.8 across eight benchmarks (including JobBench and ARC-AGI-3), RRSI improved training-task scores by up to 14.1 points, boosted performance on unseen tests by up to 4.7 points, and cut runtime token use by about 30%; the paper and code are on GitHub.

7.0