Key Takeaways

• The job market is shifting from a human-only economy to a hybrid human–agent economy, powered by protocols like A2A, MCP, and AP2.
• By the time today’s freshers graduate, AI agents will be collaborating, transacting, and working alongside humans, creating parallel economies with overlaps.
• Students in natural sciences, engineering, and humanities should prepare now by building hybrid skills, experimenting with agents, and understanding how human work integrates with the agent economy.

Introduction

This week is freshers’ week, and as I look out at the new faces joining university, I can’t help but wonder what kind of world they will graduate into. Recent evidence suggests that change is already well underway. According to Anthropic’s Economic Index, nearly half of employees in the US now use AI in their daily work, and the number of tasks fully delegated to AI agents has jumped from around a quarter to almost 40% in just a year [Anthropic, 2025]. At the same time, businesses are hiring fewer entry-level staff, leaning more heavily on experienced workers who can pair their expertise with AI tools.

To me, this isn’t a story of jobs vanishing, but of how the economy is evolving. By the time today’s freshers graduate, junior roles may look very different, but opportunities will open up for those who can demonstrate real expertise, project-building experience, and fluency in working with AI agents. This is the beginning of what many now call the AI agent economy—a shift from a human-only workforce to a world where humans and agents collaborate, transact, and create value together. The question is: how can students prepare now to thrive in this hybrid economy?

From Mainframes to Micro-Agents: A New Computing Era

In the early days of computing, only large organisations had access to mainframes. Fast-forward to the 1980s and desktops decentralised computing. The internet era then connected billions of computers worldwide, just as social media later enabled the rise of the creator economy. Today’s LLMs resemble the mainframes: powerful, centralised, controlled by a few tech giants. But the next step is clear: personal agents and micro-agents. These are smaller, task-specific AIs that individuals or businesses will own.

These agents will be connected through standards like the Agent-to-Agent Protocol (A2A) for communication, Model Context Protocol (MCP) for tool use, and Agent Payments Protocol (AP2) for transactions. Recent announcements from major tech players underscore this trend: Google’s new protocols for agent communication and payments, Coinbase’s stablecoin payment extension for AI agents, and Anthropic’s push for a universal AI integration standard.

To clarify, a micropayment is a very small digital payment—often a few pence or even a fraction of a penny—made automatically when one agent uses another agent’s service. Imagine your agent pulling a weather forecast: it could cost 0.2p, paid instantly and invisibly via AP2. These tiny, low-friction transactions are what make a scalable agent economy possible.

This evolution will unlock an economy where agents collaborate and pay each other in real time, reshaping work itself. Already, we’re seeing the emergence of agent marketplaces—similar to the early app stores—where agents can be discovered, bought, or hired. Google has even launched its own AI Agent Marketplace, signalling that the ecosystem is moving quickly.

Recent data from Anthropic’s Economic Index shows exactly this shift: the share of AI use that is “directive” (where full tasks are delegated to AI agents) has risen from about 27% to 39% in just a year [Anthropic, 2025]. This suggests growing trust in letting agents act autonomously rather than merely assist humans.

Parallel and Overlapping Economies

The coming years will see a parallel economy emerge:

• The physical economy: Human work in labs, factories, studios, and classrooms—still vital, still irreplaceable.
• The agent economy: Autonomous AI agents performing tasks, making micropayments, and negotiating services on behalf of people and organisations.
• The overlap: Humans creating the unique data, experiments, content, and cultural products that agents cannot generate alone, and agents amplifying and monetising that work.

Mini Case Story: An environmental science student collects water quality data for a dissertation. By publishing that dataset as an MCP-connected service, the student can license it for micropayments. Each time an agent from a research institute or consultancy queries the dataset, the student earns a small payment. What was once “just coursework” now becomes a monetisable micro-service in the agent economy.

Recent findings from Anthropic’s Economic Index illustrate this early adoption: the share of educational tasks carried out with AI rose from about 9.3% to 12.4% in just a few months, while scientific tasks grew from 6.3% to 7.2% [Anthropic, 2025]. This shows that universities and students are already leaning on AI agents for learning and research, signalling the shift is well underway.

What This Means for Students Across Disciplines

Natural Sciences

Agents will handle vast simulations, pattern recognition, and automated analysis. Students need to focus on data quality, experiment design, and MCP integration so their results feed reliably into the agent economy. Unique, high-quality fieldwork will remain irreplaceable.

Mini Case Story: A biology student deploys soil sensors in a local park. Their MCP-connected micro-agent collects and shares the readings. Other agents—from city councils to climate start-ups—can access this data for a micropayment, turning fieldwork into recurring income.

Engineering

Engineering students will graduate into a world where agents can design, optimise, and even source components. The role of human engineers will shift toward orchestrating systems, validating outputs, and ensuring safety. Building hybrid skills—physical prototyping plus agent orchestration—will be key.

