🎤 Anthropic puts Claude in the interviewer's chair

Good morning, AI enthusiasts. What do workers really think about AI? Anthropic just asked 1,250 of them — and used Claude as the interviewer.

The company just launched a new tool for AI-powered qualitative research, and the findings from its first study paint a complicated picture: widespread adoption, hidden usage, and growing unease about the future of work.

Reminder: Our next workshop is today at 4 PM EST. Join and learn how to leverage Google’s new Nano Banana Pro to create polished slide decks. RSVP here.

In today’s AI rundown:

Anthropic puts Claude to work as a research interviewer
OpenAI trains models to ‘confess’ when they cheat
One-Shot a Landing Page with Replit Design Mode
Google, Replit deepen partnership for enterprise vibe coding
4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

📝 Anthropic puts Claude to work as a research interviewer

Image source: Anthropic

The Rundown: Anthropic just released Anthropic Interviewer, a Claude-powered research tool that conducts and analyzes qualitative interviews at scale — debuting with findings from 1,250 professionals on how they’re navigating AI in their work.

The details:

Anthropic Interviewer handles the full research pipeline: planning questions, running 10–15 minute conversations, and clustering themes for human analysts.
In the initial study, 86% of workers said AI saves them time, though 69% noted social stigma around using it and 55% voiced concern about their future.
Creatives reported hiding AI use from peers and concerns about job loss, while scientists said they want research partners but don’t fully trust models yet.
Anthropic is releasing all 1,250 transcripts publicly and plans to run ongoing studies tracking how the human-AI relationship evolves.

Why it matters: Companies typically learn about users through analytics dashboards or feedback forms, but open-ended interviews (scaled massively by Claude) can surface what people actually feel, not just what they click. The initial data shows a workforce adopting AI, but unsure of the social and future implications of it.

TOGETHER WITH POSTMAN

🗓️ Get your 90-day AI readiness guide

The Rundown: Turn your APIs into AI-ready tools. Whether you're setting the strategy or writing the code, Postman’s free 90-Day AI Readiness guide gives you a practical 30-60-90 day plan, so you can confidently build the foundation for intelligent automation.

Here’s the playbook:

0-30: Transform chaotic API docs into machine-readable standards
30-60: Build intelligent infrastructure that scales with AI automation requirements
60-90: Deploy AI agents that manage AI collaboration at scale

Get the playbook.

OPENAI

🤐 OpenAI trains models to ‘confess’ when they cheat

Image source: Nano Banana Pro / The Rundown

The Rundown: OpenAI just published new research on a technique called “Confessions” that trains models to produce a second, honesty-only output — where the model reports rule violations, shortcuts, or deceptive workarounds.

The details:

After generating a response, the model writes a separate confession report listing all instructions it received and whether it actually followed them.
Admissions carry no penalty, with the model earning ‘rewards’ for truthful self-reporting even if the original answer was misleading or gamed the grader.
In stress tests on GPT-5 Thinking, ‘false negative’ cases where the model broke rules and hid it occurred just 4.4% of the time.
OpenAI said the Confessions research does not prevent misaligned behavior, but helps surface it as another tool to leverage in a stack of AI safety methods.

Why it matters: Visibility into model behavior is improving, but the systems themselves are improving faster. Confessions give researchers a way to catch shortcuts and deception early, though the real test is whether interpretability can keep pace as systems grow more sophisticated and subsequently harder to test and control.

AI TRAINING

🧑‍🎨 One-Shot a Landing Page with Replit Design Mode

The Rundown: In this tutorial, you’ll learn how to create a unique, high-converting landing page for your business in under 5 minutes using Replit’s Design Mode with no coding required.

Step-by-step:

Use your preferred AI chat tool to generate a detailed landing page prompt using this formula: company name, target audience, offer, conversion goal, page sections, visual style, and CTA elements.
Find a landing page design you love on Google, Behance, or Dribbble, screenshot it, and drop it into your chat tool. Ask it to rewrite your prompt to match the style of the screenshot.
Sign into Replit, click the “Design” tab, and paste your prompt. Next, type “Generate a single-page landing page site based on the attached instructions. Generate assets as needed.”
Refine your page by updating social proof sections, embedding your CRM form, and deploying to a custom domain.

