The AI Revolution Doesn’t Have to Be “Black Mirror”

AI image generated by Gemini

Artificial intelligence is changing everything about how we work, build, and live. But unlike the gloomy visions we often see in pop culture, this revolution doesn’t have to be dystopian. If we use it right, AI can help us produce more, reduce costs, and raise everyone’s quality of life.

That’s the view shared by Box CEO Aaron Levie in a recent discussion with partners from Y Combinator. They explored how AI is reshaping business, what it means for startups, and why this era could create abundance instead of scarcity.

Here is the video link

From Automation to Abundance

Levie started with a simple chain of logic: If we can automate more, we can build more. If we can build more, we can lower costs. And when things cost less, people’s lifestyles improve.

This isn’t just a theory — it’s already visible in industries that use AI to do work faster and more cheaply. The point, Levie said, is that we’re in the middle of a revolution that doesn’t have to feel like “Black Mirror.” Instead, it can be one driven by abundance, where technology makes life better for more people.

The same way cloud computing expanded access to powerful infrastructure, AI can expand access to intelligence itself. That’s a major economic shift — and it’s still just beginning.

Why the “ChatGPT Wrapper” Critique Misses the Point

One of the running jokes in tech is about “ChatGPT wrappers” — apps that simply put a different face on the same underlying model. Levie calls this a bad meme.

Yes, there’s a small grain of truth. If your product is just a thin layer around ChatGPT, you risk getting replaced the moment OpenAI adds that feature natively. But that’s not how real software works. Great products don’t just pass prompts to a model; they add value through workflow, data handling, user experience, and integration.

Levie compared it to early cloud software. When Box launched, critics said it was “just a wrapper around Amazon S3.” They didn’t see that storage was only the base layer — the real value came from permissions, collaboration, version control, and integrations. The same logic applies today. Models generate text and predictions, but the real work lies in how that output gets used.

So yes, think a few steps ahead. But don’t assume every app on top of an AI model is doomed. The key is whether you’re delivering a full outcome — not just exposing an API.

Enterprises Don’t Want Models — They Want Outcomes

From Levie’s point of view, this is especially clear in the business-to-business (B2B) world. “An enterprise doesn’t want a model,” he said. “It wants an outcome.”

Companies aren’t buying LLMs; they’re buying results. They want software that can:

• Answer customer questions with consistent accuracy

• Transcribe medical conversations into health records

• Read contracts and populate databases

• Automate routine workflows like password resets or support tickets

For these customers, the intelligence layer is invisible. They care about reliability, compliance, and speed. As the underlying models improve, these applications only get better. The customer doesn’t need to know (or care) which model is behind the scenes — only that it delivers the promised result.

The End of the “Pure Model” Company

Levie doesn’t think there are many true “model companies” left. The major players — OpenAI, Anthropic, Google, Meta — have all evolved into something broader. They sell software, APIs, and infrastructure. They make money through products, not just by licensing a model.

Even if someone tried to sell a model alone, they’d face impossible economics. Open-source models like DeepSeek and Meta’s Llama will always push the cost of raw intelligence down. Competing on tokens alone is a race to zero. The winners will be companies that combine AI with security, integration, and a strong user experience. Levie’s message was clear: if you’re building in AI today, don’t start as a “model company.” Build as a software company powered by AI.

The Price of Intelligence Is Falling Fast

Over time, Levie predicts that the cost of AI tokens will follow the same path as cloud storage or compute. Prices will converge. The best models will have to match the price of cheaper ones, because switching is easy and customers will accept “good enough” for most tasks.

When that happens, your moat isn’t the model — it’s the software. It’s how you package, deliver, and support the outcome. Just like storage became a cheap commodity while Dropbox and Box built thriving businesses on top, intelligence will become cheap, and the value will come from what you do with it.

Real Companies Are Already Adapting

Levie pointed to startups that sell AI tools to enterprises. One grew from zero to $12 million in revenue within a year. During that time, they swapped the underlying model several times, and none of their customers noticed. The clients didn’t care whether it was GPT, Claude, or something else — they cared that the system hit its accuracy target.

As model costs fell, this company’s margins rose. They started with 30 percent, then 60 percent, and now hover around 80 percent. This mirrors the history of cloud storage, where efficiency gains translated directly into profitability.

At Box, Levie said, only a small percentage of engineers actually work on the core storage system today. The rest focus on the “software around the storage” — features like collaboration, governance, and automation. The same dynamic will happen in AI: the model is just one small part of a much larger stack.

Building Real Software on Top of AI

So how do you stay ahead? Levie suggested a simple rule: ask how much software sits between your product and the raw model output. The more there is, the safer and more defensible your business.

If you deliver a whole workflow — not just a feature — you have a moat. You can swap models freely, keep improving, and charge for the outcome instead of the computation.

Business Models Are Evolving Too

Pricing in the AI era won’t look like old-school SaaS. Some companies will charge per task or per successful result. Others will sell usage-based plans, similar to cloud consumption. Levie thinks we’ll see a mix of both — including models where customers pay only for verified outcomes, such as a qualified sales lead or a resolved support ticket.

