Skip to content
StackShip

Issue #1 ·

Haiku 5.5 gets cheap, Supabase drops Docker, agents buy domains

The first issue of New this week: the releases that matter if you build with AI tools, each linked to its primary source. This week: a much cheaper small Claude model, a new Gemini model for coding, Supabase running locally without Docker, and a run of agent-friendly updates from Vercel, Lovable and Cursor.

  1. model

    Gemini 4 Argon opens to developers on the paid Gemini API

    Google says developers can build with Gemini 4 Argon on the paid tier of the Gemini API in Google AI Studio and Android Studio, and Google AI Ultra subscribers can try it in AI Studio and Google Antigravity. Google describes it as designed for long-horizon software engineering work such as code migrations.

    Why it matters: A new flagship model option for coding agents, though API access requires the paid tier.

    Primary source ↗
  2. releaseGitHub Copilot

    Visual Studio 2026 October update makes the new Copilot chat the default

    Microsoft's Visual Studio Insiders release notes (released October 13, 2026) say the new Copilot chat experience, built on the Copilot SDK shared with the Copilot CLI, is now the default and aimed at longer coding tasks. The update also adds Copilot-powered Auto Resolve for Git merge conflicts and AI-generated Change Groups for reviewing large pull requests.

    Why it matters: Agent-style Copilot is now the out-of-the-box experience for Visual Studio users.

    Primary source ↗
  3. releaseLovable

    Lovable apps can expose WebMCP tools to browser agents

    Lovable's changelog says enabling agent integrations adds WebMCP tools that let a browser-based agent act on the page a person has open, such as marking a task done. Lovable starts with read-only or easily undone tools and adds riskier ones like deleting or publishing only when you ask.

    Why it matters: Apps built in Lovable can be made usable by AI browser agents without hand-writing the integration.

    Primary source ↗
  4. releaseVercel

    Vercel CLI lets agents search, price and start domain purchases

    Vercel added CLI commands (vercel domains search, check, price and buy) so agents can find and price domains and start a purchase, with the final purchase decision handed back to the user for confirmation. The CLI also ships an agent skill covering the discovery, pricing and purchase flow.

    Why it matters: A coding agent can take a project from deploy to a named domain with less manual dashboard work.

    Primary source ↗
  5. modelAnthropic API

    Anthropic releases Claude Haiku 5.5

    Anthropic released Claude Haiku 5.5 (model ID claude-haiku-5-5) at $0.10/$0.50 per million input/output tokens for prompts up to 100k tokens, and says it costs about 75% less on average than Haiku 4.5. Anthropic positions it for narrow tasks such as subagent work, summarization and compaction, and says Sonnet 5.5 and Opus 5.5 are better for complex agentic coding.

    Why it matters: It is a much cheaper option for high-volume or helper calls inside an app or agent pipeline.

    Primary source ↗
  6. releaseLovable

    Lovable adds workspace credit limits and Drizzle migrations for Cloud databases

    Lovable admins and owners can now set credit thresholds for workspaces, projects, members, groups or tokens, with alerts or blocks. The same day, Lovable began recording Cloud database schema changes as Drizzle migration files under drizzle/migrations/ and refusing changes that could break a published app unless permitted.

    Why it matters: Better spend control and a reviewable history of database changes made by the AI.

    Primary source ↗
  7. releaseCursor

    Cursor adds remote control of local agents from iOS

    Cursor's changelog says you can now view and reply to local agents running on your computer from the Cursor iOS app, after pairing the phone with the desktop app. Agents keep running on your computer, which must stay on and online, and the feature is on by default except for Enterprise organizations.

    Why it matters: You can unblock a long-running local agent session without being at your desk.

    Primary source ↗
  8. releaseSupabase

    Supabase scoped personal access tokens are generally available

    Supabase personal access tokens can now be restricted to specific projects or organizations with only the permissions they need across the Management API, MCP server and CLI. Access is fixed at creation and existing account-level tokens keep working until they expire or are deleted.

    Why it matters: You can give an AI agent or MCP client a token limited to one project instead of your whole account.

    Primary source ↗
  9. releaseSupabase

    Supabase local development can run without Docker (alpha)

    Supabase's Select 2026 recap says local Supabase can now run as native processes, so it starts on machines without a Docker daemon, and multiple local instances can run one per directory; it is in alpha and off by default. The same post also lists Multigres (private alpha), OrioleDB (public beta) and new MCP tooling.

    Why it matters: Supabase calls out sandboxed setups like Claude Code's as a case where running local Supabase without Docker helps.

    Primary source ↗

Get New this week by email

One short email a week: the releases that matter for AI builders, each with a primary source. Unsubscribe anytime.

We'll ask you to confirm by email. See our privacy policy.