AI stopped advising. It started working for you.

084x

2024 was about chatbots. 2025 was about whether AI would take your job. 2026 is about something else entirely – about how much work AI will actually do for you while you oversee and direct it. And that shift is happening faster than most people realize.

Numbers that change your perspective

Researchers from OpenAI, Columbia Business School, and Wharton School published a study this year tracking the real behavior of hundreds of thousands of users of the agentic tool Codex. No surveys, no polls. Actual data from what people actually do.

A few numbers worth remembering:

Active Codex users grew more than fivefold in the first half of 2026. The fastest-growing adoption isn’t in IT – it’s in legal, HR, business development, and research.

The share of users submitting tasks estimated at eight or more hours of human work grew from 2.1% to 25.6% in just six months. A tenfold increase in half a year.

In June 2026, the median researcher at OpenAI generated 50x more output than in November 2025. The median lawyer, 13x more. “Output” here means real work – documents, analyses, reports.

Inside OpenAI itself, Codex now accounts for 99.8% of all AI output by employees. The chatbot has effectively disappeared from the workflow.

What’s actually happening?

A chatbot answers questions. An agent does work.

That distinction sounds simple, but it changes everything. When you ask a chatbot, you’re still doing the work – just with a faster reference. When you delegate to an agent, AI executes entire workflows: browsing files, editing documents, gathering data, writing emails.

The research describes this as a shift from “consultation” to “delegated production.” The data makes clear the shift is uneven – but where it takes hold, it takes hold completely.

At OpenAI, the legal and HR departments went from zero Codex usage to 75% in a single month. It didn’t happen gradually. It was a jump.

Why “experimenting” isn’t enough

More than 10% of active Codex users manage three or more agents simultaneously every week. These aren’t programmers. They’re managers, analysts, lawyers, researchers – people whose work has shifted from executing tasks to coordinating and overseeing what their agents do.

26.6% of users now build their own “skills” – codified workflows and instructions for recurring tasks. That’s a new form of knowledge management.

The shift to agentic tools isn’t just a technical change. It’s a reorganization of work from the ground up. A weekend experiment won’t prepare you for it.

TIP: The fastest way to figure out where you stand with AI agents is simple: pick one recurring task you do every week and try delegating it entirely – not just asking for help with it. The gap between what you expected and what you get tells you exactly what to work on next.

Expertise isn’t enough – but you can’t do without it

The data shows one more unexpected thing: agentic AI delivers the biggest advantage to those who understand what they’re delegating.

A researcher who understands methodology can tell when an agent’s results don’t add up. A lawyer who understands the case context gives more precise instructions. A manager who knows their field doesn’t need to check every step – they know which questions to ask.

AI doesn’t suppress expertise. It raises its value, by cheaply and efficiently replacing work that lacks it.

TIP: AI isn’t just changing individual workflows – it’s reshaping entire industries, job roles, and the way organisations are structured. Read more in our article.

Five things to work on this summer

Work slows down in summer. Meetings thin out, the inbox calms down. That’s exactly the time to do work there’s never room for otherwise – not urgent, but important.

Here are five of them.

Build something real – no coding required

Theory won’t get you anywhere. Try putting together a specific AI assistant: for planning a vacation, for exploring nature with your kids, for tracking expenses. Keep it simple. A project you actually have to finish will teach you more than any course.

Clean up your digital system

Summer is the right time to let AI go through your files, folders, and notes – rename and reorganize everything. Cowork or Claude Code will do it faster than you expect. Come autumn, you won’t be staring at a file called report_final_FINAL_v3.xlsx wondering what’s in it.

Stop asking AI. Start delegating.

Most people are still just consulting AI. Try shifting: give it a whole task, not just a question. Let an agent draft a proposal, analyze documents, prepare materials for a meeting. Then focus on what the agent can’t do – your judgment, context, and decisions. That’s the work that stays yours.

Codify your own workflows

Do you have recurring processes? Preparing reports, responding to typical inquiries, updating briefing materials? Write step-by-step instructions for an agent to follow. Research shows 26.6% of active AI agent users already do this. It’s called “skills” – and it’s a new form of knowledge management. Summer is the time to try it.

Start from the end

Picture the end of summer. What do you want your work to look like? What do you want set up, what do you want to know, which processes automated? Name that goal now – then ask AI to help you plan how to get there. A few minutes a day is enough. But only when you know where you’re headed.

Ready to prepare properly?

Reading about AI isn’t enough. You need to integrate it into your actual work context – with a strong domain foundation behind it. That’s exactly what the Effective MBA specializations at EDU Effective are built on.

Sign up and from day one you get access to 10 modules, each broken into 30 daily blocks of 15 minutes. Study in the morning, in the evening, whenever it fits your life – 100% online.

Choose your AI-focused Effective MBA specialization:

If the format doesn’t work for you, we’ll refund the full amount within 14 days – money-back guarantee, no questions asked.

Treat your professional development like your morning coffee. One cup a week produces no effect. A toothbrush used once a month doesn’t work either. Same logic here.

Questions about studying at EDU Effective? Get in touch or visit the FAQ.

Sources

Johnston et al. (2026) “The Shift to Agentic AI: Evidence from Codex” – OpenAI, Columbia Business School, Wharton School of the University of Pennsylvania, Duke University Fuqua School of Business.