💥 Wall Street Just Shaved $820 Billion Off AI Stocks — Here's What It Means for You

Hey, it's Oliver, here's your AI update for this week!

In today's issue:

• Why AI stocks just lost more money in a week than most countries make in a year
• OpenAI's new security agent finds bugs like a human researcher (and it's already caught 10 CVEs)
• Google's Gemini can finally read your emails and actually be useful
• And more...

🛠️ Tool of the Week: Perplexity's Comet Browser (Not Sponsored)

What it is: Perplexity just made their AI-powered browser completely free (it was $200/month for early access). Comet acts like a personal assistant that lives in your browser — it can search, organize tabs, draft emails, and shop for you.

Why it matters now: This launched free to everyone just last month (October 2025) after millions joined the waitlist. Unlike Chrome or Safari, Comet has AI baked into every action, not just tacked on.

Three ways to use it:

1. Research without the tab chaos – Ask Comet to research a topic and it automatically organizes findings across tabs, pulling together sources you'd spend an hour finding manually.
2. Email drafting that actually helps – Instead of opening Gmail and staring at a blank compose window, tell Comet what you want to say. It drafts the email in context while you keep working.
3. Shopping with less clicking – Tell Comet what you're looking for and it searches, compares prices, and can even complete purchases—all without opening 15 different tabs.

The catch: It’s new and still being refined. Some features work better than others, and it's only on desktop for now (mobile coming soon).

Pricing: Free. Seriously. They’re giving away what was a $200/month product.

Try it:Get Comet – You’ll need a Perplexity account (also free).

Bottom line: If you’re tired of juggling 47 browser tabs and want AI that actually helps instead of just answering questions, this is worth trying. It’s one of those “I didn’t know I needed this until I tried it” tools.

🛡️ 1. OpenAI Just Released an AI Security Agent That Hunts Bugs Like a Human Researcher

Last week, OpenAI quietly launched Aardvark, an AI agent powered by GPT-5 that finds and fixes security vulnerabilities in your code before hackers can exploit them.

Why This Matters

Aardvark reads your code, writes tests, and analyzes results like a human security researcher—not just scanning for obvious patterns. In testing, it caught 92% of known vulnerabilities and has already found 10 previously unknown security flaws (CVEs) in open-source projects. The real game-changer? It proposes actual patches you can implement with one click.

The Reality Check

It's currently invite-only and in private beta. OpenAI has been using it internally for months. Over 40,000 CVEs were reported in 2024, and their testing shows about 1.2% of all code commits introduce bugs. Companies like Google launched similar tools (CodeMender) this month, signaling automated security is becoming essential.

Bottom line: If you're writing code or running a dev team, this is AI that actually matters. Not flashy, but it could save your company from the next major breach.

📊 2. The $820 Billion Question: Is AI a Bubble or the Future?

In one week, AI stocks lost more than $820 billion in market value. That's more money than Saudi Arabia's GDP, gone in five trading days.

Why This Matters

Nvidia dropped 7% (still worth $5 trillion—8% of the entire S&P 500). Oracle fell 8.8% despite massive AI contracts. Palantir dropped 8% after beating earnings, crushed by a 700x price-to-earnings ratio. Investors are finally asking: when does this actually make money?

OpenAI generates billions in revenue but lost $13.5 billion in the first half of 2025. ChatGPT has 700 million weekly users but hemorrhages cash. Tech giants are spending $400 billion this year on AI infrastructure—a new Apollo program every 10 months.

The Practical Impact

Some call it a bubble like the dot-com era. Others point out these are profitable giants (Microsoft, Google, Amazon) with massive cash reserves—not startups burning VC money. Even Sam Altman admitted in August: "Are investors overexcited about AI? In my opinion, yes."

For regular people? This might be good news. Companies racing to justify their valuations means better, cheaper tools for you. The race for profitability could accelerate real innovation over hype.

🤖 3. Google's Gemini Deep Research Just Got Scarily Good at Reading Your Mind

Google's Gemini Deep Research can now access your Gmail, Google Drive, and Chat to create personalized reports that actually understand your context.

Why This Matters

Instead of just searching the web, Gemini now analyzes your team’s documents, email threads, and spreadsheets while combining it with real-time web research. It's what AI assistants were always supposed to be—tools that actually know what you’ve been working on. And it’s free for anyone with a Google account (desktop now, mobile coming soon).

The Privacy Question

You control which sources Gemini accesses—toggle Gmail, Drive, and Chat individually. The catch buried in fine print: "Human reviewers may review some of the data we collect." Real people might read your stuff to improve the AI.

If that bothers you, use Deep Research with just web sources. You'll lose the personalized magic, but your data stays private.

Bottom line: Google is betting you’ll find this useful enough to forgive them for reading your emails. And honestly? They might be right.

1. ✍️ Make AI Actually Remember What You Told It (No More Repeating Yourself)

The Real-World Moment: You've explained your work situation to ChatGPT three times this month. You've told Claude about your dietary restrictions twice. You've reminded Gemini about your learning style over and over. Every conversation starts from zero, like talking to someone with amnesia.

What This Solves: Both Claude and ChatGPT now have memory features that actually remember your context between conversations. But here's what nobody tells you: you have to explicitly tell them what to remember, and you have to do it strategically.

