Let's talk

The Agentic Shift

How AI is Changing Work

Work is moving from manual task execution to the supervision of autonomous AI agents that handle entire workflows.

By Divyaprakash JhaUpdated 5 min readTopic: AI and work

Photo for How AI is Changing Work

Tasks move before jobs do

AI is shifting from answering questions to executing workflows. Work is moving from manual task execution to the supervision of autonomous AI agents that handle entire workflows.

To understand how your role might change, you have to look at how a job is actually built. Most jobs are not single, indivisible units; instead, they are bundles of many different tasks. AI tends to move into these roles by targeting specific tasks before it impacts the entire job title.

AI first automates repetitive screen-based tasks. This includes things like data entry, assembling reports, drafting basic emails, or handling first-level customer support. When a role is made almost entirely of these repetitive tasks, that role is highly exposed to automation. However, for most people, the reality is different.

Roles that mix these repetitive tasks with high-level judgment, human relationships, and responsibility usually don't disappear. Instead, they become much more productive. You move from being the person doing the manual data entry to the person overseeing the system that handles it. This shift is a core part of how AI is changing work.

From simple chat to durable workflows

Most people currently interact with AI through chatbots. These are great for answering a quick question or summarizing a paragraph, but they are limited. Current AI agents mostly finish tasks that take only a few minutes to complete.

The next major shift involves what experts call "durable workflows." This is a forecast for the years 2027–2028. A durable workflow is a process that takes time and involves multiple steps, such as following up with a sales lead for two weeks, reconciling accounts every night, or preparing a complex tender over several days.

Building these systems is an engineering challenge. To make an agent reliable enough to handle a multi-day process, the system needs several key features:

  • Memory: The ability to remember what happened yesterday to inform what happens tomorrow.
  • Retries: The ability to try again if a step fails without needing a human to restart the whole process.
  • Schedules: The ability to perform tasks at specific times.
  • Checkpoints: Specific moments where the agent must stop and wait for a human to approve the next step.

The ladder of AI autonomy

Organizations do not simply hand over their entire business to an AI overnight. Instead, they climb a "ladder of trust." A company moves a process up to a higher rung only when the rung below it has proven to be reliable through logs and error rates.

Rung Level of Autonomy How it Works
1. Assist Low The agent drafts a response or a document; a human reads every word and sends it.
2. Act with approval Medium-Low The agent prepares an action (like a refund or an email) and waits for a human to click "approve."
3. Act within limits Medium-High The agent acts alone on tasks below a certain threshold (e.g., refunds under a specific amount) and escalates anything higher.
4. Run the process High The agent owns the end-to-end workflow; humans act as auditors, reviewing summaries and exceptions.

By 2028, many routine processes are expected to sit on rung two or three. By the early 2030s, a significant portion of back-office work in well-run companies may sit on rung four.

New work in the agent economy

As we move toward this agentic economy, we will see a massive shift in the types of jobs available. The role of a worker is shifting from "doing" each individual step to "supervising" the system that does them. This creates entirely new professional categories.

Demand is rising for people who can design these agentic workflows—breaking a complex business process down into steps that an AI can follow. We will also see the emergence of "AI audit, evaluation, and assurance" as a recognized profession. Much like financial auditors today, these professionals will test agents, review logs, and sign off that a system is behaving as it was claimed to behave.

For those worried about jobs AI might replace, it is more helpful to think about the new roles being created to manage the systems that do the work.

Skills that compound in the age of AI

In an era where AI can generate content and perform tasks, certain human skills become more valuable because they allow you to control and verify the AI. These are the skills that "compound"—they become more powerful as the technology advances.

The Core Skillset

  • Clear thinking and writing: To get a good result from an agent, you must be able to provide a precise brief. If you cannot think clearly, you cannot instruct an agent effectively.
  • Deep domain knowledge: You cannot verify if an AI's output is correct if you don't understand the subject matter yourself. Expertise is what makes AI useful in your hands.
  • Process design: This is the ability to look at a job and break it down into inputs, specific steps, and necessary approval points.
  • Verification: The critical skill of spotting "confident errors"—mistakes that look correct on the surface but are factually wrong—before they reach a customer.

For more on how to stay relevant, see our guide on skills needed in the age of AI.

What stays human

As AI makes the production of text, code, and basic proposals very cheap, the value of certain human traits will actually increase. When anyone can generate a polished-looking proposal in seconds, customers will look for the things AI cannot replicate.

The things that stay human include:

  • Responsibility: AI can propose an action, but a human must own the outcome.
  • Taste and judgment: Deciding what is worth doing at all, and deciding if a result is "good" or just "correct."
  • Trust and relationships: A trusted name and a real human relationship become more valuable when the digital world is flooded with AI-generated content.
  • Physical care and leadership: The human elements of teaching, caring, and leading teams.

Impact on small businesses

Small businesses often have more to gain from AI agents than large corporations because they cannot afford to hire massive teams for every administrative function.

In India, we expect a massive shift toward voice-first agents. Many users prefer to speak in their local language—such as Hindi, Marathi, or Tamil—rather than typing in English. By 2028, voice agents that can answer calls, take orders, or explain bills in these languages will likely be common tools for small businesses.

Practical applications for small business owners include:

  • Sales follow-ups: An agent that messages every lead on time, ensuring no opportunity is missed.
  • Automated invoicing: Generating quotations and invoices in minutes based on past orders.
  • Continuous bookkeeping: Keeping GST paperwork and accounts current every day, rather than in a rush at the end of the month.

For a detailed look at these opportunities, read how small businesses can use AI.

Questions people ask

Will AI replace my job?

AI is more likely to change the tasks within your job than to eliminate the role entirely. It automates repetitive tasks, allowing you to focus on judgment and relationship-driven work.

What is an AI agent?

Unlike a chatbot that just answers questions, an agent is designed to execute workflows—performing multi-step actions like booking freight or reconciling accounts with human oversight.

How can I prepare for AI in my career?

Focus on developing deep domain expertise, learning to write precise instructions (briefs), and mastering the ability to verify AI-generated outputs.

What are the risks of using AI at work?

Key risks include security vulnerabilities like prompt injection, a dependency that erodes human skill, and the potential for misinformation or fraud via cloned voices.

How do businesses implement AI agents safely?

Start by picking one repetitive process and piloting an agent that requires human approval for every important action, building a record of logs and audits.