News
Editorial dispatches and the latest from design tool blogs and changelogs.
Subscribe via RSSNew tools and editorial dispatches, straight to your inbox.
Lovable's apps can now call Claude directly — no API key needed
Building a chatbot or a PDF field-extractor into a Lovable app used to mean signing up for Anthropic and wiring in your own API key. Now it's a toggle, with five Claude models to pick from, Haiku up to Opus. It only covers AI inside the shipped app, not Lovable's own build process — but it closes a real gap between what a non-technical builder can ship and what used to need an engineer holding a key.
Penpot ships background blur, after rebuilding its renderer to make it possible
Glassmorphism has been requested for years, but Penpot's old SVG pipeline couldn't render it. Getting there meant swapping in a new WebGL renderer built on Rust, WebAssembly, and Skia — real engine work, not a checklist feature, and a sign Penpot is closing the rendering gap with Figma rather than just matching its feature list.
Framer's agent can now keep your rules, even after a template gets remixed
Skills lets a team write instructions once — how a component should behave, what has to stay consistent — and have Framer's agent apply them on every task after that, attached to the project so they survive being remixed into someone else's file. It's a real answer to the harder problem in agent-native design tools: holding onto judgment instead of re-deriving it every session.
Lovable splits thinking from building — chat is free, only shipped changes cost credits
Chats now runs on its own daily allowance, separate from the credits that pay for actual app changes, so you can think out loud, ask questions, or explore a dead end without touching the meter that decides whether you ship today. Most freemium AI tools charge exploration the same as output, which quietly discourages trying things — this is a rare move in the other direction.
Lovable splits its image generator into fast and final
Lovable added two OpenAI image models this week: GPT Image 2.5 Flare for quick previews, and Sunburst for the assets you actually ship. It's a quiet admission that one model can't do both jobs well - iterating fast and finishing well pull in different directions. Builders using Lovable's in-chat image generation can now ask for either by name.
Cursor's new Projects mode turns it into a standing team of agents, not a pair programmer
A coordinator agent plans a feature or migration, delegates the pieces to subagents, and keeps working in the cloud after you've closed your laptop — it can even pick up new work on its own when it spots a Slack message or a pull request. It's in beta, and the real question is whether the coordination holds up on an actual codebase rather than a demo. But the shift in posture is unmistakable: Cursor wants to be infrastructure you leave running, not a tool you sit down with.
Magic Patterns adds PDF upload, so a spec or deck can brief the agent directly
Drop in a spec, deck, or doc and Magic Patterns's agent treats it as context for what it builds next, instead of you retyping the brief into a prompt box. It's a small addition, and the category still cannot get output to shipping fidelity on its own, but starting from the document a team actually wrote is a real workflow fix, not a demo trick.
Framer's CMS gets a List field, so repeating content isn't manual anymore
FAQs, product features, case-study steps: anything repeating used to mean duplicating fields by hand in Framer's CMS. The new List field type handles it natively. Not flashy, but it's the sort of fix that saves real time once you're maintaining an actual site, not just demoing one.
Framer's agent can now move in 3D
Framer's AI agent now has full 3D transform controls: rotate on any axis, perspective, origin, depth, backface. Whatever it builds stays fully selectable and editable on canvas afterward. Most canvas-to-code agents stop at flat UI — this one reaches into actual space.
Codex 0.128.0: persistent goals and 90% faster task completion
Codex v0.128.0 ships persisted /goal workflows with TUI controls for long-running tasks, plus container caching that cuts median task completion time by 90%. Richer permission profiles and plugin marketplace improvements round out the release.
Zed 1.0: the Rust-built GPU-accelerated editor hits its first stable release
The former GitHub Atom team ships Zed 1.0, adding parallel AI agents, bookmarks, and a dedicated ‘disable all AI features’ mode for developers who want pure speed. Built on the custom GPUI framework, it stays native where Electron-based editors cannot.
GitHub Copilot for Visual Studio gets cloud agents, a Debugger agent, and chat history
The April Visual Studio update adds cloud agent sessions that create PRs while you work locally, a Debugger agent that validates fixes against live runtime behaviour, user-level custom agent definitions, and a chat history panel.
