Only 3% require industry knowledge. We read 326 Japanese job postings for engineers who take AI into the field, and the real ask is to see it through until people use it

One answer is a job whose postings are surging worldwide: the FDE, or forward deployed engineer. FDEs embed with customers and build AI into their actual operations until it works.

Japan saw a wave of these postings in 2026. But counting them runs into a naming problem straight away. Some employers say "FDE". Many others hire for the same work as an "AI engineer" or "AI transformation engineer". And manufacturers and banks such as Aisin and Mizuho are hiring for it in-house.

Job titles alone don't tell you what the job is. So we at DATA WORLD read the full text of each posting and recorded what it hands the engineer and what it requires. We read 326 postings from 168 companies.

The biggest surprise: only 10 of the 326 postings (3%) list industry knowledge as a requirement. What seven in ten do ask for is one person who carries the work all the way, from the first interview to the point where people use it every day.

We counted the phrases that come up in each posting's job description.

Source: DATA WORLD reading of 326 postings; one posting can list several.

Grouped into the three stages of the work, it looks like this.

228 of the 326 postings, or 70%, hand one person all three stages. Only 5% ask for the build alone, and another 5% mention a prototype but no production system.

The wording lines up behind the same point. The FDE posting at SB OAI Japan, the SoftBank and OpenAI joint venture, says the job is to build a production system that takes root in the business and keeps delivering results, not just a PoC. Logistics software firm Hacobu is blunter: the job is not done when the report is delivered, but when the site changes and the numbers move.

Generative AI has made prototypes easy. That is exactly why employers now pay for people who can carry a tool through each company's messy workflows and data until it is used every day.

The "FDE" title doesn't change much

Among postings with "FDE" in the title, 77% include all three stages. It's 58% for tech firms using other titles and 73% for in-house postings at non-tech companies. How far the job goes tells you more than what it is called.

There is a fair critique from the US: without a way to feed what engineers learn on site back into the product, the FDE is just consulting with a better title. Among the tech-firm postings we read, 52% mention product feedback. The other half may be closer to custom project work, which is worth checking before you apply.

The most common requirement is production coding in Python or similar (70%), followed by client-facing requirements work (46%).

What sits below that is the surprise. Only 37% require hands-on LLM or gen-AI work, and only 3% (10 postings) require industry knowledge. Most of those ten come from consultancies working on SAP, accounting or banking systems, and from overseas firms.

So a job that sends you into factories doesn't ask for factory experience. Domain knowledge is treated as something you pick up on the job. Preferred Networks, one of Japan's leading AI companies, puts "Learn or Die" in its list of values for the role and expects engineers to keep learning each client's domain. Industry experience appears as a "nice to have" in 21% of postings.

Where a minimum is stated, the median requirement is three years of experience. Half the postings state none, and some firms run junior tracks for people early in their careers.

An "FDE" title signals client-facing work

One clear difference shows up by group. Client-facing requirements work is mandatory in 68% of postings with "FDE" in the title, against 22% of tech-firm postings under other titles. The FDE label means you are expected to stand in front of the customer, not just write code.

The personality sections repeat the same words. The most common are ownership (35 postings) and getting deep into the customer's operations (35). Others include "see it through until it delivers", "gritty", and "unafraid of ambiguity".

The most common thing to build is AI agents and workflow automation (60%): AI that takes steps on a person's behalf and works through company systems to get a task done. Next is retrieval over company documents and past cases, or RAG (37%).

The settings named in the postings are full of work that has long run on paper, spreadsheets and experience.

Source: company job postings, checked October 17, 2026.

Drawings, dispatch boards, approval memos: work that is hard to see from outside the company. To bring AI into it, you first have to go and watch how it actually runs. That, in our reading, is why "get inside the customer's operations" keeps appearing in these postings.

Only 7% are on-site placements

Going into the field rarely means living at the client. Just 7% of postings mention an on-site placement, and 17% mention visits or travel. Some, like Preferred Networks, put going deep into the client's site and building there front and center; others state outright that the job is not a client placement. Go and look, take it back and build, return to check: that loop is the basic shape of the work.

Most of the postings so far come from tech firms: AI companies and consultancies. But going through the careers pages of manufacturers, banks and utilities one by one, we found at least 41 companies with 64 postings hiring for the same work in-house, under other names.

Aisin, the Toyota group auto parts maker, had the most (6), followed by Honda and Mizuho (4 each). Others include Toyota, Denso, Sumitomo Mitsui Banking, Resona, Dai-ichi Life, INPEX, Tokyo Gas, JR West, Sojitz, Sekisui House, Chugai Pharmaceutical and Kirin. Their titles are "AI engineer", "generative AI engineer" or "AI transformation engineer". The word FDE barely appears.

The content differs in a clear way. 94% of in-house postings include operations and adoption, while sales (9%) and product feedback (6%) barely appear. The weight is not on building something to sell, but on getting colleagues to keep using what was built.

Honda's posting, for example, goes beyond building a company-wide AI platform to interviewing departments, tracking usage and cost, and training staff. Kirin pitches the role as a chance to build the prototype yourself and stay until the sales team's way of working changes.

In-house postings list industry experience as a plus more often (53%, against 13% for tech firms). Even so, none of the 64 makes it mandatory.

This matters because of how Japan's tech talent is distributed. According to IPA, Japan's state IT agency, 73.6% of Japan's IT workers are at IT companies, against 35.1% in the US. Companies that own the field have few people who can work with AI, so FDEs come in from outside. Against that backdrop, manufacturers and banks have started hiring their own.

A Japanese government estimate shows why this will grow. By 2040 Japan will need 4.98 million people who can put AI and robotics to work, but on current trends will have 1.72 million, a shortfall of 3.26 million.

Of the 168 postings with a pay range, 134, or 80%, top out at ¥10 million or more. Sixteen reach ¥20 million and the highest is ¥30 million (about $200,000). The bottom of the range sits around ¥7 million at the median. Among postings with "FDE" in the title, more than 90% top out at ¥10 million or more. In-house postings at non-tech companies hold up too, at 80%.

The thin spot: measuring the impact

One gap stood out. Only 23% of postings ask the engineer to measure the effect of the AI in numbers, through ROI or KPIs. Among in-house postings it is 16%.

Employers hand over the build and the rollout, but rarely the job of showing how much faster the work got or how much money it saved. A few, like ANDPAD's Deployment Strategist role, include proving the value in numbers. They are still the minority.

EDITOR'S TAKE ── Kaito Ogasawara's view

Not who can build AI, but who stays until it's used

Reading 326 postings, I came away convinced that an engineer's value in the AI era is shifting from being able to build, to being able to carry it all the way into use.

FDE, AI engineer, AI transformation engineer. The titles vary, but seven in ten postings want the same thing: one person who carries the work from the first interview to everyday use.

If you are looking for this kind of job, searching by title will mislead you. Check whether the job description goes as far as operations and adoption.

Only 3% require industry knowledge. Manufacturing, finance and logistics are hard to see from outside, and almost nobody arrives knowing them.

That means the door is open to engineers from outside the industry. If you can write code, listen, and turn what you hear into something that works, you can walk into a factory or a bank.

Only 23% of postings include impact measurement. Without knowing how much the work changed, a company can't decide on the next investment or a company-wide rollout.

As people who work with data, we think the engineers who can own that measurement will be the most sought after next. DATA WORLD will keep re-reading this data as the market moves.