Here’s a $10 trillion question: can AI overtake most computer-based work by 2027?
Miles Brundage, who used to lead AGI Readiness at OpenAI, made a bold statement: by 2027, computers will do almost every money-making task done on computers better and cheaper than humans. This got a lot of attention.
But there’s more to the story.
Brundage added some important notes to his claim. He said he meant AI could “technically” do these tasks; not that it would be used everywhere by then. Also, he was talking about tasks where you can judge the result without needing human judgment or emotion.
So, what’s really possible by 2027? Let’s looks at how far AI has come, what it can do today, and what problems still stand in the way.
Parsing the Prediction: What Brundage Really Meant
Brundage’s meaning is more cautious and complicated when we take a closer look. His follow-up comments make clear that what sounds like a bold prediction is really more of a technical possibility than a real-world certainty.
Doable Doesn’t Mean Deployed
Brundage explained that his claim is about what AI could do—not what it will be doing everywhere. “Will be done,” he said, really means “will be doable.” In other words, AI might be capable of performing most computer-based tasks by 2027, but that doesn’t mean businesses will actually be using AI for all of them.
Why? Because technical ability doesn’t guarantee real-world use. Many things slow down wide adoption, like:
- Cost of building and maintaining AI systems
- Trust and safety concerns
- Problems with infrastructure, especially for small businesses
- Legal and ethical issues, like fairness and bias
So while AI might be able to do most tasks on paper, putting that ability into practice across the economy is another matter.
Effectiveness in a Vacuum
Brundage also gave a specific definition for “more effective.”
He said it means a task is done better when you only judge the output, not the full process. This means ignoring the value of a task being done by a trusted human or by someone with emotional understanding.
For example, an AI can write a code snippet or a short email quickly. But it can’t build trust in a client meeting or offer empathy to an upset customer. Many tasks that seem simple actually involve human connection or judgment. When AI’s output is judged “in isolation,” those deeper values are left out.
What Counts as an “Economically Valuable Task”?
The prediction focuses on tasks that are done on computers and help make money or save costs
This includes things like:
- Writing small pieces of software code
- Cleaning and analyzing data
- Creating marketing content or reports
- Managing emails or chat conversations
Some of these jobs are repetitive and follow clear rules. These are the easiest for AI to take on. But others require complex thinking, creativity, or deep understanding of context—areas where AI still struggles.
Economists and researchers use different models to sort tasks by how easily they can be automated. Generally, tasks that are rule-based and don’t require human judgment are easier for AI to take over. On the other hand, tasks that depend on people skills or real-world knowledge are much harder.
There’s also the economic side to consider. Even if AI can do a task, it may not be cost-effective.
For a small bakery, buying and setting up an AI system to inspect ingredients might cost more than just training a worker. For bigger businesses, the math might work out better.
In other words, what’s “valuable” or “affordable” depends on who’s using the AI.
Measuring “More Effectively and Cheaply”
To test Brundage’s prediction, we need ways to measure “better and cheaper.”
For “better,” researchers use benchmarks that test how well AI performs tasks like solving math problems or writing code. For “cheaper,” they look at cost savings from using AI instead of paying people.
Here’s what matters:
- Performance: AI often scores high on specific tests, like coding challenges or academic quizzes. But real-world tasks are messier, and success depends on things like reliability, security, and the ability to handle surprises.
- Cost: Running AI models is getting cheaper fast. For example, the cost of using a model like GPT-3.5 dropped by over 280x in just two years. But companies also have to pay for setup, maintenance, energy, and fixing mistakes.
So even if an AI system performs a task well in a lab, using it in the real world still involves many hidden costs—and risks.
Reality Check: Where AI Wins Today and Where It Still Struggles
AI has come a long way. It’s faster, smarter, and cheaper than ever in many areas.
Despite this progress, major roadblocks still stand in the way of AI taking over “almost every” computer-based job by 2027. These challenges fall into three main categories: technical limits, deployment problems, and readiness gaps in society.
AI Still Struggles With Complex Thinking
Right now, AI does well with simple, rule-based tasks. It’s good at writing short code, answering common questions, or summarizing reports. But when the task gets harder, the cracks begin to show.
