Welcome, humans.

We try to find a few good memes every week to share with you on Mondays, but this one could NOT wait…

This is of course referring to Monday’s news about ChatGPT 4.1, the absolutely outrageously titled name for OpenAI’s new “Quasar model” that got previewed early last week. Why not just keep the codenames, fam? Quasar is a 1000x better name!

Now, the Verge is reporting OpenAI is making its own social network. It’s still early days (that’s Love Island lingo for “prototype stage”), but it’ll work like this: image generator + social feed (separate from the new image library it just rolled out).

Kinda makes sense to capitalize on ChatGPT’s most viral feature, since other platforms run by its competitors are currently sucking up all the engagement from that.

Meanwhile, Google just made its own viral video tool, Veo 2, available in Gemini. Google never succeeded in launching its own social network—maybe now’s the time! Just have Gemini 2.5 vibe code something up, y’all!

Think about it: Make Myspace 2.0, but instead of everyone having to learn how to code to customize their page, just build a vibe coder into the page manager so ppl can spin up their own custom pages. This will work people. Don’t make us do it for you.

Here’s what you need to know about AI today:

• AI code tool Cursor let its AI run wild and almost tanked its user-base.

• OpenAI has a new image library and ~800M weekly active users.

• Perplexity tied Google in search benchmark.

• NVIDIA and Stanford developed cartoon generator.

When “vibe-coding” goes wrong… or, a parable in why you shouldn’t “vibe” your entire company.

Cursor, an AI-powered coding tool that many developers love-to-hate, face-planted spectacularly yesterday when its own AI support bot went off-script and fabricated a company policy, leading to a complete user revolt.

Here’s the short version: 

1. A bug locked Cursor users out when switching devices.

2. Instead of human help, Cursor’s AI support bot confidently told users this was a new policy (it wasn’t).

3. No human checked the replies—big mistake.

4. The fake news spread, and devs canceled subscriptions en masse.

5. A Reddit thread about it got mysteriously nuked, fueling suspicion.

The reality? Just a bug, plus a bot hallucination… doing maximum damage.

Hacker News’ reactions were gold, capturing the irony of an AI company getting burned by its own AI (“These bros are getting high on their own supply”), the “hate-use” relationship many have with buggy-but-useful tools (like Cursor), and suspicion of an AI cover-up.

But this comment summed up the situation perfectly:

Why it matters: This is what we’d call “vibe-companying”—blindly trusting AI with critical functions without human oversight.

Think about it like this: this was JUST a startup. If more big corporations continue to lay off entire departments, replaced by AI, these already byzantine companies will become increasingly more opaque, unaccountable systems where no one, human or AI, fully understands what’s happening or who's responsible.

Our take? Kafka dude has it right. We need to pay attention to WHAT we’re actually automating. Because automating more bureaucracy at scale, with agents we increasingly don’t understand or don’t double check, can potentially make companies less intelligent—and harder to fix when things inevitably go wrong.

Now, Cursor's cofounder apologized, explained the bug, and promised clearer AI labeling. But this warning is clear enough: automating without understanding or oversight is a recipe for disaster. Keep humans in the loop, or risk your AI turning your company into a zombie tornado. What’s a zombie tornado? We’re so glad you asked!

P.S: Are you a developer interested in Cursor alternatives? Some users mentioned switching to tools like Windsurf, Cline, or Zed.

FROM OUR PARTNERS

AI agents for every team

Sana helps you reduce time to value with one enterprise-grade platform that lets you build no-code AI agents integrated with all your company's apps.

Imagine working with AI that can:

• Find the exact info you need

• Answer and brainstorm any idea

• Recap meeting actions and to-dos

• Automate repetitive tasks

• Solve complex, multi-step problems in seconds.

Extensible APIs. Enterprise-grade security. With unparalleled flexibility.

