New Jobs Created by AI: The Roles Anthropic's Model Says Are Coming

Anthropic's Economics team has built a model of how AI could reshape the economy by 2030, and it includes a category almost nobody is discussing. Alongside tasks that get automated and tasks that get augmented, there is a fourth group. New tasks created by AI.

Everyone is arguing about what disappears. Very few people are looking at what appears.

I am Riten, founder of Fueler, a platform where people get hired through proof of work. I think this fourth category is the most under discussed part of the entire report, Economic Scenarios for Transformative AI (Korinek et al., 2026), and it is where a lot of the opportunity over the next five years is going to sit.

What Anthropic's Model Says About New Tasks

The model treats every job as a bundle of tasks drawn from the US Department of Labor's O*NET taxonomy. Each task can go four ways. It stays untouched. It gets augmented, meaning AI helps a human do it faster or better. It gets automated. Or it is brand new, created because AI exists.

The report is direct about the historical pattern. New technologies have also created new tasks for workers.

It uses a nurse to illustrate. For her, the new tasks might be checking how well an AI triages incoming patients, or reviewing a care plan the AI proposed. Neither of those jobs existed a few years ago. Somebody has to do them now.

The report also points out that this churn is normal and constant. Tasks leave the bundle. Hardly anyone hand writes paper charts anymore. Tasks join the bundle. Thirty years ago nobody monitored patients remotely. The bundle has never been static.

There is one important exception, and it is what makes the worst scenario the worst. In the extreme scenario, AI is more productive than humans at the vast majority of knowledge work tasks, does nearly all of them autonomously, and creates essentially no new knowledge tasks for people. That last clause is doing enormous work. The extreme scenario is not frightening only because of automation. It is frightening because the replacement mechanism switches off.

In the modest and substantial scenarios, it does not.

Five Kinds of New Work AI Is Creating Right Now

The report gives the framework. The specific examples below are my own reading of what that framework points toward, based on what I see companies hiring for.

• Checking whether the AI got it right. The report's own nurse example includes checking how well an AI triaged a patient. This is quality control for machine output, and it exists in every field where AI is now producing work. It requires knowing the domain well enough to spot what is wrong, which is why it cannot be handed to a beginner.
• Reviewing and approving AI proposals. The nurse reviews an AI proposed care plan. Substitute your own field. A marketer reviews an AI proposed campaign structure. An editor reviews an AI proposed outline. Somebody has to hold the standard, and holding a standard is a skill.
• Directing the tools well. Not typing prompts, but designing workflows. Deciding which tasks go to AI, at which stage, with what inputs, and how the output gets checked. This is process design and it barely existed as a role three years ago.
• Building the context AI needs. AI performs badly without good inputs. Someone has to assemble the briefs, the reference material, the brand rules, the constraints and the examples. That work is real and it is growing.
• Handling the tasks AI cannot reach at all. The report puts this plainly. AI cannot bathe a patient. Every field has its equivalent, the part that needs presence, physical action, or a human being trusted by another human being. These tasks do not shrink when the rest of the bundle automates. They often become a larger share of the job.

Why It Matters

This matters because of what the model says about wages and unemployment, which is less alarming than most coverage suggests.

Across all three scenarios, average wages rise. In the modest and substantial scenarios, job reallocation and unemployment stay within ranges history has already seen. Only in the extreme scenario, which the report says would likely require recursively self improving AI adopted very quickly, does unemployment spike beyond typical recessionary levels.

The economy also grows in every case. GDP ends up 1.6% higher in the modest scenario, 8.3% higher in the substantial one, and 32.4% higher in the extreme one.

A bigger economy with new tasks appearing is not a story about work vanishing. It is a story about work changing shape faster than our hiring systems can describe it.

And that is the real problem. Job titles lag behind reality by years. Job boards lag behind titles. Resumes lag behind job boards. By the time a new kind of work has an agreed name and a listed salary band, the people doing it have been doing it for three years already.

