In partnership with
Hello from The AI Night,
Today in AI:
Anthropic Redesigns Claude Code Desktop App for Parallel Coding
OpenAI Launches GPT-5.4-Cyber With Fewer Guardrails
Microsoft Launches Copilot Track Changes in Word
Here's the deal: Anthropic released a major redesign of the Claude Code desktop app, built around running multiple coding sessions simultaneously. The update adds a session sidebar, drag-and-drop workspace layouts, an integrated terminal, an in-app file editor, and a rebuilt diff viewer.
The Breakdown:
New sidebar lets developers manage parallel sessions across multiple repos, with filtering by status, project, or environment and automatic archiving when PRs merge or close.
Side chat feature (⌘ + ;) lets you branch off questions mid-task without polluting the main session context.
Integrated terminal and file editor mean you can run tests, make edits, and review diffs without switching to an external editor.
Desktop app now has full parity with CLI plugins, and SSH support extends to Mac alongside Linux.
Three view modes (Verbose, Normal, Summary) let you control how much of Claude's tool-call activity you see.
Available now on Pro, Max, Team, Enterprise plans, and via the API.
The bigger picture: Cursor and Windsurf already have multi-session interfaces. Claude Code was still a single terminal window. That gap was the main reason developers kept Claude Code for heavy tasks but lived inside competitor IDEs for daily work. This update removes the biggest reason to switch away mid-workflow. Anthropic is no longer competing on model quality alone. They are fighting for the screen developers stare at eight hours a day.
Here's the deal: OpenAI is scaling its Trusted Access for Cyber (TAC) program to thousands of verified individual defenders and hundreds of security teams. Alongside this expansion, it's releasing GPT-5.4-Cyber, a variant of GPT-5.4 fine-tuned with lowered refusal boundaries for legitimate cybersecurity work.
The Breakdown:
GPT-5.4-Cyber enables capabilities like binary reverse engineering, letting security pros analyze compiled software for malware and vulnerabilities without source code access.
Access is tiered: individuals verify identity at chatgpt.com/cyber; enterprises go through OpenAI reps. Higher tiers require additional authentication as legitimate defenders.
More permissive models may come with restrictions, particularly around zero-data-retention (ZDR) and third-party platform use where OpenAI has less visibility.
Codex Security, now six months post-beta, has contributed to over 3,000 critical and high fixed vulnerabilities across the ecosystem.
OpenAI classified GPT-5.4 as "high" cyber capability under its Preparedness Framework.
The bigger picture: Every serious security researcher has hit a moment where ChatGPT refused to help analyze malware they were actively defending against. That refusal didn't stop attackers who already have their own tools. It just slowed down the defenders. OpenAI is finally admitting that blanket safety restrictions were punishing the wrong side. The real test is whether "verified defender" stays meaningful at scale or becomes another checkbox that erodes the whole point.
Here's the deal: Microsoft introduced new Copilot in Word capabilities designed for professionals in legal, finance, and compliance who need full document control. The update lets Copilot edit documents with track changes enabled, manage comment threads, and handle structural formatting, all within Word's native collaboration layer.
The Breakdown:
Track Changes now works at word-level precision. Copilot edits are visible and auditable by default, with an easy toggle to turn tracking on.
Copilot can add, reply to, and manage contextual comments anchored to specific text in the document.
New formatting tools include table of contents generation, headers, footers, page numbers, margins, and dynamic fields that auto-refresh.
Progress messages now show what Copilot is doing during multi-step edits in real time.
All features operate within Microsoft 365's trust boundary, preserving sensitivity labels and enforcing data loss prevention policies.
Available today on Word for Windows desktop via the Frontier program (Office Insiders Beta Channel). Mac and web support coming soon.
The bigger picture: Few days ago Anthropic launched Claude for Word with tracked changes and style preservation. Now Microsoft is matching that exact feature set inside its own product. That should worry Anthropic. Microsoft doesn't need to be better here. It just needs to be good enough, because Copilot is already installed on every enterprise machine. Anthropic had a window to win legal and compliance teams before Microsoft caught up. That window just got smaller.
88% resolved. 22% stayed loyal. What went wrong?
That's the AI paradox hiding in your CX stack. Tickets close. Customers leave. And most teams don't see it coming because they're measuring the wrong things.
Efficiency metrics look great on paper. Handle time down. Containment rate up. But customer loyalty? That's a different story — and it's one your current dashboards probably aren't telling you.
Gladly's 2026 Customer Expectations Report surveyed thousands of real consumers to find out exactly where AI-powered service breaks trust, and what separates the platforms that drive retention from the ones that quietly erode it.
If you're architecting the CX stack, this is the data you need to build it right. Not just fast. Not just cheap. Built to last.
What else you need to know:
Mintlify raised a $45 million Series B led by Andreessen Horowitz and Salesforce Ventures at a $500 million valuation, as its developer documentation platform now reaches 100 million people yearly with AI agents driving nearly half the traffic.
Google launched Gemini 3.1 Flash TTS, a text-to-speech model with new audio tags for granular voice control, multi-speaker dialogue, 70+ language support, and an Elo score of 1,211 on Artificial Analysis.
Anthropic flagged "alien science" as a future risk, warning that automated researchers may eventually produce ideas too complex for humans to verify or detect if corrupted.
Cursor's multi-agent system, collaborating with NVIDIA, autonomously optimized 235 CUDA kernels for Blackwell GPUs over three weeks, achieving a 38% geomean speedup with 19% of solutions exceeding 2x improvements.
SpaceX is now using a Grok-powered voice AI assistant to handle Starlink customer support calls in real time, with similar deployments reportedly active at Tesla.
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.
3 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
Reply to this email — I read every response