Sponsored by
Hello from The AI Night,
Today in AI:
xAI Launches Grok Build Beta for Premium Users
Anthropic Launches Security Guidance Plugin For Claude Code
Cognition Hits $492M Revenue, Raises $1B Round
Here's the deal: xAI expanded its Grok Build coding agent to all SuperGrok and X Premium+ subscribers, after an initial release limited to SuperGrok Heavy users. Grok Build is a terminal-native CLI for software engineering that adds Plan Mode, image and video generation via Imagine, and scripting tools for automations.
The Breakdown:
Plan Mode drafts a structured plan before execution; users can approve, comment on, or rewrite steps, with changes shown as clean diffs.
Imagine generates images and videos inline through slash commands.
Headless mode via the -p flag lets agents run inside scripts and automations.
Full ACP support enables building custom bots and agent orchestration apps.
Larger tasks delegate to parallel subagents, each with its own context window and optional Git worktree.
AGENTS.md, plugins, hooks, skills, and MCP servers work without setup.
Installs with a single command from the terminal.
The bigger picture: xAI is borrowing the Claude Code and Codex CLI playbook: plan-then-execute, parallel subagents, MCP support, AGENTS.md conventions. The real differentiator is distribution. Bundling Grok Build into X Premium+ puts a coding agent in front of millions who never signed up for a developer tool, a reach Anthropic and OpenAI cannot match through standalone subscriptions.
Here's the deal: Anthropic shipped a security-guidance plugin for Claude Code that flags and fixes vulnerabilities as you write. It is available to all Claude Code users and installs from the plugin marketplace via the /plugins command.
The Breakdown:
The plugin runs through hooks and reviews code at three stages of development.
On file edits, it scans for risky patterns such as commonly misused dangerous libraries.
After each model turn, it reviews the full diff to catch harder-to-spot issues.
On commits, it reads surrounding code to validate flagged vulnerabilities.
Teams can add org-specific rules in a claude-security-guidance.md file, dropped in a repo or distributed via MDM.
Across internal rollout and benchmarks, Anthropic reports a 30 to 40% drop in security-related comments on PRs opened with the plugin.
The bigger picture: Fixing a security flaw during writing costs almost nothing. Fixing it during review costs ten times more. Fixing it in production costs everything. Anthropic moved the checkpoint to the cheapest moment. Thirty to forty percent fewer security PR comments is not a productivity stat. It is proof that problems are dying before reviewers ever see them.
Here's the deal: Cognition has raised over $1B in a Series D at a $26B valuation, led by Lux Capital, General Catalyst, and 8VC. The company behind Devin reported $492M in run-rate revenue, with enterprise usage up more than 10x since the start of 2026.
The Breakdown:
Round led by Lux, General Catalyst, and 8VC, with new investors Ribbit Capital, Atreides, and Layer Global.
Customers include Citi, Goldman Sachs, Mercedes-Benz, Dell, Santander, the US Army, and US Navy.
Mercedes-Benz cut an eight-month legacy modernization project down to eight days with Devin.
Itaú, Latin America's largest bank, fixes 70% of security vulnerabilities automatically using Devin.
SWE-1.6, Cognition's in-house model, is now the most used model in Windsurf, running up to 950 tok/s.
At Cognition, 89% of code committed by engineers is now written by Devin.
The bigger picture: Eight months to eight days. That Mercedes-Benz number will close more enterprise deals than any benchmark ever could. Cognition writing 89% of its own code with Devin is the ultimate product demo. Half a billion in revenue at a $26B valuation means Wall Street is not pricing a coding tool. It is pricing a labor replacement. Every firm selling developer hours just got a comparison it cannot win.
In a World of AI Agents: Intent > Identity
AI-powered bots aren’t just logging in anymore. They’re mimicking real users, slipping past identity checks, and scaling attacks faster than ever.
Thousands of companies worldwide trust hCaptcha to protect their online services from automated threats while preserving user privacy.
Now is the time to take control of your security.
What else you need to know:
NVIDIA introduced Dynamo Snapshot, a checkpoint/restore system for Kubernetes inference workloads that cuts gpt-oss-120b cold-start time by up to 21x, restoring a fully warm worker in under 5 seconds.
Sesame launched an iOS preview of its voice conversation agents in 39 countries, adding characters Simone and Charlie alongside Maya and Miles, plus search cards, notes, and Incognito mode.
Google launched Ask YouTube, a conversational search experience that compiles relevant videos across YouTube's catalogue and returns interactive, structured responses instead of standard recommendations, currently available to U.S. Premium members.
FastVideo open-sourced Dreamverse, a real-time video generation workspace based on LTX-2 and optimized for a single NVIDIA B200 GPU, releasing the full frontend, backend runtime, and Blackwell-optimized inference path.
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.
2 ways to support us:
Forward this to your AI-curious friend → https://www.theainight.com
Sponsor The AI Night and reach 500+ AI builders daily → passionfroot.me/theainight