I should note the model's limits, since the authors state them clearly. It leaves out policy responses, business cycles, aggregate demand effects, financial market disruptions and hyper capable robots. It does not follow individual workers. Several reviewers pushed back, some arguing the extreme scenario is better understood as a thought experiment, others arguing the modest scenario understates what is already visible in the data.

What This Means for Your Portfolio and How You Get Hired

Here is the practical problem with new work, and it is one I think about constantly.

If you are doing a job that does not have a name yet, how do you apply for it? How does anyone find you?

Not through a resume. A resume needs a title, and the title does not exist. You end up writing something vague like "AI Operations" that means nothing to the person reading it, or you shove the work under an old title that undersells you completely.

The only format that handles unnamed work is one that shows the work.

Document new tasks the moment you start doing them. If you built a review process for AI generated content, that is a project. Write it up. What was broken, what you built, what it caught, what changed. This is exactly what a proof of work portfolio built step by step is designed to capture.

Make your AI workflow a first class part of your projects. Not a footnote. The workflow is the new skill. On Fueler you can add your AI stack to a project and show the full process, which is the clearest possible signal that you are doing the new work rather than talking about it.

Do not wait for a job posting to legitimise it. The posting will arrive two years late. Build the evidence now and you will be the obvious candidate when it does. If you are early in your career, the student guide to building a first portfolio shows how to build this kind of proof without formal experience.

Use a repeatable structure. New work is messy to describe, which is why people put off writing it up. The 6 step portfolio formula gives you a frame so the messiness does not stop you.

Final Thoughts

There is a version of the AI conversation that is entirely about loss. Which jobs go, which skills die, how much time we have left. I understand why it dominates. Loss is more gripping than gain.

But Anthropic's model has four task categories and only one of them is automation. The others are augmentation, tasks untouched, and tasks created. In the two more likely scenarios, that fourth category keeps producing work.

The people who do well over the next five years will not be the ones who guessed the future correctly. They will be the ones who noticed what they had started doing that nobody had a name for yet, and wrote it down clearly enough that someone else could see the value.

That is a small habit with a large payoff. Notice the new work. Document it. Make it findable.

Go build a career portfolio that actually gets jobs and put the work nobody has named yet right at the top of it.  

Frequently Asked Questions

What new jobs will AI create?

Anthropic's model includes a category for tasks created by AI, and gives examples such as checking how well an AI triaged a patient and reviewing an AI proposed care plan. These are quality control, review and oversight tasks that require enough domain knowledge to judge whether machine output is correct.

Does AI create more jobs than it destroys?

The report does not settle that question. It notes that historically new technologies have created new tasks for workers, and its modest and substantial scenarios keep unemployment within historically normal ranges. Its extreme scenario assumes AI creates essentially no new knowledge tasks, which is a major reason that scenario produces high unemployment.

What skills will be in demand because of AI?

Based on the task categories in the report, demand is likely in verifying AI output, reviewing AI generated proposals, designing workflows that combine human and AI work, preparing the context and inputs AI needs, and performing the physical or in person tasks AI cannot do at all.

How do I get hired for a job that does not have a title yet?

Show the work directly. Document what you built, the problem it solved and the results, then publish it where a hiring manager can read it. Job titles lag several years behind real work, so evidence is the only signal available in the meantime.

Will AI create new jobs in creative fields?

The report does not break its findings down by creative industry. Its framework suggests new tasks appear wherever AI is adopted, which in creative work points toward directing tools, setting quality standards, building briefs and reference material, and reviewing generated output before it ships.

Why 100,000+ professionals use Fueler

Fueler helps professionals showcase proof of work through projects, assignments, case studies, and achievements.

• Thousands of professionals use Fueler to create their digital portfolio
• Thousands of projects are published on Fueler. Check here
• Startups and Companies hire through proof of work on Fueler
• Used by freelancers, creators, marketers, video editors, writers, designers, and product managers

Our mission is to help the next 100 million professionals build a verified professional identity through proof of work