Eternal Inc. — Osaka, Japan

Start with the website.
Then make the work itself a system.

Your company has moved on. The website hasn't. First, a site that shows the company you are now. From there we use AI to turn everyday work into systems.

In most companies, "AI adoption" ends the moment the tools are handed out. We build something different: AI staff that carry out routine work — and hand every decision that matters to a person.

We test every system on our own company first, across Finance, Strategy and Sales. What fails is recorded alongside what works, and both feed straight back into the design.

What we bring to clients outside Japan is not a translated deck. It is the system we run on our own company — starting with one business process at a time.

AI Company

A company where an AI president and AI staff do the work.

At Eternal we use our own company as the testing ground. An AI president and AI staff receive the work, sort it, carry it out, and follow it until it is finished. Work that can be handed to AI goes to AI. People keep the decisions that matter, and the work only a person can do.

Owner / Person

Seiko Kubota, Representative Director

Decides the direction, and whether we go ahead or not.

AI President

Looks across everything, sorts what to do, sets the order, hands it to the AI staff, and follows it to the end.

Watches the moneyReads the flow of cash and supports the business with numbers.
Reaches the marketPublishes, measures the response, and corrects it.
BuildsWebsites, design, implementation — and the checks that prove them.
Turns work into systemsTakes the everyday work and makes it repeatable.

The boundary is set. AI researches, organises, drafts, checks, and keeps the record. Anything that cannot be undone — publishing, sending, paying — a person confirms before it happens. What does not work is written down too, and becomes the next rule.

What gets easier

Three jobs our AI staff already do for us.

Plain examples from our own company, as of 14 September 2026.

In trial

An inbox summary on your phone

A single daily message on LINE (the messaging app most used in Japan) is being tested. It sorts the last two days of email into three groups: written by a person, for information, and sales or newsletters. Everyone who wrote to you personally is listed by name and subject, so they are not lost among automated notices.

It only reads and labels. It never replies, archives, marks as read or deletes anything. Whether a reply is needed is always your decision, not the AI's.

Live

Unpaid invoices and double entries, flagged

Invoices are matched against incoming payments. Anything billed but not received is flagged once it passes its due date, and the same payment recorded twice is caught before it distorts the numbers.

Bank data still arrives as a CSV file that a person exports; the matching and flagging after that are automatic. An earlier version missed two of three real double entries in our own books, so we tightened the rule on 12 September 2026.

Live

Nothing becomes final without a person

When the AI finds something that needs a decision — an unrecognised payment, a possible duplicate — it does not fix it on its own. It places the item in a review queue, and a person approves or rejects it.

Our AI does not send email or messages outside the company on its own. This is built into the system, not left to good behaviour.

How it works

Three steps. None skipped.

Every one of our AI staff moves through the same sequence.

Collect
They gather the facts — bank records, payments, invoices, email — on a fixed schedule, without being asked.
Check
They point out what an owner cannot afford to miss: overdue payments, unrecorded transactions, and the shortfall against the monthly target.
Record
Only what a person has approved becomes a task. No task can be marked complete without supporting evidence.

Measured, not promised

What is running today

Figures measured 12–14 September 2026.

Every 15 minutes · while the founder's Mac is on

Cash-flow monitoring

Our finance checks run every 15 minutes while the founder's Mac is on. When it is off, a cloud job takes over, but in practice it runs only every few hours (6 runs in 24 hours, measured 14 September 2026).

678 automated tests passing

Automated test coverage

Across our four internal systems, 678 automated tests pass as of 13 September 2026.

Evidence required to close any task

No task closed without supporting evidence

An AI proposal becomes a task only after the CEO approves it. A task marked complete without evidence is rejected by the database itself — not by policy, but by design.

And what we have not yet built. Among our departments, only Finance runs scheduled checks. In Strategy and Sales, AI drafts and a person decides.

We would rather tell you that now than after you have signed.

How we build

Precision, by default.

Four rules we keep on every project.

Human approval remains required. None of our AI sends anything outside the company on its own.
No evidence, no completion. "It's done" is a claim, not a record.
Least privilege, always. AI staff sign in under their own dedicated accounts. None of them holds administrator rights.
Failures are not swept away. When the same process fails three times in a row, it escalates to a person.

Where we actually are

Every feature shows its current stage.

Each feature is clearly labeled with its development stage, so nothing here reads as larger than it is.

Live	AI President (runs the company: routes work to each AI department and brings only the decisions that need the CEO)
Testing	CONTROL TOWER (sees the company) · AI Finance (protects cash) · AI PR (takes our work to the market)

We are not selling a finished product. We work as the partner who builds it and corrects it in front of you.

The first step

One business process. One MVP.

The scope is fixed in writing before we begin. We do not redraw the line afterwards.

¥300,000 (tax included, approx. USD 2,000 as of Sept 2026)

Three companies only · paid in full up front

Included: Online discovery sessions with your team, problem definition, requirements definition, MVP design, MVP development, testing, supervised use in your day-to-day operations, feedback, and minor refinements within the agreed scope. All sessions are held online.
Not included: Enterprise core-system replacement, unlimited revisions or feature additions, large-scale external integrations, App Store release, rollout across multiple departments, large-scale CRM development, and production development beyond the agreed scope.

Founder

Why this company, and why now.

Only what we can substantiate about ourselves.

Tested on our own company

We test every system on our own company first, with AI staff working in Finance, Strategy and Sales. 678 automated tests pass as of 13 September 2026.

Founder — Seiko Kubota

Twenty years spent supporting people. Twenty-five years spent chasing the technology. AI is what finally joined them into one line of work.

More than ten years of one-to-one support and operational redesign since going independent in 2015 — years spent listening, on the front line, for where people get stuck and what they cannot yet put into words. He founded Eternal Inc. in March 2019.

Contact

Tell us the one process that hurts most.

We work in English and Japanese. As a rule, you will hear back from us by the next business day, Japan time.

Email info@kubota-seiko.com → 日本語サイトへ →

English replies are drafted by AI and reviewed and sent by Seiko Kubota, our founder, personally.

Eternal Inc. · Kyoei Bldg. 309, 2-8-11 Shibata, Kita-ku, Osaka 530-0012, Japan

Start with the website.Then make the work itself a system.