For example, AI has trouble with complex reasoning. Many tasks—like planning a business strategy, managing a big project, or designing a new product—require thinking through many steps, making judgments, and solving new problems.
Current AI tools often can’t handle this. They get confused, make mistakes, or take shortcuts that don’t make sense.
AI also lacks common sense. AI may know a lot of facts, but it doesn’t understand the world like people do. It often misses obvious points or makes silly assumptions that a human would never make.
Finally, AI doesn’t plan well over time. Many jobs involve setting long-term goals and adjusting as things change. AI tends to focus on short-term outputs. It isn’t good at thinking ahead or dealing with surprise situations.
Some researchers believe these are problems that can be fixed with more training and bigger models. But others think we may need new kinds of AI altogether—something closer to how the human brain works. Either way, it’s a long road, and 2027 is coming fast.
Reliability and Trust Are Ongoing Problems
Another big issue is that AI still makes too many mistakes and often can’t explain why.
- Hallucinations: AI sometimes invents facts. It might give an answer that sounds right but is totally wrong. This is a huge problem in areas like healthcare, law, or finance where errors can have serious consequences.
- Lack of explainability: Many AI systems are “black boxes.” Even their creators don’t fully understand how they reach decisions. That makes it hard to trust the results or hold anyone accountable when something goes wrong.
- Not great at learning new things: Most AI tools don’t learn as they go. They’re trained on old data and struggle when faced with new or unfamiliar situations—something humans deal with every day.
Even when AI gets 80–90% of a task right, the last 10–20% is often the most important. This “last mile” usually involves tricky judgment calls, creative thinking, or ethical choices. That’s where humans are still essential.
Scaling AI Is Harder Than It Looks
Even if AI is technically able to do a job, getting it to work in the real world isn’t easy or cheap. For one, energy demands are rising.
Training and running large AI models takes a lot of power. Data centers use enormous amounts of electricity. If AI is going to take over most computer tasks, it could drive up energy costs and increase carbon emissions.
Besides it’s energy demand, AI needs massive, high-quality datasets to learn from. But many industries don’t have that kind of data available. If the data is biased, incomplete, or low-quality, the AI will make poor decisions. And preparing that data takes time and money.
Finally, deploying AI is complicated.
Most businesses still struggle to plug AI into their daily work. They may have old systems that don’t connect well with new AI tools. They also need experts, time, and money to train staff, test systems, and manage the change.
In fact, studies show that around 80% of AI projects fail to deliver on their goals.
This means that even if AI could do a task more cheaply in theory, the total cost of making it work may still be higher; especially for small businesses.
Society Isn’t Fully Ready—Ethically or Practically
Beyond tech problems, there are serious human and ethical concerns. Even if AI can do the job, many people won’t accept it unless it’s safe, fair, and well-regulated.
- Trust is fragile: Many people worry about AI making up facts, spreading bias, or replacing human workers. If companies rush to use AI without proper safety checks, public trust could collapse.
- Bias and fairness: If AI is trained on biased data, it can make unfair decisions—like favoring one group over another in hiring or loans. That’s a legal, social, and ethical risk.
- Lack of rules: Laws haven’t caught up with the speed of AI development. Governments are working on it, but policies are still patchy. Without clear rules, companies may hesitate to use AI for important work.
Even Miles Brundage warned that society isn’t ready.
He listed five key “AGI readiness gaps”—from weak regulations to lack of public input. One of his biggest concerns is that AI capabilities are racing ahead faster than our ability to control or guide them. He compared it to building a rocket engine without knowing how to steer it.
If AI keeps advancing without proper guardrails, we could face serious setbacks such as big mistakes, misuse, or a public backlash. That would slow down or even block deployment in many areas.
Cutting Through the Noise: What the 2027 AI Prediction Really Tells Us
After looking closely at both the hopeful forecasts and the tough realities, one thing is clear: the debate around AI taking over computer-based work by 2027 isn’t black and white.
There are strong reasons to believe AI will keep improving fast. But there are also big reasons to doubt it will truly replace humans in “almost every” job done on computers—at least not that soon.
Let’s unpack both sides and see what really holds up.
