AI Coding Tools

Oct 28

GitHub has taken its HydraFusion research preview to general availability as Copilot Fusion. You choose an effort level and Copilot routes each prompt across models using single, cascade or critique workflows. Here is what is confirmed, what GitHub claims, and how to use it.

Oct 27

Heal, a runtime-testing agent, published a benchmark claiming AI pull-request reviewers miss 70 percent of the user-facing bugs it proves with failing tests. The numbers are interesting and the method has real limits. Here is what is claimed, what the page itself concedes, and how to test it on your own repo.

Oct 27

Microsoft published microsoft/foundryx-plugin on October 27, 2026: a GitHub Copilot plugin with a foundry-agent skill and a preview Agent canvas that deploy and operate Foundry agents through a separate foundryx CLI. Here is what is confirmed, what is missing, and how to try it.

Oct 27

Ramp turned its internal background coding agent into a product on October 27, 2026. Its co-founder says 74% of merged code now comes from Inspect, up from near zero last November. Here is what is known about how it works and what is still a vendor claim.

Oct 26

On October 26, 2026 Cognition relaunched Devin Security with deep scans powered by Claude Mythos and other frontier cyber models, plus a pledge of hundreds of millions of dollars in Devin credits. Here is what the announcement says, what it leaves out, and how to evaluate the offer.

Oct 26

Oct 24

A controlled study of 8,032 trials finds that swapping the coding-agent harness (Claude Code, OpenCode, mini-SWE-agent) moves SWE-bench Verified pass rates about as much as rerunning the same setup, while cost per task can triple. Here is what the numbers say and what to change in your own setup.

Oct 23

Cursor announced a Computer Use tool on October 23, 2026 that lets its agents click and type across any app on your computer, not just the browser. Here is what Cursor has said, what its docs already confirm about cloud agents, and the security questions to ask before turning it on.

Oct 21

Oct 20

Self-HPO, from Microsoft Research and the University of Virginia, alternates two steps: the model rewrites the harness around it, then reinforcement learning trains the model inside that harness. Starting from Qwen3.5-35B-A3B, it lifts the average resolve rate over three SWE-bench variants from 57.8% to 64.8% with no stronger teacher model.

Oct 20

As Steam Next Fest began on October 19, 2026, IGN reported players complaining about AI-generated art and vibe-coded games and asking Valve for a filter. Valve has required AI content disclosure since January 2024, so a filter is technically close, but tags, definitions and enforcement are the real problem. Here is the situation, the options, and what developers and players can do now.

Oct 20

Spinoza Labs launched Vortex Code, a coding CLI with Agent, Build and Architect modes, and published demo counts such as 173,773 lines and 1,111 commits for one finance terminal. We separate what the company claims from what can be checked, and give you a way to test it.

Oct 19

Oct 18

A solo developer published git-ai (gitai), an MIT-licensed server that implements the source-control design Anthropic described for Scion. It gained 1,890 GitHub stars in three days. This post explains what it does, how to try it, and what is still unproven.

Oct 17

Oct 17

OpenCode announced a security breach tied to a large-scale distillation attack that affected 1.4% of its Console users, specifically those with external provider keys linked. This post separates the maker's own statement from second-hand reports and gives a rotation checklist.

Oct 16

PostgreSQL 19 is now expected around 19 November 2026. A core committer says LLM-assisted code review found defects faster than the project could fix them, forcing 8 to 10 features out of the release. Here is what happened and what it signals for AI-assisted engineering.

Oct 15

Oct 15

At TechCrunch Disrupt on October 15, Bluesky CEO Toni Schneider announced that Attie, the company's experimental AI app, can now build interactive sites, quizzes, games and polls for communities on AT Protocol. Users can bring their own model or use open-weight options.

Oct 15

Cognition announced Devin Oncall on October 15: Devin automatically investigates alerts and bugs to help teams diagnose and resolve production incidents. The headline number, thousands of alerts a day at some of the world's largest companies, is Cognition's own claim.

Oct 15

A report says Cursor engineers raised alarms in July that confidential client data could flow into SpaceX AI training through an internal tool. No breach is confirmed. Here is what the sources say, what they do not, and what Cursor users should check.

Oct 15

OpenCode now lists DeepSeek V4.1 Flash Free at zero cost on its Zen pay-as-you-go provider, while the paid Go plan counts the same model against an $80 monthly allowance. Here is exactly what the official pages say, what they do not, and how to use it without getting burned when the promo ends.

Oct 15

XERJ, a Rust search engine that indexes your code, docs and logs in one command, gained over 700 GitHub stars in a day. Its headline claim is 40x fewer tokens than grep, but the README body measures 2.7x against grep in its own study. Here is what it does, how to try it, and where it is risky.

Oct 14

Epic Games released a preview of a rebuilt Epic Games Launcher for Windows, Mac and Linux and said in its own blog post that it was vibe coded. Epic claims it starts faster and reaches the store 46 percent faster than Steam; outlets have not yet tested that. Here is what is confirmed and what is not.

Oct 14

The Last Secure Code Benchmark builds 450 tasks from 150 real vulnerabilities across seven languages and asks agents only for functionality. Functional pass rates reach 91.3%, yet the best agent produces code that is both working and secure on just 25.1% of tasks.

