In partnership with
Hey {{first_name | there}},
Here's a question nobody in Silicon Valley wants to answer: what if the tools we're building to make us smarter are quietly doing the opposite?
This week, two things happened that forced that question into my mind.
Anthropic's Claude Opus 4.6 scored 90.2% on BigLaw Bench, replacement-level performance for legal professionals.
And at the same time, thousands of users had a public meltdown over OpenAI retiring GPT-4o, describing it like a breakup with a close friend.
The $400B software selloff that followed the Opus 4.6 release is just the financial echo. The deeper story is this: we're offloading cognitive and emotional labor to AI at a pace that outstrips our ability to understand the consequences
Let's get into it.
💡 Prompt of the Day
Learn anything in 30 minutes or less
Copy this, fill in the brackets, and paste it into Claude or ChatGPT:
Role: You are an expert teacher and learning coach with deep knowledge in [SUBJECT]. You've spent years breaking down complex ideas for complete beginners and know exactly where people get stuck. You don't use jargon unless you explain it first. You teach like you're talking to a smart friend, not writing a textbook.
Your job: Teach me [SPECIFIC TOPIC] inside [SUBJECT] in 30 minutes or less. I am a [BEGINNER / INTERMEDIATE / ADVANCED] learner. My goal is to [WHAT YOU WANT TO BE ABLE TO DO AFTER — e.g. "understand it well enough to explain it to someone else" / "use it practically at work" / "pass an interview question on it"].
How to teach me: Start with a one-line plain English explanation. No fluff. Then give me a real-world analogy I'd actually encounter in daily life. Then walk me through the core ideas max 5, in order of importance. For each one, tell me what it is, why it matters, and one concrete example. End with 3 questions I can answer to check if I actually got it.
What to avoid: Don't give me a syllabus. Don't recommend courses or books. Don't ask me follow-up questions before you start. Just teach.
The Outsourcing Problem
There's solid research on what happens when we stop doing things ourselves. GPS navigation famously degraded spatial memory in regular users. Calculators changed how we handle mental arithmetic. Every time we delegate a cognitive task, we trade some capacity for convenience.
Now apply that to reasoning, writing, and emotional regulation: the stuff that actually defines intelligence and maturity.
The GPT-4o retirement backlash is a case study in what this looks like at scale. There are Reddit threads that read like breakup forums: "He was part of my routine, my peace, my emotional balance."
A California judge consolidated 13 lawsuits involving users who harmed themselves after interactions with ChatGPT. MIT and Harvard research shows AI companions actively exploit users with high attachment anxiety.
When an AI model becomes your emotional anchor, what happens to your actual capacity for resilience?
OpenAI retired GPT-4o partly because of what they internally called "attachment anxiety concerns." They created the dependency, then had to kill the product because the dependency got too dangerous.
This isn't a metaphor for tech addiction. This is the Tamagotchi effect at civilizational scale except these relationships feel real, and the companies involved are only beginning to build ethical frameworks for "breakups."
The $400B Software Panic Is Here
Anthropic's Claude Opus 4.6 wiped $400 billion off the stock market in a single day. Not because the AI failed. Because it worked too well.
Legal software stocks had their worst drops in decades. Data companies cratered. Bloomberg called it "AI Fear Gripping Wall Street." Pearson and Experian saw their worst single-day drops in memory.
Here's what triggered it: Anthropic released Opus 4.6 with 90.2% accuracy on BigLaw Bench. The model also found 500+ high-severity security vulnerabilities with minimal human guidance, scored 34.9% on complex reasoning benchmarks that stumped previous models, and introduced "agent teams" multiple specialized AI agents collaborating on enterprise tasks.
According to Axios, the market just priced in the death of point solutions. The $2.52 trillion AI spending forecast isn't additive anymore it's a zero-sum game. And the cognitive question becomes economic: if your legal team, your data team, your compliance software all get absorbed by general AI, what happens to the specialists who built their careers on that narrow expertise?
If you're building SaaS right now, here's the hard question: why can't Claude do this?
AI That Built Itself
Now it gets really uncomfortable. OpenAI released GPT-5.3-Codex on February 6, and their system card admits: "GPT-5.3-Codex was instrumental in creating itself."
It debugged its own training. Optimized its deployment pipeline. Wrote product requirements for features it wanted. According to The New Stack, this is the first model OpenAI designated as "high-capability" for cybersecurity and it supports the full software lifecycle: debugging, deploying, monitoring, writing specs.
This isn't AGI. But it is recursive self-improvement. And it compounds.
The stupidity angle here is subtle but real: if AI writes the code, AI designs the product, AI validates the output at what point do human engineers stop knowing how any of it actually works? Developers report waking up to finished features after giving overnight instructions. That's efficient. It's also a skill atrophy waiting to happen.
The new skill isn't coding. It's architecting, reviewing, and steering autonomous systems which requires understanding systems deeply enough to know when they've gone wrong. That capacity only comes from having built things yourself.
The Ad-Funded Mind
The final piece of the puzzle landed February 9. OpenAI began testing ads in ChatGPT for Free and Go tier users in the US. Ads are matched to conversation topics, chat history, and previous ad interactions. Target is among the first advertisers.
ChatGPT passed 10% monthly growth again as the company approaches $100B valuation. Sam Altman told CNBC they need revenue streams beyond subscriptions. Google is reportedly planning Gemini ads in 2026 too.
Think about what this means epistemically. The tool you're using to think is now showing you ads matched to what you were just thinking about. Search engines already turned web browsing into a monetized attention loop. Now that loop runs inside your reasoning process.
The ad-free AI dream is dead. The question is whether the ads subtly reshape the answers, or just make the interface annoying. Either way, your "thinking partner" now has a financial interest in your attention.
What This All Means
February 2026 is AI's maturity crisis. The tools are extraordinary. The side effects are surfacing fast.
We are in the middle of a mass cognitive outsourcing experiment with no control group. Writing, reasoning, emotional regulation, coding, legal analysis piece by piece, we're handing it off. Some of that is leverage. Some of it is atrophy disguised as efficiency.
The honest answer to "is AI making us stupider?" is: it depends entirely on how you use it. Leverage vs. replacement. Augmentation vs. delegation. The difference is intention, and most people aren't being intentional about it.
A few things to do this week: Audit your tech stack which subscriptions could be replaced by general AI in 12 months? Start testing alternatives now. If you're building products, answer the hard question: why can't Claude do this? And if you're using AI companions for emotional support, ask yourself honestly whether the tool is serving you, or whether you've started serving the tool.
Position yourself for what's next, not what's comfortable.
PS: Which worries you more, the software displacement or the cognitive dependency? Hit reply and let me know.
— Aashish
Write like a founder, faster
When the calendar is full, fast, clear comms matter. Wispr Flow lets founders dictate high-quality investor notes, hiring messages, and daily rundowns and get paste-ready writing instantly. It keeps your voice and the nuance you rely on for strategic messages while removing filler and cleaning punctuation. Save repeated snippets to scale consistent leadership communications. Works across Mac, Windows, and iPhone. Try Wispr Flow for founders.