Pro tip: Deploy your Replit site to an existing domain’s subdomain for targeted campaigns like paid ads or outbound marketing.

PRESENTED BY SYNK

🔐 Navigate the new OWASP Top 10

The Rundown: The 2025 OWASP Top 10 has officially launched with new categories and shifting priorities that reflect today’s modern application landscape. Join Snyk’s live session on Dec. 11 at 11 AM ET to discover what’s changed and how it impacts your security strategy.

OWASP Leader Vandana Verma Sehgal will cover:

What’s new in the 2025 OWASP Top 10 and why it matters
How the updated priorities impact developers and AppSec teams
Strategies for adapting your security approach to the new landscape

GOOGLE & REPLIT

🤝 Google, Replit deepen partnership for enterprise vibe coding

Image source: Midjourney

The Rundown: Google and Replit just announced a multi-year expansion of its partnership with vibe-coding startup Replit, with the agreement focused on bringing the tooling to large enterprise partners running on Google infrastructure.

The details:

Replit will integrate Google’s recently released Gemini 3 into the platform, as well as its Imagen 4 text-to-image model for multimodal efforts.
The companies will market jointly through Google Cloud Marketplace, aiming to get Fortune 1000 teams building apps without dedicated engineers.
Replit’s annualized revenue jumped from under $3M to $150M in less than a year, with a September funding round valuing the startup at $3B.
The deal lands as rivals gain ground, with Claude Code crossing $1B in run-rate revenue and Cursor hitting the same milestone at a $29B valuation.

Why it matters: Vibe coding has been more of a solo-dev phenomenon so far — but Google’s deal is a bet that the workflows can scale for enterprises too. It’s also another example of the surging coding battle between frontier labs, with Google fighting Claude Code, OpenAI’s Codex, and others for a bigger slice of the developer workflow.

QUICK HITS

🛠️ Trending AI Tools

📲 Fastshot - Ship mobile apps to the App Store from ideas or Figma designs in minutes. No code, just by chatting with AI*
🍌 Nano Banana Pro - Google’s SOTA image-generation tool
🎥 Kling 2.6 - Kling’s new AI video model with native audio capabilities
🤖 4.5 Opus - Anthropic’s most powerful model, now available in Claude Code

*Sponsored Listing

📰 Everything else in AI today

Google rolled out Gemini 3 Deep Think to its Ultra ($250/mo) tier subscribers, the company’s most advanced reasoning model that hit gold-medal performances at the IMO and ICPC competitions for math and programming.

Microsoft open-sourced VibeVoice, a new small text-to-speech model that offers real-time streaming and long-form speech generation that can handle as long as 90 minutes of speaking and 4 distinct voices.

Snowflake and Anthropic announced a $200M multi-year partnership to deploy Claude-powered AI agents to Snowflake’s 12,600+ enterprise customers.

OpenAI announced plans to acquire Neptune, a startup that builds tools for tracking and analyzing AI model training during development.

Anthropic’s CEO seemingly took a jab at OpenAI CEO Sam Altman during an interview at NYT’s DealBook Summit, saying some AI companies may be overextending themselves with leaders who “just want to ‘YOLO’ things, or just like big numbers.”

AI legal startup Harvey raised $160M at an $8B valuation, while reporting that roughly half of the Am Law 100 firms now use the AI tool.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Dean F. in Atlanta, GA:

“I’m using AI to guide patients through at-home hemodialysis, and to help me build the technology behind it. I created an iPad app for my father’s dialysis that uses AI to speak customized step-by-step instructions, listen for voice commands, and adapt the flow in real time based on branching medical logic. It also uses patients’ own photos and videos to reinforce memory and reduce errors in a sterile environment.”

How do you use AI? Tell us here.