The 2-Minute Setup: Start your next conversation with any AI by saying: "Before we begin, here's what I need you to remember about me going forward..." Then list 5-10 key facts: your role, your goals, your constraints, your preferences, whatever matters most.

For example:

• "I'm a marketing manager for a B2B SaaS company"
• "I prefer direct communication without corporate jargon"
• "I work with a remote team across 3 time zones"
• "I'm particularly interested in AI applications for workflow automation"
• "I learn best through concrete examples, not abstract theory"

The Immediate Result: The AI will store this context and reference it automatically in future conversations. No more explaining yourself repeatedly. It’s the difference between talking to a stranger and talking to a colleague who actually knows you.

Pro Tip: Check your memory settings monthly and update them. Your priorities change, your role evolves, your goals shift. Keep the AI's understanding current.

💬 Copy-Paste Prompt to Try:
"Save this to memory: [Your 5-10 key facts]. From now on, reference this context in our conversations without me having to repeat it. Confirm you've saved this information."

🛠️ Tools Mentioned:
ChatGPT Memory – Available to Plus subscribers
Claude – Memory coming to all users
Gemini – Has profile settings that work similarly

2. 🔍 Stop Doing Research the Hard Way (Let AI Do the 4-Hour Deep Dive)

The Real-World Moment: You need to understand a complex topic for work—a new regulatory change, competitor strategy, or emerging technology. The old way? Open 47 browser tabs, read contradictory blog posts, give up after two hours and just guess.

What This Solves: Gemini's Deep Research and Perplexity’s Pro Search do actual multi-step research for you. They create research plans, follow leads, cross-reference sources, and synthesize findings like a human researcher would.

The 2-Minute Setup:

• Gemini Deep Research (free): Go to gemini.google.com → Click Tools menu → Select "Deep Research" → Choose your sources → Ask your complex question → Wait 5–10 minutes → Get a comprehensive report with citations.
• Perplexity Pro Search (free tier available): Go to perplexity.ai → Enable "Pro Search" → Ask your question (it’ll ask clarifying questions first) → Get sourced answers.

What Makes This Different: These aren't glorified search engines. They run actual research workflows: identifying sub-questions, searching multiple times, synthesizing findings. What you'd pay a junior analyst to do, except in minutes.

💬 Copy-Paste Prompts That Work:

For industry analysis:
"Analyze the current state of [your industry] focusing on: competitive landscape, emerging trends, key challenges, and opportunities for growth. Include specific examples and data points from the past 6 months."

For technology evaluation:
"Compare [technology/tool A] vs [technology/tool B] for [your specific use case]. Focus on: real-world performance, cost analysis, integration requirements, and long-term viability. Include recent case studies."

🛠️ Tools:Gemini Deep Research (free) | Perplexity Pro (free tier, $20/month unlimited)

Why This Matters Now: These are free because companies want market share. As profitability becomes essential, expect free tiers to shrink. Use them heavily now.

3. 💬 Build Your Own AI "Expert Panel" (Get Multiple Perspectives Instantly)

The Real-World Moment: You’re facing a tough decision at work and need multiple perspectives—your boss, a peer, someone from another department. But they’re all busy, and you need to think this through now, not in three days.

What This Solves: Use AI to simulate different expert perspectives on the same problem. It's like having a panel of advisors available 24/7, each with different expertise.

The 2-Minute Setup: Use any AI tool (ChatGPT, Claude, Gemini) with specific personas:
"I need to analyze [your decision/problem] from three different perspectives:

1. A risk-averse CFO focused on financial implications
2. A creative marketing director focused on growth opportunities
3. A technical product manager focused on implementation feasibility

For each perspective, provide: their main concerns, the questions they’d ask, and their recommendation.

The Result: You get three distinct viewpoints that challenge your assumptions and highlight blind spots. Not a replacement for real human advice, but remarkably good for thinking through different angles quickly.

💬 Copy-Paste Prompt Template:
"Act as three different advisors analyzing [your situation/decision]. For each advisor:

• Advisor 1 - The [Role/Expertise]:
  • Key concerns:
  • Critical questions:
  • Recommendation:
• Advisor 2 - The [Role/Expertise]:
  • Key concerns:
  • Critical questions:
  • Recommendation:
• Advisor 3 - The [Role/Expertise]:
  • Key concerns:
  • Critical questions:
  • Recommendation:

Then synthesize: what are the areas of agreement and disagreement? What decision would balance all three perspectives?"

Advanced Version: Ask the AI to roleplay a debate between these personas. "Now have these three advisors debate this decision. Have them challenge each other's assumptions and reasoning. After the debate, summarize the strongest arguments from each side."

🛠️ Tools:ChatGPT (best for personas) | Claude (nuanced perspectives) | Gemini (pulls from your work context)

Why This Works: Large language models are trained on vast amounts of expert reasoning. When you ask them to adopt specific perspectives, they can approximate how different experts would think about a problem. Not perfect, but way better than deciding in a vacuum.

A quick note before you go

Thanks for reading this week’s Brain Bytes — I hope something here helped you move faster or think better.

How’d this one land?

• A ✨ Super Helpful!
• B 🤔 Some Good Stuff!
• C 😕 Could be better.

See you next week, — Oliver