When Marketing Dreams Meet Reality's Receipts
Greg Brockman, OpenAI's cofounder and president, has a vision. He sees a future where we abandon software interfaces entirely. No more clicking, typing, or learning how applications work. Instead, invisible AI agents will handle everything while we sit back and.. well, presumably evolve into helpless digital infants.
This revelation came during a recent interview where Brockman cheerfully admitted that ChatGPT's plugins, which OpenAI marketed aggressively throughout 2023, failed spectacularly. "The models weren't ready," he confessed with the casual air of someone announcing they forgot to water the office plant.
I checked. Twice. These are the same plugins that were going to revolutionize how we interact with software. The ones that would make ChatGPT the ultimate productivity hub. The ones that would finally justify all those enterprise subscriptions.
But rather than reflect on why overpromising might be problematic, Brockman has pivoted to an even grander vision. He envisions AI agents so sophisticated that regular humans won't need to learn software anymore. The computer will simply understand what you want and make it happen.
This sounds delightful until you remember that OpenAI's current flagship product still hallucinates with the confidence of a politician during campaign season.
The Plugin Graveyard. A Brief Memorial
Let us pause to honor the fallen plugins. They arrived with fanfare in March 2023, promising to transform ChatGPT into a Swiss Army knife of digital productivity. Connect to Expedia! Integrate with Zapier! Order food through OpenTable! The future was here, and it had APIs.
The reality proved less inspiring. Users discovered that asking ChatGPT to book a restaurant often resulted in suggestions for establishments that had closed during the Carter administration. Flight searches returned options that existed only in the fevered dreams of travel algorithms. Shopping queries produced items that cost twice the listed price and arrived in colors that physics hadn't invented yet.
By December 2023, OpenAI quietly buried the plugin world in favor of "GPTs," which turned out to be plugins with a different name and identical problems. Filed under. things that were always going to happen.
"We're not building better tools. We're building a generation that won't know how tools work at all. This strikes me as roughly equivalent to teaching children that food appears when they press a button, without mentioning farms, cooking, or why the button occasionally produces rubber chickens instead of dinner."
Brockman's explanation for the plugin failure reveals the core issue. The models weren't sophisticated enough to handle the complexity of real-world software integration. This is rather like announcing plans for a Mars colony while admitting you haven't quite figured out how to keep the rockets from exploding on the launch pad.
The Interface-Free Fantasy
Undeterred by recent setbacks, Brockman now proposes eliminating interfaces altogether. In his vision, AI agents will be so contextually aware and capable that we'll simply speak our desires into existence. Need to edit a video? Tell the agent. Want to analyze data? Ask politely. Require a presentation for tomorrow's meeting? The AI will handle it while you sleep.
This represents a fascinating evolution in technological thinking. Rather than making software more intuitive or powerful, we'll simply hide it entirely. Users won't need to understand how applications work because the AI will serve as an invisible intermediary.
The appeal is obvious. Who wouldn't want a digital butler that handles all the tedious computer tasks? Learning software is genuinely difficult. Photoshop has more buttons than a nuclear submarine. Excel's formula syntax was apparently designed by sadists with computer science degrees.
But there's a rather substantial gap between Brockman's vision and current reality. OpenAI's Codex, which powers GitHub Copilot, still struggles with basic programming tasks. It can write simple functions and occasionally guess what you're trying to accomplish. But ask it to architect a complex system or debug subtle logic errors, and you'll quickly rediscover the value of human expertise.
If AI can't reliably write code that works, how exactly will it manage entire software workflows on our behalf?
The Expertise Apocalypse
Brockman's interface-free future raises uncomfortable questions about human capability. If nobody learns how software works, what happens when the AI agents inevitably malfunction?
Consider a parallel scenario. Imagine if GPS navigation became so prevalent that nobody learned to read maps or understand basic geography. Then imagine the navigation system fails during a critical journey. You're not just lost; you lack the fundamental skills to figure out where you are.
This is precisely what Brockman proposes for software literacy. Future generations won't understand how applications function, how data flows between systems, or why certain tasks are computationally expensive. They'll live in a world where technology works by magic, right up until it doesn't.
The implications extend beyond individual inconvenience. Entire industries depend on people who understand how systems interact. Software engineering, data analysis, digital marketing, graphic design. These fields require practitioners who comprehend the underlying tools, not just their outputs.
What happens to innovation when the next generation treats software as an incomprehensible black box? How do you improve systems you don't understand? How do you debug problems in tools you've never learned to use?
The Control Problem Nobody Mentions
There's another issue lurking beneath Brockman's vision. agency. When AI handles all your software interactions, you're no longer the primary user of your own computer. You've become a passenger in your own digital life.
