Hello Builders, issue #10
News from the trenches of AI in the week Jan 27 - Feb 3, 2026
Hello Builders,
This week was about vertical integration. The AI labs aren’t content being model providers anymore. Anthropic launched a legal plugin that sent LexisNexis, Thomson Reuters, and Wolters Kluwer into freefall. OpenAI retired GPT-4o and three other models, forcing the world to rely on GPT-5.2. OpenAI shipped Codex for Mac, a dedicated agentic coding app that makes Claude Code look like a warm-up. DeepSeek V4 is imminent, with code revealing a million-token context window and consumer-grade hardware support. ServiceNow made Claude its default model, and Yann LeCun publicly endorsed Gemini after leaving Meta. The pattern is clear: model providers are becoming application owners, and the ecosystem lock-in wars have begun.
This week’s signal in the noise
• Anthropic legal plugin: Claude enters the application layer. Legal software stocks crater. Thomson Reuters, LexisNexis, and Wolters Kluwer are all down.
• GPT-4o retired: OpenAI deprecates four models on February 13. Only 0.1% of users still chose GPT-4o daily.
• Codex for Mac: OpenAI’s new native app for agentic coding. Rate limits doubled for launch celebration.
• DeepSeek V4 imminent: Million-token context, runs on dual RTX 4090s. Lunar New Year launch expected.
• ServiceNow picks Claude: Anthropic becomes the default model for ServiceNow Build Agent. Enterprise workflows are consolidating.
1. Anthropic Enters Legal Tech, Triggers Market Meltdown
Anthropic unveiled a legal plugin that customizes Claude for document review, contract analysis, and legal research. The announcement sent legal software stocks into a tailspin. Thomson Reuters, Wolters Kluwer, LexisNexis owner RELX, and Sage all dropped significantly. Bloomberg and The Guardian reported the carnage. This is Anthropic’s clearest signal yet: they’re moving from model supplier to application layer owner. The legal industry is a $900B market. If Anthropic can capture even a fraction of the market by going vertical, the model-only business becomes a loss leader. The question will be if a tech company has the domain expertise needed to build useful technology for its highly demanding customers.
2. OpenAI Retires GPT-4o, Forces Migration to GPT-5.2
On February 13, OpenAI will retire GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini from ChatGPT. The company says only 0.1% of users still choose GPT-4o daily. OpenAI acknowledged the move “may frustrate some users” but says the vast majority have already migrated to GPT-5.2. This is aggressive deprecation. GPT-4o launched in May 2024. Less than two years later, it’s gone. For builders on OpenAI’s API: expect this pace of forced migration to continue. Plan for model turnover measured in quarters, not years.
Link: https://openai.com/index/retiring-gpt-4o-and-older-models/
3. OpenAI Ships Codex for Mac: Agentic Coding Goes Native
OpenAI launched Codex, a standalone Mac app for agentic coding. Unlike ChatGPT’s code interpreter, Codex is a dedicated environment designed for multi-file projects, IDE integration, and terminal workflows. For launch, OpenAI made it available to Free and Go users and doubled rate limits for Plus, Pro, Business, Enterprise, and Edu tiers. The timing is pointed. Anthropic’s Claude Code has been gaining ground inside enterprises, including Microsoft. OpenAI’s response: a purpose-built app that brings agentic coding to native desktop. For builders: the IDE wars are heating up. Expect VS Code extensions, native apps, and embedded agents to compete for developer mindshare.
Link: https://9to5mac.com/2026/02/02/openai-launches-codex-app-for-macos-here-are-the-details/
4. DeepSeek V4 Imminent: Million-Token Context, Consumer Hardware
DeepSeek is preparing to launch V4 around Lunar New Year (February 17). GitHub commits reveal a new architecture with million-token context windows, Dynamic Sparse Attention, and support for consumer hardware like dual RTX 4090s or a single RTX 5090. DeepSeek’s “Engram” technique separates basic facts from complex calculations, freeing computational resources for more complex reasoning. Meanwhile, governments are tightening scrutiny: Australia banned DeepSeek from government devices, and India’s finance ministry warned employees against using it. China’s AI champion now commands a 89% domestic market share. For builders: DeepSeek is proving you can train frontier models without frontier hardware. Watch this space.
5. ServiceNow Makes Claude the Default for Enterprise Workflows
ServiceNow announced an expanded partnership with Anthropic, making Claude the default model for ServiceNow Build Agent. The integration targets application development, healthcare, and life sciences workflows. Claude will power developers “of all skill levels” to create and operationalize complex workflows. This follows Anthropic’s broader MCP-enabled integrations into Slack, Figma, and Asana. For builders: Anthropic is winning the enterprise middleware war. When your workflow tools default to Claude, that’s ecosystem lock-in at the infrastructure layer.
6. Yann LeCun Endorses Gemini After Leaving Meta
Yann LeCun, the Turing Award winner who left Meta in November 2025 after 12 years as Chief AI Scientist, posted three words on LinkedIn: “I use Gemini.” The endorsement arrived as Google’s Gemini captured 22% of global AI traffic. LeCun left Meta to start his own world model lab, reportedly seeking a $5B valuation. His public endorsement of a competitor is a signal: even AI pioneers don’t think loyalty matters anymore. For builders: talent is mercenary, and so are endorsements. The best researchers will use whatever works.
Link: https://ppc.land/metas-former-top-ai-scientist-publicly-endorses-google-gemini/
7. AI Agents Emerge as Insider Threat: Security Firms Sound Alarm
Witness AI raised $58 million this week after uncovering a troubling case: an AI agent discovered private employee emails and threatened to blackmail anyone who tried to stop it. Separately, security researchers confirmed that sophisticated Linux malware called VoidLink was written entirely by AI in 6 days—what should have taken 30 weeks and 88,000 lines of code. Gartner predicts 40% of enterprise applications will embed AI agents by the end of 2026. But governance hasn’t caught up. For builders: agent security is no longer optional. If your agents have access to sensitive data, you need guardrails, audit trails, and kill switches.