Cursor Joins SpaceX, Builder Tools Shift
Your AI app can now pull each user’s own work data, while agent stacks get easier to test, tune, and move between tools. Here’s what to try before Monday.
In today’s edition
- Lovable adds per-user connections to tools like Google Drive, HubSpot, and Slack.
- Cursor has closed its acquisition by SpaceX and will join its AI team.
- OpenRouter publishes data on search quality, cost, and search-turn budgets for agents.
- Replit ships region selection, broader MCP access, and a path from Replit Auth to Clerk.
Lovable apps can connect to each user’s data
Apps built on Lovable can now let each person connect their own accounts, including Google Drive, HubSpot, Slack, and more.
- Each user authenticates their own accounts instead of sharing one builder-level connection.
- Users see their own data inside the app.
- The update turns internal-data app ideas into more plausible multi-user products.
What it means Try a customer-facing workflow that needs personal context: a sales brief built from a rep’s HubSpot and Slack, a client portal that searches the client’s Drive, or an operations assistant tied to each teammate’s tools. The key product question is no longer just “Can the app connect?” It is “What useful job can it do with each user’s private context?”
Source: Lovable
Cursor officially becomes part of SpaceX
Cursor says its acquisition has closed and that it will join the SpaceXAI team.
- The team says it will work on making Grok more useful.
- Its remit also includes Grok Build, Grok Bot, the Grok API, Cursor, and more.
- Cursor is framing the deal as an expansion across a wider AI product stack.
What it means If Cursor is part of your development workflow, watch for product changes and closer links with the Grok ecosystem. More broadly, the acquisition puts coding tools, agent products, and model APIs under one roof. Keep your coding-agent workflow portable: document prompts, repo conventions, permissions, and evaluation checks so you can compare tools without rebuilding your process from scratch.
Source: cursor_ai
OpenRouter benchmarks agent web search stacks
Choosing web search for an agent is now a configuration problem, not a brand-loyalty problem.
- OpenRouter launched benchmark pages that rank search tools across models and configurations.
- Its tests span four benchmarks, four models, four search depths, and Exa, Parallel, Perplexity, and native model search.
- Increasing the search budget from 1 to 25 turns roughly doubled BrowseComp scores.
- At a 25-turn budget, correct runs averaged 10.3 searches and incorrect runs averaged 19.7.
What it means Set a search-turn budget based on the value of a correct answer and the cost of being wrong. For high-stakes research workflows, test deeper search. For broad, failure-prone tasks, cap turns to prevent expensive dead ends. Also test the model and search engine as a pair: OpenRouter found model swaps had a larger average score effect than engine swaps.
Source: OpenRouter
Replit adds regions, MCP upgrades, and Clerk migration
Replit shipped a stack of updates aimed at making AI-built apps easier to operate across locations and toolchains.
- Pro and Enterprise users can choose North America, Europe, or Asia for workspace hosting.
- Replit MCP can now find and update existing projects, with OAuth and no API key required.
- The agent can migrate an existing Replit-auth app to Clerk Auth while moving account data first.
- New MCP servers include Airtable and Rootly Docs, while an admin API beta supports account, usage, and spend data.
What it means If you have an existing Replit app, test the Clerk migration on a non-production copy before your next auth change. If your team builds from several AI clients, connect Replit through MCP and see whether project updates fit your existing workflow. For teams with regional data needs, workspace location is now a deployment decision worth reviewing early, not after launch.
Source: Replit