Cursor launches Security Review with AI agents that scan every PR and codebase for vulnerabilities
Cursor Security Review enters beta for Teams and Enterprise plans, adding two always-on AI agents: Security Reviewer (checks every PR for auth regressions, prompt injection, and privacy risks inline at the diff) and Vulnerability Scanner (scheduled codebase scans with Slack alerts). Both support MCP integration with existing SAST and secrets scanners.
Midjourney V8.1 launches on Discord and the web with improved sharpness
After the alpha period, V8.1 is now live on both Discord and midjourney.com with improved sharpness and quality in HD mode, SREFs, and Moodboards. SD is the temporary default while the team manages server capacity.
Rive AI Agent goes free — Ask mode and hourly-recharging credits
Rive updated its in-editor AI Agent with more efficient models that cost significantly fewer credits to run, and added an Ask mode for quick questions that do not need full agent invocation. Free plan users now get AI Agent access with credits that recharge every hour.
Midjourney opens 2K high-res rating to tune V8.1 and V8.2
Midjourney is running its first community image-ranking session at full 2K resolution, asking users to compare pairs of images to help tune upcoming aesthetic improvements in V8.1 and V8.2 toward natively HD output.
GitHub Copilot drops premium requests in favour of AI Credits from June 1
GitHub is replacing premium request units with token-based AI Credits from June 1. Subscription prices stay the same — $10/mo Pro, $39/mo Pro+ — but usage is now metered on tokens consumed, giving organisations pooled budgets and finer cost control.
Tokens Studio v2.11.5: variable scopes sync with Studio platform
Variable scopes and code syntaxes now sync from the Studio platform, and per-mode rem-to-pixel scaling is available. A focused release for teams running serious design token pipelines.
Google open-sources DESIGN.md, a file format for feeding brand systems to AI coding agents
Google Labs released DESIGN.md as an Apache 2.0 open-source format — a single YAML + markdown file that encodes a brand's design tokens and explains how to use them, so AI coding agents like Claude Code, Cursor, and Copilot generate on-brand UI from the project root.
Displacement Shaders add organic wave distortion to Jitter animations
Jitter added displacement shaders — a new animation type that applies organic, wave-like distortion to any element. You control amplitude, offset, repeat, and direction for both start and end states, making it useful for fluid reveals, ambient backgrounds, and motion-graphic transitions.
Anthropic launches Claude Design for quick visual creation
Powered by Opus 4.7, Claude Design lets non-designers generate prototypes, slide decks, banner ads, and one-pagers from text prompts — reading codebases and design files to apply brand guidelines automatically. Exports as PDF, URL, PPTX, or directly to Canva. Now in research preview on Pro, Max, Team, and Enterprise plans.
Affinity gets Canva Brand System and a Claude AI connector
Canva Create 2026 brings the Canva Brand System (colors, fonts, imagery, brand voice) directly into Affinity apps, plus an AI Connector powered by Claude for batch editing and print prep automation. The update also adds native Capture One and DaVinci Resolve file support.
Windsurf 2.0 ships Devin cloud agent and Agent Command Center
Windsurf 2.0 integrates Devin—Cognition's autonomous cloud coding agent—directly into every self-serve plan, letting you delegate complex tasks with one click while Devin works in its own cloud VM. A new Kanban-style Agent Command Center unifies local and cloud agent sessions in one view, with Spaces grouping related work across PRs, files, and context.
Sketch 2026.1.1 adds selection colors, independent borders, and full layer list in Inspect
Sketch's first 2026 update lands with selection colors, independent border controls, corner smoothing, and a revamped eyedropper with color variable support. The web app now shows a full layer list in Inspect mode, making it easier to navigate document structure without the Mac app.
Midjourney V8.1 Alpha: HD mode 3× faster and 3× cheaper
V8.1 replaces V8.0 Alpha with HD mode now 3× faster and 3× cheaper, standard resolution 50% faster and 25% cheaper, and image prompts and weights restored. A new prompt shortener handles long inputs, and V8.0 will be deprecated once V8.1 stabilises.
Claude Code gets a redesigned desktop app and Routines in research preview
The Claude Code desktop app (Mac and Windows) received a full visual overhaul alongside Routines — repeatable agentic workflows you define once and run on demand — now in research preview. The April sprint also shipped a Vertex AI setup wizard, Monitor tool, and /team-onboarding command.
