What AI Update Did You Miss Today?

Cursor 3.0: The Agents Window and New Interface

Cursor released version 3.0 on April 2, 2026, marking a fundamental shift in the product's identity: from an AI-augmented code editor to an agents-first workspace. The new Agents Window lets developers run multiple agents simultaneously across local repositories, git worktrees, cloud infrastructure, and remote SSH connections — all from a single unified interface. Key additions include Agent Tabs for side-by-side multi-chat views and new commands like /worktree and /best-of-n for isolated execution and parallel model comparison.

Cursor

Gemma 4: Google's Open-Weight Models Now Available via Gemini API

Google DeepMind released Gemma 4 on April 2, 2026 — a family of four open-weight models built on the same research as Gemini 3, now available under the Apache 2.0 license. The two server-scale variants, gemma-4-26b-a4b-it (26B Mixture-of-Experts with 4B active parameters per token) and gemma-4-31b-it (31B Dense), are immediately accessible via the Gemini API and Google AI Studio. The 31B model ranks #3 on the Arena AI text leaderboard among all open-weight models, while the 26B MoE secures #6 despite activating only a fraction of its weights per token. All Gemma 4 models are natively multimodal, supporting text, image, and video input, with 256K context windows on the server-scale variants.

Gemini CLI

Cursor Design Mode: Visual Browser Annotation for Agents

Cursor introduced Design Mode as part of the Cursor 3.0 release, enabling developers to directly annotate and target UI elements inside a live browser while working with agents. Rather than describing interface changes through text alone, users can select specific elements, highlight areas, and add them directly to an agent chat — making UI feedback significantly more precise. Design Mode operates as two loops in one: a visual loop for direct style adjustments and a code loop where the agent finds and edits the underlying source automatically.

Cursor

GitHub Copilot SDK: Public Preview

GitHub has released the **Copilot SDK** into public preview, giving developers a multi-language toolkit to embed Copilot's agentic runtime directly into their own applications and services. The SDK exposes the same production-tested agent engine that powers the Copilot cloud agent and Copilot CLI — providing tool invocation, streaming, file operations, and multi-turn sessions out of the box. Available for Node.js/TypeScript, Python, Go, .NET, and Java, the SDK supports custom tool definitions, BYOK (Bring Your Own Key) for OpenAI, Azure AI Foundry, and Anthropic, and OpenTelemetry distributed tracing. GitHub Copilot Free subscribers can access the SDK at no extra cost, with paid subscribers drawing from their premium request quota.

GitHub Copilot

GitHub Copilot in Visual Studio: Custom Agents via .agent.md Files

The March 2026 Visual Studio update brings **custom Copilot agents** to Visual Studio, allowing developers to define specialized agents as `.agent.md` files directly in their repositories. These agents gain full workspace awareness, access to preferred models, and connections to external knowledge sources via MCP — and appear automatically in the agent picker for team use. Alongside custom agents, **agent skills** arrive as reusable instruction sets that teach agents to perform specific repeatable tasks, with Copilot auto-discovering them from repos or user profiles. Together, these features turn GitHub Copilot in Visual Studio into a fully extensible platform for team-specific AI workflows.

GitHub Copilot

Warp: Interactive Agent Conversations with Clarifying Questions and Follow-Up Suggestions

Warp introduced a significant upgrade to its agent interaction model, enabling agents to ask users clarifying questions mid-task when additional context is needed. Warp agents now also proactively suggest follow-up actions upon completing a task, creating a more guided and conversational development workflow. These two changes together transform Warp's agent from a one-shot executor into a true interactive collaborator capable of iterative back-and-forth dialogue.

Warp

Warp: Send Code Review Context Directly to Claude Code, Codex, and Other CLI Agents

Warp now lets developers send code review comments, attach diff hunks as context, and forward substrings of output to third-party CLI coding agents — including Claude Code, Codex, and Opencode — directly from within the terminal. This integration bridges Warp's visual code review panel with the growing ecosystem of terminal-based AI coding agents, enabling seamless handoff between reviewing code and instructing an agent to act on what was reviewed.

Warp

Lovable: Six New Connectors — Aikido, GitLab, AWS S3, Hex, Confidence & PostHog

Lovable expanded its connector ecosystem with six new integrations across security, DevOps, cloud storage, and product analytics. The **Shared Connectors** — Aikido, GitLab, and AWS S3 — enable AI-powered penetration testing, two-way repository sync, and direct S3 file access respectively. The **Personal Connectors** — Hex, Confidence, and PostHog — give the Lovable AI agent access to data notebooks, feature flag experiments, and product analytics for more context-aware development.

Lovable

Lovable: Visual Edits Now Works Across the Entire App

Lovable extended its **Visual Edits** feature to work across the entire application, including pages that display database-driven and API-populated content. Previously limited to static UI elements, Visual Edits now allows users to select and update any UI component — including those rendered from live data — without writing code. The feature remains free within daily usage limits, making visual UI editing broadly accessible.

Lovable

Lovable: Project Comments for Collaborative Feedback

Lovable launched **Project Comments**, a collaborative feedback feature that lets users leave inline annotations directly on app preview elements. Comments support threaded discussions and can be sent to the AI agent using @Lovable mentions, turning feedback into actionable development tasks. Collaboration features — including Project Comments — are now free for all Lovable users.

Lovable

Warp: Cloud Agents Now Accept Image and File Attachments

Warp's cloud agents (Oz) can now accept images and file attachments as context — a capability previously limited to local agent conversations. This removes a long-standing limitation for developers who rely on cloud-based agent sessions, enabling richer, visually-grounded prompts for remote tasks such as UI debugging, design-to-code workflows, and documentation-driven development.

Warp

Figma: Make Kits — Design System Context for AI Prototypes

Figma introduced Make kits, a new feature that lets design system teams package their components, styles, variables, and tokens into reusable kits that teach Figma Make how to properly use an organization's design system when generating AI prototypes. Design system authors can import code components via npm packages (public or private to a Figma organization) alongside Figma library styles and tokens, then publish these kits organization-wide. The result is that AI-generated prototypes start from production-aligned components rather than generic placeholders, closing the gap between AI output and real codebases. Make kits began rolling out to all paid plans on March 26, 2026.

Figma

Lovable: Generate Files and Analyze Data

Lovable introduced the ability to execute code and generate downloadable files — including PDFs, Excel spreadsheets, and PowerPoint presentations — directly within conversations. Users can perform data analysis and chart creation without modifying project source code, with generated files scoped to the conversation and refinable through follow-up prompts. The feature is available across all plan tiers, marking a significant expansion of Lovable beyond app building into general-purpose data workflows.

Lovable

Lovable: App Emails from Custom Domains

Lovable added support for sending transactional emails from custom domains, allowing apps built on the platform to deliver order confirmations, receipts, shipping updates, and security notices using branded sender addresses instead of generic Lovable domains. The platform handles all email authentication infrastructure — including DNS, SPF, DKIM, and DMARC configuration — automatically. The feature is available on paid plans for Lovable Cloud projects.

Lovable

Lovable: In-App Domain Purchasing

Lovable introduced the ability to search for, purchase, and connect custom domains directly within the platform, eliminating the need to use external registrars like Namecheap or GoDaddy. Powered by Stripe checkout, the feature automatically handles domain registration, DNS configuration, SSL setup, and WHOIS privacy for both root and www versions. Domains are managed at the workspace level, allowing them to be connected to different projects as needed. The feature is available on paid plans for Lovable Cloud projects.

Lovable