This dependency creates fascinating power dynamics. The companies that control these AI agents will effectively mediate every interaction between humans and digital tools. Need to create a document? The AI decides which software to use and how to structure the content. Want to analyze data? The agent chooses the methodology and presents conclusions.
We're not just outsourcing tasks; we're outsourcing decisions about how those tasks should be accomplished. This represents a profound shift in human autonomy that Brockman presents as convenience.
Current evidence suggests this transition is already underway. Many users interact with ChatGPT as if it were infallible, accepting its responses without verification. They've begun treating a probabilistic text generator as an authoritative source of information. What happens when that same credulous trust extends to complex software operations?
Reality's Stubborn Persistence
The most amusing aspect of Brockman's prediction is its timing. OpenAI just admitted their previous interface revolution failed because the technology wasn't ready. Their current models struggle with basic reasoning tasks and frequently produce confident nonsense.
Yet Brockman confidently predicts a future where these same systems will be sophisticated enough to handle arbitrary software tasks without human oversight. This is rather like a chef burning toast and immediately announcing plans for a seven-course molecular gastronomy experience.
The technical challenges are substantial. Context-aware AI agents would need to understand not just what users want, but how different software applications work, how they integrate with each other, and how to handle the countless edge cases that arise in real-world computing.
They'd need to manage user authentication across dozens of platforms, handle software updates and compatibility changes, troubleshoot connection issues, and maintain security protocols. All while being transparent enough that users can verify the results but simple enough that those same users never need to learn the underlying systems.
This is approximately as realistic as teaching a goldfish to perform tax preparation.
The Enterprise Reality Check
Consider how Brockman's vision would work in practice. An executive wants to analyze quarterly sales data. Instead of opening a spreadsheet application and creating pivot tables, they'll ask their AI agent to handle it.
The agent needs to access the sales database, understand the company's data structure, identify relevant metrics, choose appropriate visualization methods, and present results in a format that supports decision-making. It also needs to handle security permissions, data privacy requirements, and integration with existing business intelligence tools.
This assumes the AI understands the executive's industry, their specific business context, and the nuances of their decision-making process. It must navigate corporate data governance policies and produce results that comply with regulatory requirements.
Meanwhile, the executive has no way to verify the analysis methodology or catch subtle errors in the AI's reasoning. They're making business decisions based on conclusions they can't audit, using data they can't access directly.
This represents either remarkable trust in AI capabilities or remarkable indifference to business risk. Possibly both.
The Innovation Dead End
Perhaps the most troubling aspect of Brockman's vision is its implications for technological progress. Innovation in software typically comes from users who understand the tools deeply enough to imagine improvements.
Photoshop's advanced features exist because professional designers pushed the software's boundaries and requested capabilities that would enhance their work. Programming languages evolve because developers identify inefficiencies and propose better approaches. Productivity software improves because users understand their workflows well enough to suggest optimizations.
If future users never learn how software works, who will drive innovation? How do you improve tools when your user base treats them as incomprehensible magic?
Brockman seems to assume that AI systems will handle innovation autonomously, continuously improving software without human guidance. This requires AI that not only understands current tools but can imagine better ones.
We're still waiting for AI that can reliably use current tools correctly.
The Middle Path That Nobody's Taking
There's an obvious alternative to Brockman's interface-free fantasy: better interfaces. Instead of hiding software complexity behind AI agents, we could design applications that are genuinely intuitive to use.
This approach would maintain human agency while reducing learning curves. Users would understand their tools well enough to verify results and suggest improvements, but wouldn't need computer science degrees to accomplish basic tasks.
Unfortunately, better interface design is harder than training AI models. It requires understanding human psychology, workflow optimization, and the messy realities of how people actually work. Much easier to build a chatbot and claim it will solve everything.
Naturally.
Conclusion: The Future We're Actually Getting
Brockman's prediction reveals more about Silicon Valley's worldview than about technology's future. Rather than empowering users with better tools, the industry increasingly prefers solutions that increase dependency on centralized AI systems.
This isn't driven by user needs or technological inevitability. It's driven by business models that profit from intermediating every digital interaction. Interface-free computing isn't a feature; it's a subscription service.
The real future will likely be messier and more mundane. AI will handle some routine tasks while humans continue managing complex ones. Software interfaces will gradually improve, but won't disappear. Most people will learn enough about their tools to accomplish their goals, but won't become experts.
We'll muddle through with a combination of human skill and AI assistance, rather than surrendering completely to invisible agents or rejecting artificial intelligence entirely.
But that future doesn't generate headlines or venture capital funding. So we'll continue getting predictions about interface-free utopias from executives whose last revolutionary prediction failed because "the models weren't ready."
I suppose the next revolutionary prediction will be ready any day now.
Everything is fine.
