Anthropic CEO: Within five years, AI will truly replace humans. Who will be the first batch?
Last week, at the Dreamforce 2025 summit, Dario Amodei, co-founder and CEO of Anthropic, made a statement that caught wide attention:
I'm still optimistic about the complementary nature of AI in the short term, but I must be honest: real replacement will start to occur in two to five years.
This isn't the first time someone in the tech industry has talked about "AI replacing humans."
The difference is that when Amodei made this statement, there had been earth-shaking changes within Anthropic: Over 90% of the code in their team is now automatically written by Claude, and tasks such as bug location, system debugging, and even product reconstruction are all completed by agents. The role of human engineers is no longer to write code but to review Claude's work.
This means that "AI replacement" is no longer a threat in science fiction movies but has started to become a reality in real production systems, beginning with lines of code in Anthropic's office.
But Amodei's warning isn't just for engineers.
"This won't only happen to developers," he added. "Enterprise customers in industries such as insurance, finance, and healthcare are already using Claude to perform end-to-end tasks. The first people to be replaced are often those in positions in the workflow that are most easily automated."
The idea of "AI replacing humans" has entered the countdown from theory.
Next, the question is no longer whether AI will replace you, but who will be replaced first? How are they being replaced step by step? And how should you redefine your own role?
Section 1 | The Starting Point of Replacement: End-to-End Capability
Many people today say that Claude can write code, but Dario Amodei's focus isn't on code writing at all. Instead, it's about its ability to complete a whole task.
He observed that in the past, AI only helped you write a few lines of code, but now it can debug systems, fix bugs, and complete the entire deployment process.
In other words, it's no longer you taking the lead and AI providing assistance. Instead, Claude completes a task from start to finish, and humans review and modify. The roles have changed, and so has the relationship.
Within Anthropic, this transformation has already started to take effect.
Dario gave a real - life example: When we recently released a model, there was a bug in the cluster. Engineers searched for it for several days but couldn't find it. Later, we asked Claude to investigate, and it actually found a hidden problem that everyone had missed.
It has evolved from AI doing something according to your instructions to AI being able to judge, execute, and fix on its own, completing the entire closed - loop process.
Moreover, this isn't an isolated incident.
Dario clearly stated:
"Currently, 90% of the code in our team is written by Claude. The role of humans is more like that of an editor or supervisor."
In fact, engineers are no longer the executors. Instead, they treat AI as an "intern" or an "outsourcer" to assign tasks, inspect results, and ensure quality.
Not only within Anthropic, but also Claude's enterprise customers are verifying this point.
Mike Krieger, co-founder of Anthropic, also revealed in an earlier interview: We have a customer who let Claude run continuously for 30 hours to complete a complex system reconstruction.
30 hours, without human on - duty, and the task didn't stop.
This is no longer just a conversational model but a long - running virtual executor.
It can remember the goal, track the context, and discover and correct problems. What it does is no longer just language generation but task delivery.
And Dario's advice to enterprises is more forward - looking:
In a conversation on October 20th with Diogo Rau, the Chief Information Officer of pharmaceutical giant Lilly, he said:
"Don't be limited by the idea of what small things we can do with AI. There is an existing process with 20 parts. If you want to introduce AI in the 5th and 12th parts, it's actually very difficult. But in a year, AI may be able to complete the entire process from the 0th to the 20th part end - to - end."
If the model isn't powerful enough until a year later and you start deploying it then, you'll delay for another two years. Have confidence in the speed of technological progress. Start preparing for end - to - end transformation now.
This statement reveals the essence of replacement: Because replacement doesn't happen suddenly at a certain moment but starts from the moment when AI changes from "assistance" to "delivery."
When Claude doesn't just answer questions and generate text but can:
- Understand the task goal
- Invoke tools and agents
- Run autonomously and correct errors
- Deliver the final result
Do you still need several people to do this? We used to think that AI only helped you work faster, but Dario's words clearly indicate that the moment Claude can perform end - to - end tasks means you no longer need these intermediate steps.
Code is just the beginning.
What really replaces people isn't that AI has become smarter but that it starts to work like a system: continuously executing, connecting processes, autonomously invoking, and correcting and optimizing.
This set of execution capabilities is the watershed for AI to transition from a tool to a job position.
Section 2 | The First to Be Replaced: Those in Intermediate Positions
We often think that the positions to be replaced by AI will be those with low technology requirements, low educational qualifications, and low entry barriers, such as clerks, assistants, and customer service representatives.
But Dario Amodei's judgment is exactly the opposite.
He said: It's not that humans have nothing to do, but in two to five years, the entire economic system will be deeply reshaped. The most affected isn't individual positions but that all links may be compressed, replaced, and reorganized.
In other words, what AI replaces isn't those with poor skills but those who transfer information in the process rather than create results.
As long as a job doesn't involve core creation but only transfers information from A to B and then organizes it into C, Claude can do it for you, and it may be faster, cheaper, and error - free.