The Case for a Fast AI Takeover
Some experts—and the data—say we might be closer than we think. AI is improving at a breakneck pace. New models are doubling their task-solving ability every few months. Some tools already beat humans in narrow areas, like solving coding problems under time pressure.
Costs are also dropping fast. Running powerful AI systems is getting cheaper by the day, and new open-source tools are making this tech more accessible to small teams, not just big tech companies.
On top of that, businesses are pouring money into AI. Most companies are already using it in some form. Many are betting big that it will unlock major savings and productivity boosts.
A few experts even think we’re on the edge of something bigger—like AI designing better versions of itself. If that happens, progress could speed up even more, potentially triggering what some call a “recursive loop” of rapid innovation.
The Skeptics Push Back
Still, there are plenty of reasons to be cautious.
First, Brundage himself added important footnotes to his claim. He said that just because something is technically possible doesn’t mean it’ll be used widely. He also said his idea of “more effective” only applies to tasks where you judge the outcome alone, without factoring in the human element.
That’s a big caveat. In real life, a task’s value often depends on trust, creativity, or human judgment—not just speed and cost.
There are also big technical gaps. AI still struggles with logic, planning, and adapting to new situations. It often makes mistakes and can’t explain its thinking. And while it might handle parts of many jobs, it rarely does the whole job without human help.
Deployment is another mountain to climb. Most AI tools today are hard to install, expensive to train, and even harder to scale. Most AI projects don’t meet their goals. And smaller businesses usually can’t afford the time or money to adopt them.
Then there’s the hype problem. Experts warn we may be in a “hype cycle,” where people overestimate what AI can do in the short term. This has happened before with other tech trends—and it leads to a period of disappointment before real progress continues.
Finally, some experts argue today’s AI—especially large language models—has built-in flaws. They say these tools don’t really understand the world and may never be able to match human intelligence unless we find a whole new approach.
So, How Likely Is the 2027 Vision?
Here’s a balanced take on where we’re headed by 2027:
- AI doing more tasks? Very likely. AI will become even more capable at handling routine, data-heavy, and clearly defined jobs.
- AI doing “almost every” task? Unlikely. Many tasks still require creativity, trust, or judgment—things AI hasn’t mastered. And many niche jobs lack the clean data AI needs to learn.
- AI being used everywhere? Very unlikely. Deployment is messy and expensive. Businesses still need time, people, and trust to make it work.
What’s more realistic? AI will keep reshaping jobs—but mostly by helping humans, not replacing them outright. Expect more “augmented work,” where AI handles the simple stuff and humans focus on the hard parts.
Some jobs will change a lot. Some may disappear. But most will evolve, not vanish. Companies that adapt, reskill their teams, and think carefully about where AI fits will come out ahead.
So Why is the Debate Is So Messy?
Much of the confusion around Brundage’s prediction comes down to one thing: unclear language. Even he said he was being a bit “cheeky” with how he phrased it.
Depending on how you read it, his statement could mean:
- Computers will get better software.
- AI will automate more of the work humans do on computers.
- AI will replace most human thinking jobs.
Each version has a very different level of impact and likelihood. That’s why having clear definitions really matters when we talk about the future of AI.
2027 Will Be the Year of Augmented Work; Not an AI Takeover
By 2027, AI will be more capable, cheaper, and widely used; not dominant. It will likely handle more routine, computer-based tasks, especially in coding, content creation, and data processing.
But for jobs that require creativity, empathy, or judgment, humans will remain essential.
The real shift won’t be about replacement; it will be about redesign. Most jobs will evolve as AI takes on support tasks, freeing people to focus on higher-level work. This is less a takeover, and more a transformation of how we work.
To manage this change:
- Policymakers should invest in AI safety, update laws, support education, and strengthen social safety nets.
- Business leaders should adopt AI where it adds value, retrain staff, and design jobs for human-AI collaboration.
- Individuals and schools must prioritize lifelong learning, teach digital and human-centric skills, and prepare students for an AI-augmented world.
Ultimately, the winners won’t be those who predict the future perfectly, but those ready to adapt. 2027 will mark a major turning point, but not the end of human work.