I have been thinking about this a lot lately, the public markets are about to get absolutely wrecked in slow motion.
Here's the thing: the people predicting when AGI shows up aren't just random Twitter doomers anymore. Sam Altman says end of decade. Jensen Huang says 2028. Kurzweil has been saying 2029 since before most of us were writing code.
If they are directionally right, we're about to watch the stock market re-sort itself based on one brutal question: When intelligence is basically free, what's actually worth building?
Remember Andreessen's "Software is eating the world" essay from 2011? Well, the food chain just changed. Intelligence is now eating software.
Let me break down how this plays out across public companies - because honestly, this is an important lens we use when we invest at Eight Capital.
Phase 1: Right Now to 2027 - Powerful AI
The Extinct: Wrapper Apocalypse
These companies are already dead. They just don't know it yet.
The profile: You built a UI on top of data or human labor that an LLM can now access directly.
Real examples:
- Chegg - Literally the canary in the coal mine. Stock cratered because ChatGPT explains calculus better than their homework answers ever did. They were a "wrapper" around answers. Now the model IS the answer.
- Teleperformance / TaskUs - They sell human hours for customer support. Voice AI agents are about to handle complex calls with empathy and zero hold time. Why do you need 50,000 people in a call center?
- Fiverr / Upwork - The supply side is getting automated. You used to pay $50 for someone to write a blog post. Now it's $0.01. Transaction volume goes to zero.
Founder Lesson #1: Don't Be a Wrapper
If your entire product is 'we made a nice UI for [thing LLM can do]' - you're already dead. Doesn't matter how good your UX is.
The test: Can someone get 80% of your value by just prompting Claude or ChatGPT? If yes, you're a wrapper.
What to do instead:
- Own proprietary data - Not public internet data. YOUR data that no one else has.
- Build workflow lock-in - Integrate so deeply into how companies operate that ripping you out breaks everything.
- Create network effects - Every new user makes the product better for existing users. AI can't replicate your user base.
- Go vertical, not horizontal - 'AI for X' loses to the operating system for a specific industry that happens to use AI.
Real talk: If you're building a 'ChatGPT for X' right now, you have maybe 18 months before OpenAI I or Anthropic just adds that feature. Either find your moat or pivot.
Phase 2: 2028 to 2030 - AGI Arrives
The Compressed: Race to Zero
These companies survive but become commodities.
When AGI shows up - meaning software that can genuinely act as an autonomous employee - differentiation in generic B2B SaaS just evaporates.
The profile: Single-feature tools and "middle-office" software.
Real examples:
- Docusign - Digital signature is a feature, not a company. AGI agents will handle verification and signing as one step in a larger workflow. Why pay $30/month for a feature?
- Zoom - Video transport is a commodity. When AGI can attend meetings FOR you and give you a perfect summary, the value shifts to whoever hosts the intelligence (Microsoft, Google). Standalone tools get squeezed.
- Dropbox / Box - Storage is a commodity. The premium was for smart features. When AGI makes all your data instantly searchable and organized, who cares where it lives?
Founder Lesson #2: Features Get Eaten, Platforms Survive
The entire SaaS playbook of ‘do one thing really well’ is about to get inverted. Single-feature tools become line items in someone else's platform.
The test: Is your product a verb or a noun? ‘Sign this document’ is a verb or feature. ‘Salesforce’ is a noun or a system of record.
What to do instead:
- Become the System of Record - Own the data that other tools depend on. If you're the source of truth, you survive.
- Build the platform, not the app - Let others build on top of you. Stripe isn't a payment button; it's financial infrastructure.
- Create switching costs - Years of historical data, integrations with 50 other tools, trained workflows. Make leaving painful.
- Bundle aggressively - If you are a single feature, you will get commoditized. If you're 10 features that work together seamlessly, you're a platform.
Real talk: Look at your product roadmap. If you're adding "AI features" to a single-purpose tool, you're playing defense. The question isn't ‘how do we add AI?’ - it's ‘how do we become essential infrastructure that AI plugs into?’
The Survivors: Distribution Moats
These companies take margin hits but persist.
The profile: Systems of Record and massive distribution.
The logic: AGI automates the WORK inside these systems, but the systems themselves stick around because of switching costs and data gravity. They're already installed on every Fortune 500 laptop.