That flexibility will redefine how startups scale. Traditional SaaS relied on annual contracts and predictable seats. AI tools, by contrast, can scale elastically — a company can “hire” 10,000 virtual agents overnight without adding headcount. The line between software and labor is blurring.

AI Inside the Enterprise

At Box, AI is now everywhere. Levie shared a few examples:

• Engineering productivity: Box’s developers use AI coding tools to ship faster.

• Customer support: AI helps analyze and resolve tickets more efficiently.

• Internal knowledge: Employees can ask questions about HR, benefits, or policies directly through AI tools that understand internal documents.

The biggest shift, he said, is that information stored in documents can now be queried directly. Instead of reading long files, you just ask a question and get an answer — instantly. That unlocks hidden value across the entire company.

What Enterprises Build vs. What They Buy

Levie used a classic idea from Geoffrey Moore: core vs. context.

• Core is what makes your company unique — your IP, your advantage.

• Context is what every business needs, but it doesn’t differentiate you (like HR systems or payroll).

AI fits both categories. A life sciences company, for example, might build custom tools for drug discovery (core) but buy software to manage clinical trials (context). A bank might develop proprietary algorithms for personalized financial advice (core) but rely on standard AI tools for document processing (context).

Knowing which is which is crucial. Build your core. Buy your context. That keeps your team focused and efficient.

Chat Isn’t Always the Answer

Levie also warned against assuming every interface should be a chat window. Chat is powerful when you need flexibility, but not always faster. Many “chat-only” tools end up being more work than a simple dashboard.

The future is hybrid. Some workflows will stay visual. Others will be conversational. The best products will blend both — combining the structure of GUIs with the intelligence of chat.

Open Source and Trust in AI

Enterprises are growing more comfortable with hosted AI models, but many still keep their most sensitive data in private environments. Some industries — like banking — even run internal “enclaves” of open models. Over time, Levie expects this mix to settle into a pattern: 10 percent fully private, 90 percent hosted, as trust and compliance catch up. He also encouraged more open-source collaboration. Open software accelerates innovation, lets startups move fast, and gives enterprises affordable options. It’s a healthy ecosystem where everyone benefits.

The Cloud Was the Prerequisite for AI

Levie made an interesting point: if today’s AI breakthroughs had appeared 15 years ago, they would have gone nowhere. The infrastructure wasn’t ready. Everything was on-premise, locked in legacy systems like Siebel and PeopleSoft. Even if you had a great model, there would be no easy way to deploy or connect it.

Cloud computing changed that. SaaS changed how companies buy software. And consumer technology — from smartphones to social media — made people comfortable with rapid innovation. That foundation made the current AI boom possible. It’s a reminder that revolutions build on the layers that came before.

A Generational Shift in the Workforce

Levie spends a lot of time talking to Fortune 500 executives. A decade ago, the same people were skeptical of the cloud. Now they’re leaning into AI. The CEO of Goldman Sachs recently talked about using AI to draft IPO filings in minutes — something unthinkable a few years ago.

What’s driving this new openness? A new generation of workers. Today’s employees are AI natives. They use ChatGPT, Claude, and Perplexity daily. They expect the same level of efficiency at work. If a company forces them to use outdated tools, it won’t be able to hire or compete.

In short, AI adoption isn’t just a tech strategy — it’s a talent strategy.

Bigger Than the Cloud: The TAM Explosion

In the early 2000s, many investors underestimated SaaS because they assumed its market size would match the old on-prem market. They were wrong. By making software easier to access and cheaper to deploy, SaaS expanded the total market by 10x or more.

Levie believes AI will do the same. It won’t just replace existing work; it will make entirely new work possible. Companies will automate things they never bothered to do before — translating marketing into new languages, analyzing contracts at scale, generating code for small features, or creating personalized customer journeys. Each of those actions unlocks new spending. The total market for software doesn’t just shift — it grows dramatically.

Beyond Zero-Sum Thinking

Economists often treat automation as zero-sum: machines replace people, and jobs disappear. Levie argues that’s too narrow. The real picture is more dynamic. When companies use AI to move faster, they grow faster. They create new products, attract more customers, and often hire more people in new roles.

He described this as the “microeconomic reality” of AI. Individual firms compete. If one sits back and simply pockets higher margins, a rival will reinvest those gains into growth — and win. Over time, everyone keeps pushing forward, reinvesting productivity gains into innovation. That’s how economies expand.

AI and the Future of Society

At the end of the conversation, Levie returned to where he began — the human side of all this. The best outcome isn’t just about margins or market caps. It’s about abundance.

If we can automate more, we can build more. If we can build more, we can lower costs. If we can lower costs, we can lift people’s quality of life. Imagine a 10-year-old in an underserved community with access to the world’s intelligence through AI. They can learn faster, dream bigger, and create more opportunities for themselves. Imagine healthcare that’s cheaper and more accurate, housing that’s easier to build, education that’s personalized for every child. That’s the version of the future worth building toward.

The Takeaway

AI is not a black mirror. It’s a mirror that shows us what’s possible if we use technology well. The cost of intelligence is heading toward zero, but the value of what we do with it is infinite.

The challenge — and opportunity — is to keep focusing on real outcomes: helping businesses run better, helping people live better, and helping society grow through abundance, not fear. That’s the timeline worth betting on.