Book an intro today

Prompt Tip of the Day

A Redditor just shared OpenAI’s latest Prompt Guide for GPT 4.1 and it goes like this:

1. Role: Define what the AI is (“You are a helpful research assistant”).

2. Instructions: Set behavior and tone (“Respond concisely, avoid speculation”).

3. Sub-Instructions: Add focused control sections (prohibited topics, phrases).

4. Reasoning: Encourage structured thinking (“Think step-by-step”).

5. Format: Specify response structure (“Summary: [1-2 lines], Key Points: [bullets]”).

6. Examples: Show sample inputs/outputs.

7. Final: Reinforce key points at the end.

Treats To Try.

1. Tableau Pulse delivers personalized data insights directly in your workflow, automatically detecting trends and explaining why metrics change without requiring you to build visualizations.

2. Notion Mail automatically sorts emails, creates custom views, and drafts replies in your style—free to try with Gmail.

3. Extrovert puts all your LinkedIn prospects' posts in one feed and helps you write personalized comments that sound like you wrote them—free 10-day trial, then $35/month.

4. Windsurf is offering free unlimited GPT-4.1 for a week, then at a discounted rate of $0.25 credits per use.

5. n8nChat helps you create, edit, and debug n8n workflows via chat, generating complete automations and custom code so you can automate anything.

6. DeepCoder is a new 14B parameter open model achieving top-tier coding performance with enhanced GRPO and 64K context generalization. It's available on OpenRouter for those wanting a free, specialized coding assistant.

7. This is cool—tech blogger Lenny Rachitsky is offering 10 free premium tools (Perplexity Pro, Notion, the dreaded Cursor, & more) with his annual newsletter subscription—$200 a year, super good deal.

See our top 51 AI Tools for Business here!

Around the Horn.

• OpenAI now has somewhere between 800M-1B weekly active users after the launch of its image generator.

• DeepSeek announced plans to open-source parts of its inference engine by porting optimizations to popular frameworks like vLLM, llama.cpp, and kobold, rather than releasing the full stack (that’s nerd for coders get cool free tools).

• Perplexity's Sonar-Reasoning-Pro-High model tied with Gemini-2.5-Pro-Grounding for first place in LM Arena's Search Arena, scoring 1136 and 1142 respectively.

• Google reportedly nerfed Gemini 2.5 Pro's tool calling function, rendering it unable to execute tool calls—probably due to cost concerns—but its still useful for UI design and data processing tasks.

• Samsung announced a partnership with Google to power its Ballie home robot with Google's Gemini and Samsung's own multimodal AI models.

• NVIDIA and Stanford researchers unveiled a new AI technique to generate consistent, minute-long cartoons.

FROM OUR PARTNERS

🛠️ Unstructured API: Full ETL+ Power, Now Programmatic

The Unstructured API delivers full feature parity with the UI, so you can connect to data sources, partition unstructured documents, enrich them, generate vector embeddings, and deliver structured data. Entirely in code.

Ready to level up your unstructured data workflows? Try it today!

Mid-Week Wisdom

We love writing our Intelligent Insights Friday section…and tbh, we wish we could run the section every day (instead of just once a week). So we’re going to trial “mid-week wisdom.” If you like getting Intelligent Insights 2x a week, let us know in the feedback!

• Tech recruiters are reporting a hiring crisis as candidates use AI tools like Interview Coder to fake technical skills, forcing companies to revive in-person interviews despite the $2K in travel costs to fly someone out (which could lead to more local hires).

• LinkedIn is leading a major organizational shift where product teams are restructured around products rather than roles, encouraging “full-stack builders” who use AI to work beyond their traditional job descriptions—designers now code, engineers design, and everyone focuses on business impact regardless of title. This is the future, y’all.

• Gavin Leech explains why he barely uses AI despite being an AI PhD—they make confident errors that kill his appetite, he prefers writing himself, needs precision that AI lacks, and worries about being deskilled by relying on them.

• Ed Zitron (famous AI skeptic) wrote a great argument about OpenAI being a “systemic risk to the tech industry” because it burns $14B+ annually with unsustainable finances, depends on SoftBank's shaky $40B funding commitment, faces capacity constraints despite begging for GPUs, relies on unproven startups for data centers, and if it collapses, will trigger an industry-wide “contagion” affecting Microsoft, CoreWeave, Oracle, and NVIDIA.

A Cat's Commentary.