Hello from The AI Night,
Today in AI:
Anthropic Launches Claude Computer Use to Control Your Desktop
Tesla, SpaceX and xAI Unveil $25B TERAFAB Chip Factory
Cursor Speeds Up AI Agent Search in Large Codebases
Here's the deal: Anthropic launches Claude Computer use that lets Claude directly use your Mac to point, click and navigate apps on your screen. The feature works inside Claude Cowork and Claude Code and pairs with Dispatch to let you assign tasks from your phone.
The Breakdown:
Claude tries connected integrations first (Slack, Google Calendar), then falls back to controlling your browser, mouse, keyboard and screen when no connector exists.
The system scans model activations automatically to detect prompt injection, and Claude asks explicit permission before accessing new applications.
Some apps are blocked by default. Anthropic recommends avoiding sensitive data while the feature is in early stages.
Available to Pro and Max subscribers, macOS only. Desktop app must be running.
Dispatch enables recurring automations; daily email scans, weekly report pulls, or spinning up Cowork/Code sessions on schedule.
Complex tasks may need a second attempt and screen-based control is slower than direct integrations.
The bigger picture: This moves Claude from a chat-based assistant to one that can operate across your desktop environment. For builders and founders, it turns repetitive multi-app workflows into assignable tasks, even when direct API integrations do not exist.
Tesla x SpaceX
Tesla, SpaceX and xAI Unveil $25B TERAFAB Chip Factory
Here's the deal: Elon Musk announced plans to build two chip factories under one facility in Austin, Texas. One fab will produce chips for Tesla vehicles and Optimus humanoid robots. The other will make chips designed for AI satellites in space built to handle harsher environments and higher temperatures.
The Breakdown:
SpaceX's involvement was previously undisclosed; SpaceX recently merged with Musk's xAI and social media platform
Target output: one terawatt of computing capacity per year, roughly double current total U.S. chip output
Musk framed the project as existential: current global production would cover only a fraction of his companies' future needs
No timeline was given. Musk has a track record of delays on ambitious projects
He acknowledged existing suppliers (Samsung, TSMC, Micron) but said demand will eventually exceed total global supply
The bigger picture: This would be one of the largest vertical integration plays in AI hardware, spanning cars, robots and orbital compute. The space chip component opens a new hardware category entirely. Execution and timeline remain open questions.
Here's the deal: Cursor shipped a new indexing system that speeds up regular expression searches for AI agents working in large monorepos. Instead of scanning every file with ripgrep, the system builds local indexes on the developer's machine that narrow down which files could match before doing a full scan.
The Breakdown:
The core technique uses sparse n-grams, variable-length character segments weighted by frequency, to build an inverted index of source code
Posting lists are stored on disk with a memory-mapped lookup table, keeping editor memory usage minimal
Index freshness is maintained by anchoring to Git commits, with user and agent changes layered on top
Indexes run client-side to avoid network roundtrips, latency costs, and data privacy concerns around syncing codebases to servers
The approach draws on decades of research, from 1993 trigram indexes through GitHub's Project Blackbird and suffix arrays
The bigger picture: For enterprise teams with massive repositories, grep latency is one of the few agent operations that scales with codebase size. Removing that bottleneck means faster iteration loops and fewer wasted tokens when agents search for specific text patterns.
Here’s how I use Attio to run my day.
Attio is the AI CRM with conversational AI built directly into your workspace. Every morning, Ask Attio handles my prep:
Surfaces insights from calls and conversations across my entire CRM
Update records and create tasks without manual entry
Answers questions about deals, accounts, and customer signals that used to take hours to find
All in seconds. No searching, no switching tabs, no manual updates.
Ready to scale faster?
What else you need to know:
Anthropic added Projects to Claude Cowork, letting users organize tasks into persistent local workspaces with dedicated files, instructions and project-scoped memory inside the Claude Desktop app.
Anthropic added scheduled, recurring cloud-based tasks to Claude Code letting users automate workflows like sweeping PRs, building features from issues and syncing docs without running the tool locally.
Cursor launched Composer 2, a coding model scoring 61.3 on CursorBench (up from 44.2) priced at $0.50/$2.50 per million input/output tokens with a faster variant available.
Luma launched Uni-1, a unified model that reasons and generates pixels simultaneously, ranking first in human preference Elo for overall quality, style, editing and reference-based generation.
Anthropic detailed a three-agent architecture (planner, generator, evaluator) inspired by GANs that produced full stack applications over multi-hour autonomous coding sessions, solving self evaluation and context anxiety issues.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
If this was useful, you’ll get the same signal here tomorrow.





