AI Agents in the Workplace 2026: What Changes Now and What Stays Human

Japan’s working-age population is shrinking by about 600,000 people a year. Meanwhile contract review, due-diligence summaries and regulatory filings, the work that used to train junior associates, now goes to AI systems that do it faster and at a fraction of the cost. Japan has had more than 1.3 job openings for every applicant for three years, so the firms doing this are mostly filling seats they could not staff anyway.

What an Agent Does

An AI agent takes an objective, breaks it into subtasks, carries them out with whatever tools and data it has, checks its own output and keeps going until the objective is met. It runs without anyone prompting each step.

In 2026 these systems are writing production code, run financial analyses that used to take teams of analysts, handle customer service across languages and time zones, draft legal briefs, produce marketing campaigns and schedule supply chains. Deloitte estimates that 40% of the tasks knowledge workers in OECD countries do today could be automated with AI that already exists, with current models rather than the next generation.

Deloitte’s exposure list now includes radiologists and paralegals alongside warehouse and assembly-line workers. A law degree, an MBA or a CFA charter used to guarantee a salary. Contract review and financial modeling were the tasks those credentials certified, and both are now on the list above.

The Japanese Numbers

Japan’s working-age population peaked at 87 million in 1995 and had fallen to 73 million by 2025. The National Institute of Population and Social Security Research projects 59 million by 2040, a loss of nearly a third in 45 years. The ratio of job openings to applicants has been above 1.3 for three years running. Healthcare, logistics, construction, hospitality and eldercare all have labor gaps that immigration alone will not fill.

So in Japan the question of AI and employment runs the other way. The Ministry of Economy, Trade and Industry (METI) has said as much: the goal is to lead automation rather than resist it, and to use AI to keep productivity, public services and living standards up in a society without enough people to do the work by hand.

That is also the reasoning behind Society 5.0, the national plan for a “super-smart society” that folds digital systems, IoT, robotics and AI into economic and social life. The term dates from the Fifth Science and Technology Basic Plan in 2016, and ministries now budget against it.

Where It Has Already Happened

Banking went first. JPMorgan’s COiN platform processes in seconds the commercial loan agreements that once took 360,000 hours of legal review a year. Goldman Sachs cut its equity trading desk from 600 traders to two, with AI systems behind them. In Japan, Nomura, MUFG and SBI are using AI in risk assessment, fraud detection and customer advisory, mainly to absorb volume their understaffed teams cannot.

Law firms followed. Contract-analysis platforms such as Harvey and CoCounsel do document review, case research and regulatory compliance at a scale that used to need floors of associates. On a cross-border transaction, one system can read the regulatory requirements of dozens of jurisdictions at the same time, work that used to be split across local counsel in each country.

In medicine, AI diagnostic systems now match or beat specialists at spotting conditions from diabetic retinopathy to early-stage cancers. Japan, with the oldest population in the world, is using AI for triage, paperwork and remote patient monitoring so clinicians can spend their hours on patients.

McKinsey, BCG and Bain each built internal AI platforms for research, data analysis and slide production, the work that took 60–70% of a junior consultant’s week. The firms now sell judgment and relationships with fewer people producing more output at higher margins.

Schools are later to this. Adaptive learning platforms tailor instruction to each student in a way no teacher can across a class of thirty. Japan is short of teachers in rural prefectures and is piloting AI-assisted classes there, so that one educator can handle a larger group while each student still gets work pitched at their own level.

Three Forecasts

One camp says AI makes people more productive without making them unnecessary: one lawyer with AI tools does the work of five, a developer with a copilot writes code ten times faster, and wages and new kinds of work follow. The history is on its side so far. Bank teller numbers rose for two decades after the ATM arrived. Spreadsheets did not end accounting.

A second camp says AI agents break that pattern, because earlier technologies amplified one narrow human skill while agents reproduce general cognitive labor. Once a system can reason, plan, write, code, analyze and communicate across most knowledge domains, the “new jobs” argument has less to stand on. Oxford Economics puts the number of full-time jobs worldwide exposed to AI automation by 2030 at up to 300 million.

A third goes further and asks what employment is for if machines can do most productive labor. Sam Altman has put $375 million into Universal Basic Income research. Andrew Yang ran a presidential campaign on it. Finland, Kenya and several U.S. cities have run or are running UBI pilots.

Where People Are Still Needed

An agent told to reduce a hospital’s costs will reduce them; deciding what it is not allowed to cut is still done by people. The same goes for accountability. A court, a regulator or a board wants a named person behind an AI-generated diagnosis or trade.

Policy

School and university systems were built to produce standardized knowledge workers, which is now the category most exposed to automation. Japan has expanded STEAM education and put programming into the national curriculum. Both are moving more slowly than the software.

The shelf life of professional skills is shrinking from decades to years. Singapore’s SkillsFuture program gives every citizen credits for lifelong learning. Japan’s Human Resources Development programs are growing, but the gap between how fast AI is deployed and how fast people retrain is still widening.

Governments will also have to pay people who lose work in the transition. The options on the table are UBI, a negative income tax, expanded public services, or designs nobody has tried yet. Each has to keep people secure without removing the reason to work.

The Next Five Years

The software is moving faster than labor law, school curricula or pension systems can be amended. Whether AI agents end up spreading prosperity or concentrating it will be settled largely by decisions executives, policymakers, educators and technologists take before 2031, and Japan, which cannot wait for its workforce to recover, is taking them first.

Join the Conversation

On April 26, 2026, the Tech for Impact Summit brings senior executives, policymakers and technologists to Tokyo Garden Terrace Kioi Conference under the theme “Beyond Boundaries: Building 2050 Together.” The future of work is one of the sessions.

Confirmed speakers include Taro Kono (former Minister of Digital Affairs), Yoshito Hori (GLOBIS), Charles Hoskinson (Cardano), Kathy Matsui (MPower Partners), Ken Suzuki (SmartNews), Jesper Koll (Monex Group), Sota Watanabe (Astar/Startale) and Hiroshi Aoi (Marui Group).

The session is for executives reorganizing a workforce around AI, for the companies building the agents, and for the policy institutions designing what comes after.

Explore partnership and membership opportunities →

Watch highlights from previous summits: youtu.be/ujy7ZXflrt4

The Tech for Impact Summit is an invitation-only executive gathering taking place April 26, 2026, in Tokyo as a partner event of SusHi Tech Tokyo. Learn more at tech4impactsummit.com.