The last two years felt like the AI wave had arrived. Chatbots appeared everywhere. Everyone tried ChatGPT. But that was the splash before the wave.
The actual wave — the one that restructures how knowledge work happens — is just starting to roll in. It is called agentic AI. Instead of chatbots that respond to prompts, agents are systems that pursue goals autonomously, use tools, hand off work to each other, and complete entire workflows with minimal human direction.
According to Gartner, 40% of enterprise applications will have task-specific AI agents integrated by the end of 2026, up from less than 5% in 2025. That is an 8x jump in roughly 18 months. The market for these agents is projected to expand from $7.6 billion in 2025 to over $50 billion by 2030.
The chatbot era was knowledge workers using AI. The agent era is AI doing the work. Those are very different things, and the second one is what professionals should actually be preparing for.
The shift from prompts to goals
A chatbot answers a question. The interaction is bounded. The user is still doing the work, just with a smarter search engine.
An agent is given a goal. It plans the steps, calls the tools it needs, pulls data from one system, processes it in another, sends an output to a third. It can run for hours or days. It hands off subtasks to specialised agents that handle pieces of the workflow a human would not even think to design separately.
In the old model, AI was a tool. In the new model, AI is a coworker under supervision.
CIO recently described how the engineer of 2026 will spend less time writing foundational code and more time orchestrating a portfolio of AI agents, reusable components, and external services. That is not a future prediction — it is already starting at companies early on this curve.
For any role that mostly involves executing predictable workflows, the wave coming is not a marginally better chatbot. It is a system that can do the whole workflow without human involvement.
What the agent era actually changes about jobs
Tasks that involve coordinating across multiple systems become prime candidates for automation: customer onboarding, invoice processing, data reconciliation, standard report generation, routine code reviews, first-pass customer support — anything that follows a knowable sequence with defined inputs and outputs.
The first wave of generative AI excelled at one-shot tasks — drafting an email, summarising a document. The second wave excels at multi-step tasks. Receive a customer complaint, classify it, pull relevant account history, draft a response, route to the right person if it exceeds a threshold, log the interaction. That entire flow used to require five or six emails and several people. Now it is an agent, supervised by one person who handles exceptions.
Roles centred on the routine version of those flows today will change dramatically in the next 24 to 36 months. Not necessarily disappear, but transform. The work at the end of that window will look almost nothing like the work being done now.
The supervisor role, not the worker role, is the position of value
When the system is about to change, the most valuable position to occupy is the one that designs the new system — not the one that performs the old work better.
The professionals who will benefit most from the agentic wave are those positioned to supervise agent workflows: defining what the agent should do, setting guardrails, monitoring outputs, handling edge cases, and improving the system over time.
This is not exclusively a technical role. It is a judgment role. Someone has to decide what should be automated, what should not be, what good output looks like, and what failure modes matter. Someone has to sit between AI speed and actual business needs. Someone has to know when to override and when to trust. That function is going to remain valuable for the foreseeable future, and the work involved is not replaceable by the same wave displacing routine workers.
The skills the agent era actually rewards
Systems thinking. The agent era rewards people who can see how work flows across functions — where the dependencies are, where the friction points sit — and design where an agent fits into the whole pipeline.
Judgment under uncertainty. Agents will produce outputs that are 85% right. Knowing whether the remaining 15% matters in a specific context is the entire game. That is a human skill, and it is becoming more valuable, not less.
Domain depth. Agents need to be configured for specific contexts, and configuring them well requires genuine understanding of the domain. A marketer who knows their industry deeply will be far more valuable directing AI marketing agents than a generalist with no specific expertise.
Communication and translation. Supervising agentic systems means communicating intent clearly — both to agents and to the humans who depend on the outputs. Professionals who can write a precise spec, explain a complex result, and translate between technical and business audiences will be in extraordinary demand.
Practical moves to make this year
Start using agentic tools now, even in their imperfect current form. The fluency built over the next twelve months will be hard to replicate later.
Identify one workflow that could plausibly be agent-driven and start designing what that would look like. Even if an organisation is not ready to deploy it, the exercise of mapping it out teaches more than any course.
Build relationships with the technical side of the organisation. The agent era will require significant cross-functional collaboration between domain experts and the people deploying agents. Being the person who speaks both languages is a quietly powerful position.
Get specific about individual value. Vague positioning will not differentiate against 30 other professionals saying the same thing while their roles get partly automated. What specific judgment, what specific decisions, define the edge? Sharpening that answer matters now.
For a data-driven assessment of where existing capabilities map against the direction of the labour market, Trajectory uses real labour-market data to help professionals navigate exactly this shift.
The bottom line
The first wave of AI gave professionals time. The second wave will test what was done with it.
Most people used the first 24 months to either ignore AI entirely or integrate chatbots in modest ways. Both positions are about to look insufficient. The agent era does not care whether someone became comfortable with prompting. It cares whether they can supervise systems, exercise judgment, and design work that humans and agents do together.
The real wave is coming. Those who position themselves now, before it fully lands, will look prescient by 2028. The window is shorter than it looks.
