Explore what's new in AI

Supervibes bundles app building, backends, and App Store releases

The hosted app-building agent combines a simulator, backend provisioning, an AI gateway, and TestFlight and App Store delivery to reduce setup friction.

A code review workflow uses before-and-after agent diagrams

Mapping changed flows onto a canvas gives reviewers a place to question agent-produced architecture changes, making control flow easier to inspect than a diff alone.

Zed 1.25.2 adds direct editing of inlay hints

Double-clicking an inlay hint or invoking an editor command now applies its text edit, a small workflow improvement in Zed's latest stable release.

Amp adds a BYOK option for GitHub Copilot subscriptions

Developers can connect an existing Copilot subscription inside Amp, offering another way to use a coding assistant without switching plans.

A developer recommends rebuilding legacy WordPress sites with coding agents

For small sites facing PHP upgrade work, one proposed alternative is an agent-assisted rebuild in a modern framework rather than preserving a heavier legacy stack.

Decks turns slide comments into agent-driven presentation edits

Decks lets reviewers annotate slides and hand edits to coding agents, with per-slide history and clarity checks that make revisions easier to inspect.

PandaCode runs multiple coding CLIs in one desktop workspace

PandaCode combines coding agents with Git, terminal, and browser panes, plus an architect agent that can delegate work to workers.

Devin configures and tests iPhone Duo apps through terminal and simulator

A demo shows an agent moving between terminal commands and a mobile simulator to configure and test an iPhone app, illustrating an automated native-app workflow.

FrankenTorch attempts a PyTorch-compatible tensor stack in Rust

The project targets PyTorch behavior including autograd and serialization, with parity tracking; it is an ambitious experiment rather than a proven drop-in replacement.

VS Code defaults new Copilot sessions to the SDK-based harness

New Copilot sessions in VS Code default to the SDK-based harness and can be shared between the editor and Agents window, with additional model comparison and diff features.

Pierre releases Highlights, an AST-based syntax highlighter

Highlights supports more than 70 languages and aims to reduce highlighting latency and memory compared with Shiki, offering another option for code-rendering systems.

Zed says v1.25 binaries are about 30% smaller than v1.23

Zed's release reduces binary size and adds ACP context compaction, session notices, editable inlay hints, and more model providers for developers using the editor.

Amp adds interactive mini-apps to agent threads

Amp can present interactive mini-apps within agent conversations, allowing users to provide input and explore results beyond a plain-text response.

T3 Code reports processing 36.5 trillion tokens in 72 hours

The coding product says it processed about 1.5 times Gemini 4 Argon's reported global demand in the same window, a striking usage signal for agentic coding.

OpenAI describes how ChatGPT ads stay separate from answer generation

A separate ad orchestrator retrieves, ranks, and auctions placements, while timing and relevance checks prevent ads from changing answers or delaying responses.

scriptc and tsc-ts now compile and type-check each other

The TypeScript compiler and type checker have reached a mutual self-checking loop, a concrete example of bootstrapping language tools and validating their implementations.

OpenAI details the retrieval and auction tools behind ChatGPT ads

The ad orchestrator turns conversation context into intents, retrieves candidates, checks relevance and eligibility, then handles auction and pricing separately from answer generation.

Robert Martin describes agents building and checking a system from a spec

His workflow has agents draft a specification, implement it, then run checks to find bugs and dead code, illustrating a verification-heavy development loop.

JetBrains offers Junie Lite free for everyday coding tasks

The free Junie CLI model handles routine edits such as renaming fields and scaffolding tests, giving developers a low-friction way to try an agent.

Genspark recaps ten updates, including its GenCode CLI assistant

Recent additions span project memory, desktop AI for documents, a CLI coding assistant, and messaging-app tasks, giving builders a map of workflow integrations.