Real examples:
- Salesforce - They own the customer data record. AGI will write the emails and update the CRM, but no one's ripping out Salesforce. It's the source of truth.
- Microsoft - The ultimate distribution moat. Copilot is baked into Word, Excel, Windows. They own the OS where AGI lives. They capture all the value that standalone apps lose.
- Adobe - Creatives are locked into the file formats and workflows. Adobe survives by embedding AI (Firefly) directly, preventing the escape to standalone generators.
Founder Lesson #3: Distribution > Product - Now More Than Ever
In the AGI era, the best product doesn't win. The most distributed product wins. Microsoft isn't the best at anything - they're just EVERYWHERE.
The test: If a startup built your exact product with better AI tomorrow, would your customers switch? If yes, you have a product. If no, you have distribution.
What to do instead:
- Get embedded early - Be the default choice before the market gets crowded. First-mover advantage matters more when switching costs are high.
- Own the workflow, not the task - Zoom is a task. Microsoft Teams is a workflow (chat + calls + files + calendar). Workflows are stickier.
- Build for the enterprise - Consumer apps get disrupted easily. Enterprise apps with 3-year contracts and 50 integrations don't.
- Create data gravity - The more data you store, the harder it is to leave. Salesforce doesn't have the best product; they have 10 years of your customer data.
Real talk: Stop optimizing for Product Hunt launches. Start optimizing for how I become the default tool that IT installs on Day 1?’ Boring distribution beats sexy products every time.
Phase 3: 2030+ Superintelligence
The Thrivers: This Is the Counterintuitive Part
When superintelligence shows up, digital cognition becomes commoditized. Value flees digital and accumulates in companies dealing with real-world constraints.
Companies that control the physical manifestation of AI or assume legal liability for its actions? They become the new kings.
Infrastructure & Energy -The Food for AI:
- NVIDIA / TSMC - The shovel sellers. Superintelligence needs infinite compute. Until physics changes, these monopolies print money.
- NextEra Energy, Inc. / Dominion Energy - Data centers are power-hungry af. The constraint on AI scaling in the 2030s won't be code - it'll be electricity. Utilities with green capacity become critical tech infrastructure.
Robotics & Physical World -The Body of AI:
- Tesla - Not a car company. A robotics company. Superintelligence needs a body. Optimus is the physical labor force that AGI will power.
- John Deere - Autonomous agriculture. AGI applied to farming requires actual heavy machinery and land. Deere owns the fleet that AGI will drive.
Vertical & Liability - The Trust Layer:
- Palantir Technologies - They operate where "hallucination" is not an option - war, defense, government. They provide the liability layer that lets institutions use superintelligence without everything collapsing.
- UnitedHealth Group / HCA Healthcare - Healthcare requires navigating regulation, liability, and physical care. AGI can diagnose, but it can't legally treat or insure. These giants will use AI to crush costs while keeping the premium for actual care delivery.
- Eli Lilly and Company / Novo Nordisk - Bio-acceleration. Superintelligence will simulate biology and invent drugs. But someone still needs to own the labs, run the clinical trials, and navigate the FDA. That's where the value lands.
Founder Lesson #4: Regulation Is a Feature, Not a Bug
Everyone complains about regulation slowing things down. In the AGI era, regulation is your moat.
The test: Would your business be easier if there were no regulations? If yes, congratulations - you just described exactly why you'll get disrupted.
What to do instead:
- Embrace compliance - HIPAA, SOC2, FDA, FINRA. Every regulation is a barrier to entry that AI alone can't cross.
- Build the compliance layer - Someone needs to help companies use AI safely. Audit trails, governance, permissions. Boring? Yes. Essential? Absolutely.
- Target regulated industries - Healthcare, finance, government, defense, legal. AI accelerates the work, but humans and their lawyers still sign off.
- Assume liability - If you're willing to be legally responsible for AI outcomes, you become the trust layer. That's valuable.
Real talk: The YC companies that will matter in 2030 aren't building 'AI for X.' They're building 'the legally compliant, auditable, insured way to deploy AI in X. Less sexy, way more durable.
What's the biggest hole in this framework?
I have been thinking about this, but I am sure I am missing something. Would love founders and investors to pressure-test it in the comments