You'll find that many intermediate positions that seem to require technical skills are actually just information carriers:
- Organize meeting content into a report
- Summarize customer service feedback into a weekly report
- Fill data into a form and then export it as a PPT
- Break down a marketing plan into specific work orders and distribute them to different departments
The common characteristics of these jobs are:
- The information already exists
- The steps are clear
- They don't involve complex judgments
- The results can be verified
And this is exactly what AI is best at.
Mike Krieger added a real - life example of an Anthropic customer: A customer in the finance department uses Claude to automatically generate analytical Excel tables. Claude can understand the data, make inferences, and draw charts on its own.
It's not about making Excel a tool for humans to use, but Claude directly takes over the entire process.
So this isn't a war between programmers and AI but a replacement between "those in intermediate positions" and "those who complete the whole thing."
Looking further, as long as a company uses AI to:
- Review contracts → Integrate documents → Write meeting minutes
- Archive work orders → Generate FAQs → Automatically send emails
- Write budgets → Analyze expenses → Produce annual reports
How many people are still needed for the positions specifically dedicated to these tasks?
Because in the past, a task required three or four people to take turns, but now AI can complete it in one step.
This is the real - life basis for the first group of people to be replaced.
Dario's judgment isn't based on emotions but on the changes that have already occurred within their company:
We haven't laid off engineers, but the role of each person in our team is being redefined.
The meaning of this statement is clear: AI doesn't eliminate jobs all at once but marginalizes the original human resources.
The first to be replaced are exactly those positions without decision - making and creative power.
Section 3 | The New Way of Working: Not Doing the Work but Directing AI to Do It
The moment AI really starts to replace people, the most intuitive concern for many people is whether we'll be out of work.
But Dario Amodei gave an unexpected answer at the Dreamforce summit: You may need more people because they can have a greater leverage effect. Engineers can become ten times more productive.
If AI is replacing people, why do we still need more people?
- IG Group case: Their analysis team saves 70 hours per week, and these hours are reinvested in higher - value strategic work. The productivity of certain use cases has doubled, and ROI has been achieved in 3 months.
- Cox Automotive: After using Claude in VinSolutions CRM, the number of consumer inquiries and test - drive appointments has more than doubled.
- Palo Alto Networks: After 2500 developers used Claude, junior developers with no prior Claude experience completed complex integration tasks 70% faster.
The common point of these cases is that people aren't "replaced" but their "roles" are upgraded.
But this transformation isn't easy.
A study of 16 experienced open - source developers found that when they used AI tools, the time to complete tasks actually increased by 19%. Developers predicted that AI would make them 24% faster, but instead, they were 19% slower.
Why? Because experienced developers have a lot of context, while AI doesn't. They need to transform their problem - solving strategies into a form that AI can understand and also spend time debugging AI's output.
But learning to manage AI takes time and practice.
And there may not be much time left.
Research from Stanford University shows that the employment rate of 22 - 25 - year - old software developers has dropped by nearly 20% since the end of 2022 because AI tools are taking over the routine coding and data tasks that were previously assigned to junior employees.
Christopher Stanton, a professor at Harvard Business School, warns that as employers redefine early - career roles, wages may decline. However, at the same time, the Chief People Officer of Salesforce said that the company is recruiting new graduates on a large scale, taking in 1000 interns in the summer of 2025. She pointed out that there are now many new positions that didn't exist six months ago.
In the future, not everyone will be unemployed; instead, job roles are being redefined.
So, what kind of people can adapt to this change? Truly high - value people can:
- Clearly express their requirements
- Reasonably assign tasks to AI
- Verify AI's work results
- Know how to correct AI when it makes mistakes
This is the "AI coordinator."
Dario clearly stated at Dreamforce: I'm very worried, especially about people's ability to adapt because jobs are changing too fast.
This isn't alarmist talk but a reality that's happening.
The time window is only two to five years.
Conclusion | It's Not About Whether but When
Dario made it clear: "It's not about replacement but re - arranging the division of labor."
AI is no longer just a tool but a virtual colleague capable of independently completing work. It can complete a process from start to finish, find problems that humans miss, and work continuously without a break.
What's really replaced isn't the position but the person who can only do one part of the work and doesn't know how to use AI.
Mike Krieger, CPO of Anthropic, said: What we're building is a reliable virtual colleague.
It's not about providing assistance; it's about taking on the job.
It's not about helping you; it's about finishing what you can't.
In the next two to five years, the changes may be faster than most people expect.
📮Reference Materials:
https://www.youtube.com/watch?v=wUOjTR1511M&t=1s
https://www.youtube.com/watch?v=uhJJgc - 0iTQ&t=309s
https://reg.salesforce.com/flow/plus/df25/sessioncatalog/page/catalog/session/1744144502316001kd1E?utm_source=chatgpt.com
https://www.businessinsider.com/anthropic - ceo - warning - world - ai - replacing - jobs - necessary - 2025 - 9?utm_source=chatgpt.com
https://www.techradar.com/pro/live/dreamforce - 2025 - were - live - in - san - francisco - for - all - the - latest - salesforce - news - as - it - happens?utm_source=chatgpt.com
This article is from the WeChat official account "AI Deep Researcher." Author: AI Deep Researcher. Republished by 36Kr with permission.