Mini Case Story: A mechanical engineering student builds a prototype drone. Alongside it, they develop a micro-agent that provides flight performance data as a service. Drone manufacturers’ agents can purchase insights through AP2 micropayments, creating a bridge between student innovation and industry.

Humanities

For humanities students, the concern that “AI will write essays” misses the bigger picture. Agents will summarise, translate, and remix, but they will still need authentic content, cultural insight, and ethical framing. Historians, writers, and philosophers will be needed to create content and frameworks that agents cannot, and to shape the narratives by which society makes sense of this new economy.

Mini Case Story: A history student digitises archival letters and creates a small MCP-linked collection. Literary agents and education platforms can pay tiny fees per access to use excerpts in their work. The student’s archival project suddenly has a global reach and revenue stream.

Anthropic’s Economic Index also highlights a labour market shift: entry-level hiring is declining in many AI-exposed sectors, while experienced workers who can combine domain expertise with AI tools are increasingly favoured [Anthropic, 2025]. This reinforces the need for students to build expertise, product-building experience, and AI fluency, rather than assuming traditional junior roles will exist in the same way.

What Might a Real Person Do in the Agent Economy?

Here are some roles humans will play while agents transact and collaborate:

Role / Activity	What they’d do	Why it’s valuable in agent economy
Content creator / domain expert	Produce original data, content, analysis, case studies, field observations; produce proprietary datasets. Set licensing conditions.	Agents need reliable, accurate inputs. Your content can generate micropayment revenue. Your content powers agents and tools.
Agent service provider	Build micro-agents: specialized summarizers, translators, data cleaners, educational aids, domain-specific consultants (e.g. environmental risk evaluator agent).	These agents can be “called” via A2A by other agents; you earn via micro fees. You’re offering your skills in an automated, reusable form.
Agent tool / integration developer	Build or maintain MCP servers: connectors to data sources, tools (weather APIs, lab databases, geographic info systems), document archives. Develop standard interfaces.	As MCP becomes common, people who know how to integrate tools safely will be in demand. Also good basis for entrepreneurial projects.
Ethics / policy / rights specialist	Define licensing models, copyright policies, oversight of how agents use content; mediate disputes; work on regulation around mandates, digital identity, accountability.	Transactions between agents and content creators need policy guardrails. Someone has to design them, audit them.
Hybrid engineer / scientist / storyteller	Combine domain knowledge + technical agent knowledge: e.g. environmental scientists who build agents that monitor ecosystems, generate real-time reports, decide when to trigger alarms. Humanities grads who build educational agents, storytelling agents, cultural heritage agents.	They bring what agents need: context, nuance, human insight. Agents can’t fully replace these, so those with the skills and domain background + agent-tool fluency will be valuable.

Preparing for the Seismic Shift

Here’s how students can act during their time at the university:

• Learn the fundamentals of AI literacy: If you haven’t used AI much before, start now. Learn how to prompt responsibly, evaluate outputs, and understand limitations.
• Use AI responsibly in your learning: Treat AI as a partner, not a shortcut. Use it to explore, draft, and test ideas—while ensuring your own understanding comes first.
• Learn the protocols: Get familiar with A2A, MCP, and AP2. Think of them as the TCP/IP, USB-C, and PayPal of the agent world.
• Build a portfolio agent: Even a simple summariser or scheduling assistant shows you can work with this new medium.
• Understand licensing: Learn about copyright, Creative Commons, and how micropayments will change content economics.
• Experiment early: Join hackathons, contribute to open-source agent tools, or create an MCP connector for a dataset you care about.
• Adopt the right mindset: Don’t prepare to be the worker agents will replace—prepare to direct, design, and collaborate with them.

Risks and Ethics

Of course, challenges remain. There’s the risk of inequality between those who own agents and those who don’t. Copyright disputes will flare up, even with micropayments. Security and privacy will need constant attention. But these are precisely the areas where thoughtful graduates can make their mark—by building systems that are fair, safe, and accountable.

Anthropic’s Economic Index also shows important geographic and sectoral differences. AI use is concentrated in higher-income countries, with Singapore showing usage over four times higher than expected, while India and Nigeria remain far lower. In high-adoption regions, AI tends to augment human work and diversify into areas like education and science, whereas in lower-adoption economies it is more often used for full task automation. This highlights that the agent economy will not unfold uniformly, and graduates should stay mindful of local variations [Anthropic, 2025].

Conclusion

When I stand in front of freshers this week, I want them to see that their degree is not just about surviving exams—it’s about preparing to thrive in an economy where human and agent work are deeply intertwined. By 2028, when they graduate, agents will be negotiating, transacting, and creating value at scale. But the graduates who succeed will be those who combine their human insight with fluency in this new agentic infrastructure. They will not just compete with agents—they will partner with them, profit through them, and design the rules of the game they play in. And that, I believe, is an exciting future to step into.

Acknowledgment

This article was created with the assistance of AI tools, including ChatGPT, Gemini, Google NotebookLM, and Napkin AI for research, structuring, and image generation.