You can talk to us before requirements are fixed.
Guide: problem framing from ¥300,000 (about 2 weeks), then a monthly quote|See pricing
Forward Deployed Engineering
From deciding what to build to running it
A Forward Deployed Engineer works inside your organisation. Rather than building to a fixed spec, we listen to the people doing the work, agree on what success looks like, build the generative AI or business system ourselves, and stay until it is in use. It suits new businesses without fixed requirements and AI work where specs depend on measured accuracy.
Common Situations
- You ran a generative AI PoC but have nothing to decide whether to go to production
- Nobody in-house can design AI systems, so you cannot tell a vendor what to build
- Your current vendor builds only what is specified, and the system has become a black box
- A legacy system has no documentation, so nobody dares to change it
- A new business has no fixed requirements yet, so fixed-price quotes are impossible
- You cannot produce the schedule and risk material leadership needs to approve investment
How an FDE Works
Start from the problem and KPIs
Before "what to build" we agree on "what success looks like" — the purpose and the numbers to measure — and work back to what to build and in what order.
Work inside your team
We join your meetings and chat, interview the people doing the work, and sit where decisions are made — preparing the information each decision needs.
Build it ourselves
Not just advice: our engineers implement generative AI, RAG, AI agents and business systems, and put them into operation.
Prepare the next decision
We do not stop at delivery: we provide the numbers and options to decide whether to continue, expand or stop — including material for leadership.
We develop with AI tools such as Claude Code and Cursor, keeping specs and design decisions as documents in the repository so both AI and the next engineer can read them.
Where FDE Fits
Launching a new business or service
- ・Requirements are not fixed yet
- ・You want to go PoC → MVP → full build
- ・You can set a milestone like "MVP by month X"
Generative AI and AI agents
- ・RAG, AI agents, OCR, image analysis
- ・Specs cannot be fixed until accuracy is tested
- ・You need cycles of test, improve and implement
A second opinion on your current vendor
- ・Development is slow and proposals never come
- ・The specification has become a black box
- ・You want to sort out roles before replacing anyone
Changing an undocumented legacy system
- ・No specification and the original authors have left
- ・We read the whole codebase with AI and write the spec before changing it
- ・You want the next person to be able to read it
* If requirements are fixed and you only need extra hands, standard system development will usually cost less.
Examples
Client names are withheld; only the industry is shown.
An AI prototype for an internal new-business review
- Starting point
- Passing an internal new-business review needed a working AI prototype people could try, not slides.
- What we did
- We shaped the experience from the owner’s concept, built and ran an access-controlled test environment, and agreed the next round of changes from trial feedback.
- Where it stands
- Delivered about two weeks after signing; now in cycles of internal trials and revisions.
Adding AI dialogue to a service for schools
- Starting point
- An AI dialogue prototyped in a no-code tool had to become production-ready, and how student data would reach the AI was undecided.
- What we did
- We documented the operations, moved the AI logic into code on cloud infrastructure, defined a layer that masks personal data before it reaches the AI, and sorted responsibilities with other vendors.
- Where it stands
- In development.
Requirements for an AI chatbot and integrations
- Starting point
- The platform had a request for proposal, but the AI chatbot part had no settled scope.
- What we did
- As a partner to the prime contractor, we owned requirements for the chatbot and integrations, working in their issue tracker and weekly meetings.
- Where it stands
- Requirements agreed; now in design.
A portal that makes care facilities easier to find
- Starting point
- Care facility availability was hard to find locally, and the budget assumptions changed mid-way.
- What we did
- We reshaped the scope into a minimum build plus optional add-ons, created initial facility data from public open data, and built map search and facility-side sign-up.
- Where it stands
- Preparing for launch.
Choosing the Contract Type
We match the contract to how uncertain the work is, using two kinds of quasi-mandate and fixed-price contracts, and can switch along the way.
| Quasi-mandate (time-based) | Quasi-mandate (outcome-based) | Fixed-price contract | |
|---|---|---|---|
| Basis of payment | Time and effort spent | Delivery of agreed outcomes | Delivery of the finished product |
| Obligation to complete | No | No (depends on how outcomes are defined) | Yes |
| Liability for non-conformity | No | No | Yes |
| Best for | Ongoing development with moving requirements, joining an existing team, operations | PoCs, MVPs and AI accuracy work with a clear goal per stage | Work with fixed requirements and deliverables |
* The two types of quasi-mandate follow the Japanese Civil Code as amended in April 2020 (Articles 648(3) and 648-2). What counts as an outcome is set in each contract.
How We Start
Problem framing (about 2 weeks)
Through interviews and a review of existing material and systems, we set the problem, the KPIs and what to test first. If you stop here, you keep everything we organised.
Validation (PoC)
We build small and test the most uncertain part first. For AI, we set up a way to measure accuracy on your real data before anything else.
Build and run
We reset the scope based on the results, build for production, and keep improving after release based on how it is used.
Next decision
Using the numbers, we decide together whether to expand, move to the next process, or stop. If you take it in-house, we leave it in a form you can take over.
How costs work
Problem framing starts from ¥300,000 (about two weeks). After that, we quote a monthly fee based on how many days a week we work with you. Stages with a clear endpoint, such as a PoC, can be paid on outcomes.
Combined with training
If you want to build in-house AI-driven development skills, you can combine this with our separate AI-Driven Development / FDE Training. Development and training are contracted separately.
Related Services
Pricing guide
Prices are pre-tax guides and vary with scope, integrations, and post-launch operations. Formal quotes are free.
Related work
Projects close to this service. Those with challenge, scope, and stack described come first.
Case study株式会社Nature Innovation Group
傘レンタルアプリ「アイカサ」
A LINE app, rental-spot hardware, and payments — launched as one system
Scope|デザイン / コーディング / システム開発
Case study学校法人松本学園（広島桜が丘高等学校・広島県瀬戸内高等学校）
学食モバイルオーダー
Students pre-order and pay on their phones, easing cafeteria queues and cash handling
Scope|デザイン / コーディング / システム開発
CHIeru株式会社
英語学習サービス「トイテイク」
Scope|デザイン / コーディング / システム開発
Forward Deployed Engineering
How is FDE different from staffing (SES)?
Staffing takes on defined tasks by the hour. An FDE starts from deciding what to build and stays until what we built is actually used. Both are usually quasi-mandate contracts, but the scope of responsibility differs. Under a quasi-mandate you cannot direct our engineers directly, so instructions go through our lead.
Can we start without fixed requirements?
Yes. In the first two weeks or so we agree on the problem, the KPIs and what to test first. Projects without fixed requirements are where FDE fits best.
How do you choose the contract type?
Time-based quasi-mandate for ongoing work with moving requirements, outcome-based quasi-mandate for stages with a clear goal such as PoCs and MVPs, and fixed-price contracts when requirements and deliverables are set. We can switch along the way and will propose one after problem framing.
Can you work alongside our current vendor?
Yes. We do not assume replacing your vendor. We first sort out roles, and can start by taking on what is missing — design, validation or decision material.
Can you work on-site?
We work online by default and visit for interviews, regular meetings and anything that needs to be in person. If you need full-time on-site presence, tell us the period and frequency and we will discuss it.
What does it cost?
Problem framing starts from ¥300,000 (about two weeks). After that we quote a monthly fee based on how many days a week we work with you. Stages with a clear endpoint, such as a PoC, can be paid on outcomes instead.
You can talk to us before requirements are finished.
We can start by sorting what to do first. We run businesses ourselves, so the conversation stays about labor and ops. Consultation and estimates are free.
Online meetings / reply by next business day / free consult & estimate