Oct 14

A new Microsoft repo, vscdeck, is a small always-on-top Windows utility that reads Copilot session files to show which VS Code windows are idle, running or waiting for you. Here is how it works, how to build it, and where it can break.

Oct 13

On October 13, 2026, Google said users 18 and older in select markets, including the US, can build personal Android apps in AI Studio and launch them privately on their own devices through Google Play with only a standard Google account. Sharing with friends is promised for the coming months.

Oct 12

A Harvard job-market paper using Jellyfish data from over 700 software firms finds that AI coding agents raise lines of code, commits and pull requests, but the rate of finished Jira issues does not move. Code review absorbs the gain. Here is what the study measured, what it cannot prove, and what to change in your own workflow.

Oct 12

Bloomberg reports that Epic Games has delayed its Disney-licensed shooter and told the team to adopt an "AI first" production plan, with no hand-written code and art made through full AI workflows. Epic confirms it is embracing AI-accelerated development. Here is the claimed-versus-verified breakdown, the context, and what studios and developers should take from it.

Oct 12

OpenCode 2.0 turns the open-source coding agent into a programmable platform: one shared background server for terminal, web and desktop, a JavaScript Code Mode for chaining MCP tools, background agents, hot-reloading plugins and a new SDK. 1.0 plugins break.

Oct 10

Kotaku reported on October 9, 2026 that hundreds of AI-decompiled or vibe-coded game clones now run in web browsers, including Halo: CE, GTA: Vice City and Call of Duty: Black Ops. Here is what is claimed, what is verified, why AI changes the economics of reverse engineering, and why takedowns are a whack-a-mole.

Oct 9

AgentPlane, announced on October 8, 2026, is a local web app that runs many coding agents on your own machine, with one approval inbox, per-turn revert and optional git worktrees. We read the README. Here is how it works, and how it differs from AgentsInTheCloud.

Oct 9

At Open Source Summit Europe, Linus Torvalds said he really likes using AI, defended vibe coding for newcomers, and described AI review and bug reports as both improving Linux and stressing maintainers. Here are his words, in context.

Oct 9

The pingdotgg team ported the Go-based TypeScript 7 compiler to Rust almost entirely with LLMs. The README claims 181,711 ported tests pass and about half the type-check time of Go. It also says nobody read the code. We read the repo and the Hacker News thread.

Oct 7

Oct 5

Cognition gave Devin persistent memory stored as a git repository of markdown files, plus a daily Dreaming pass that merges and prunes it. The format is an open spec that other agents can use too. Here is how it works, how it compares, and where it can go wrong.

Oct 2

On October 2, 2026, Andrej Karpathy argued that as intelligence and code become abundant, you can ask for large, custom, discardable software artifacts, such as web apps and video explainers, that would never have made sense to build before. The shift is economic, not a new model capability.

Sep 27

José Valim's September 24, 2026 essay asks what programming languages should optimize for once coding agents are users. The practical answer is three surfaces: explicit types, a queryable program database, and runtime state an agent can inspect.

Sep 27

Priyan R spent months handing Prince of Persia to frontier coding agents and only playing the result. Opus 5.5 got a level-1 screen from 8,429 differing pixels down to 2. The useful part for anyone grading agents is the oracle and the diff, not a leaderboard of model names.

Sep 26

On September 25–26, 2026, Google Antigravity shipped a dedicated planning mode: type /plan and the agent researches, drafts an implementation plan for your review, and waits for approval before editing code. The same flow exists on the Antigravity CLI. You can also ask for a plan in plain English for a lighter variant — explainx.ai maps how /plan fits next to /boost, Teamwork, and harness design patterns builders already use elsewhere.

Sep 25

Sep 24

Sep 22

Sep 19

Sep 19

Sep 18

Sep 18

Sep 16

Sep 15

Sep 14

Sep 12

Sep 11

Sep 11

Sep 11

Sep 10

Sep 9

Sep 8

Sep 5

Sep 5

Sep 3

Sep 3

Sep 3

Sep 1

Sep 1

Sep 1

Aug 29

Aug 29

Aug 29

Aug 28

Aug 27

Aug 27

Aug 27

Aug 25

Aug 24

Aug 24

Aug 23

Aug 22

Aug 22

Aug 22

Aug 22

Aug 20

Aug 20

Aug 20

Aug 20

Aug 19

Aug 18

Aug 17

Aug 16

Aug 14

Aug 8

Aug 5

Aug 5

Aug 5

Aug 4

Aug 2

Aug 2

Aug 1

Jul 31

Jul 29

Jul 27

Jul 27

Jul 26

Jul 24

Jul 23

Jul 23

Jul 23

Jul 23

Jul 22

Jul 22

Jul 22

Jul 20

Jul 19

Jul 17

Jul 15

Jul 13

Jul 13

Jul 9

Jul 5

Jul 3

Jul 1

Jun 30

Jun 29

Jun 29

Jun 27

Jun 27

Jun 27

Jun 27

Jun 25

Jun 24

Jun 23

Jun 19

Jun 19

Jun 19

Jun 17

Jun 16

Jun 11

Jun 11

Jun 8

Jun 7

May 21

May 16

Apr 27