Figma Weave community workflows now live
Weave workflows — reusable, shareable multi-model AI pipelines — are now available in Figma Community. Teams can duplicate and adapt workflows for generating images, videos, and illustration sets at scale.
Replit agent can reproduce a UI from a screenshot in a single pass
Replit's agent can now take a screenshot of any interface and reproduce the design at high fidelity in one shot—no iteration required. The same update adds Databricks Apps integration for enterprise, Razorpay payments for India, and geography-based deployment region selection.
Bolt switches default model to Claude Sonnet 4.6, adds layer picker and @ file references
Bolt’s April 6–10 update swaps the default model to Claude Sonnet 4.6, drops 4.5 variants, and adds two workflow improvements: a “Pick from layers” selector for overlapping elements and @ syntax for tagging project files directly in prompts.
Replit integrates RevenueCat so vibe coders can add subscriptions with a text prompt
Replit partnered with RevenueCat to let creators configure App Store billing, paywalls, and subscriptions through natural language — no manual SDK integration required. RevenueCat processes $1B monthly across 80,000+ apps, so the infrastructure behind the prompt is battle-tested.
Sketch 2026.1 "Dublin" adds selection colors, independent borders, and corner smoothing
The first major Sketch release of 2026 brings simultaneous color editing across selected layers, per-side border controls, a new corner smoothing panel with squircle and capsule support, and an eyedropper that can pick Color Variables. Over 150 additional improvements are included.
Windsurf launches Adaptive — an AI model router that picks the right model per task
Windsurf added an Adaptive model option that dynamically chooses the best AI model for each coding task, reducing premium model burn without sacrificing output quality. Available on Pro, Max, and Teams plans.
Design directly on production code, no translation layer, no intermediate files. The canvas sits on top of your actual codebase — design changes propagate to code, and code changes show up in the canvas. Founded by ex-Bookpad and ex-Leap founders, backed by First Round and Chemistry.
Noon is now live — a minimal, fast design tool built for the way designers actually work today. No subscriptions to start.
Tokens Studio v2.11.4 adds theme-set filtering and direct platform sync
The latest plugin release lets you filter token sets by enabled, disabled, or reference status inside the Theme editor — a real help for large multi-brand systems. It also gains the ability to pull tokens, sets, and themes directly from the Tokens Studio platform.
Compound shapes with inner cutouts, importable from Figma
Jitter now supports multi-path compound shapes with holes and inner cutouts — perfect for complex icon and illustration animations. Shapes can be built directly in Jitter or imported from Figma, and all paths stay individually editable for morphing.
Paper now lets you snapshot a live URL and paste it into the canvas as fully editable layers — built from the real HTML/CSS, not a screenshot.
New Layout Mode lets you rearrange and resize existing elements, add components, and get structured design feedback from the agent — all directly in the browser.
Spline ships Omma, a new product built on top of its 3D engine. Details still emerging but positions Spline further into interactive/generative territory.
Google Stitch gets infinite AI canvas, voice design, and MCP server
Google Labs updated Stitch with an infinite AI canvas, voice commands, and vibe design — a mode for starting from abstract ideas rather than wireframes. The tool now generates up to five screens at once and ships an MCP server for connecting AI coding tools like Cursor and Gemini CLI directly to your canvas.
Stitch relaunches as AI-native vibe design platform
Google overhauled Stitch into an infinite canvas you design by talking to. Voice Canvas, DESIGN.md for portable brand rules, Agent Manager for parallel directions, and an MCP server connecting to Cursor, Claude, and AI Studio.
Affinity 3.1: Convert to Curves, Live Tone Blend Groups, Light UI
First meaningful update since Canva relaunched the suite for free. Convert to Curves turns pixel selections into editable vectors, Live Tone Blend Groups add non-destructive compositing, and Light UI mode is finally available.
Sketch ships local MCP server and 2026.1 beta
Sketch quietly launched a local MCP server — connect Claude, Codex, or any MCP-compatible tool to your designs for code generation and design queries. Off by default, local only, so you control what gets access. The 2026.1 beta also lands with 150+ improvements: selection colours, a smarter eyedropper that picks up colour variables, and a redesigned workspace invite flow.
Tokens Studio 2.11.2 adds Studio platform sign-in to the Figma plugin
Version 2.11.2 adds a Subscription tab that lets you log in with your Studio platform account directly from the plugin, laying groundwork for consolidated account management.