TL;DR
Economist Luis Garicano argues young professionals should prioritize roles that bundle many interdependent tasks—what he calls 'messy' jobs—because single, well-specified tasks are increasingly automatable. He illustrates how regulatory choices, firm incentives and local, non-codified knowledge will shape which occupations persist and where opportunities for entrepreneurship may arise.
What happened
In a New Year’s essay Luis Garicano revisits career advice for students in light of rapid advances in artificial intelligence. Drawing on work with Jin Li and Yanhui Wu, he frames knowledge work on a spectrum from single, rule-based tasks to complex, multi-task “messy” roles. Single-task roles—such as routine programming, basic contract drafting or simple customer support—are increasingly vulnerable as AI systems improve and error rates fall. By contrast, jobs that require local knowledge, in-person execution, organizational navigation and relationship-building are harder to automate. Garicano cites examples including radiology (where clinicians spend much time on non-image tasks), factory engineering, construction contractors and post-merger integration managers. He also highlights how AI lowers the fixed costs of starting firms, giving the Base44 case—an app built largely with Claude that scaled quickly and was acquired by Wix—as an example of new entrepreneurial paths. He notes that firms and regulators will influence the pace of adoption and that cultivating deep domain knowledge remains essential.
Why it matters
- Career planning: Roles made up of many interdependent tasks are likelier to persist as AI automates discrete tasks.
- Policy and firm choices will shape how quickly automation displaces workers; regulation can delay or block adoption.
- Entrepreneurship: AI reduces overhead and specialization costs, enabling smaller teams to compete with incumbents.
- Workforce strategy: Deep domain knowledge and execution skills become key assets in an AI-assisted economy.
Key facts
- Author: Luis Garicano; essay published Jan. 2, 2026; work referenced with collaborators Jin Li and Yanhui Wu.
- Garicano defines work along a 'messiness' spectrum from single-task to complex multi-task roles.
- AI excels at codified, single tasks; error rates are falling and models are improving, increasing automation risk for simple tasks.
- Radiology example: a cited 2013 study found radiologists spend 36% of their time looking at scans; the remainder involves interaction and other duties.
- Regulatory and institutional constraints matter: Garicano notes many European jurisdictions limit services like Uber and protect notary roles.
- Case study: Base44, built by Maor Shlomo using Claude for about 90% of frontend code, reached 250,000 users, reported $189,000 monthly profit and was acquired by Wix for $80 million after a roughly $15,000 personal investment.
- Some legal and professional roles differ in exposure—Garicano expects corporate law to be more automatable, while trial attorneys are likely less exposed.
- AI implementation itself creates complex organizational work that may be difficult to automate and will require domain-savvy practitioners.
What to watch next
- The pace at which AI model error rates decline and models move from assisted to unsupervised operation.
- Regulatory and firm-level choices that mandate or remove human oversight (e.g., rules keeping humans in the loop).
- Adoption patterns across sectors—whether risk-averse fields shift to unsupervised AI or retain human checks.
- The emergence of startup examples that scale using AI to absorb specialist functions and lower fixed costs.
Quick glossary
- Messy job: A role composed of many interdependent tasks, often requiring local knowledge, coordination, judgment and real-world execution.
- Single-task job: Work defined by a clearly specified, repeatable task that can be described by rules or code.
- Codified knowledge: Information and procedures that can be written down, formalized and encoded so that machines can learn or execute them.
- Autonomy threshold: The point at which AI systems are deemed reliable enough to operate without human supervision.
- Tribal knowledge: Informal, context-specific know-how held by people in an organization or community that is hard to formalize.
Reader FAQ
What should students prioritize when choosing a career?
Garicano recommends favoring roles with deep, domain-specific substance and tasks that require execution, relationship-building and local knowledge.
Are radiologists likely to be replaced by AI?
Not according to the essay: while imaging interpretation can be automated, much of a radiologist’s time involves non-image tasks—Garicano cites a study showing 36% of time is spent on scans—so the role is a bundle of tasks that remain valuable.
Is starting a company a safer path in an AI world?
The essay argues that entrepreneurship is especially viable because AI lowers fixed and variable costs, giving small teams leverage to scale; the Base44 example is offered as evidence.
Can governments or firms prevent automation?
Yes. Garicano emphasizes that regulatory choices and firms’ adoption decisions can slow or block automation; examples cited include limits on Uber and protected notary roles.
Discover more from Silicon Continent Essays about Europe, economics, innovation, and growth Over 6,000 subscribers Subscribe By subscribing, I agree to Substack's Terms of Use, and acknowledge its Information Collection…
Sources
- A New Year's letter to a young person
- AI Means the End of Entry-Level Jobs
- The Entry-Level Hiring Process Is Breaking Down
- How to AI-proof your job applications in 2026
Related posts
- Nightshade tool poisons images to deter unauthorized AI model training
- Anatomy of BoltzGen: Designing protein binders from PDB targets
- Beyond Benchmaxxing: Why the Future of AI Is Inference-Time Search