Contact

Give that systemits next chapter.

Technical debt.
No more putting it off.

  • The latest updates
    never get applied.

    You want to update the OS and the libraries. But nobody knows whether the old setup will still work, so even security fixes are left for later.

  • The developer has left.
    The spec lives in the code.

    The existing code has become a black box, and every fix breaks another feature. Nobody can check what a change will affect, so nobody touches it.

  • A rebuild is needed.
    But there is no budget.

    A conventional estimate is out of reach. You are asked to save with AI, yet there is no plan that also covers quality and migration.

A rebuild,
and what comes after.

“Couldn’t we do this in-house with AI?”
What Samurai brings is a way of building that combines spec-driven AI development, security by design and continuous updates.

Spec-driven AI development

AI development,
on our own workflow.

From writing the spec to implementation and pull requests, the work is standardized with agent skills we built ourselves. Cross-review by a different model and steps where a person decides are built in.

More on the development workflow

We generate the spec from business requirements, then implement against the agreed spec and acceptance conditions. Our skills define how specs are written, the coding rules and how pull requests are made. Different models, such as Claude Code and Codex, take the generating and reviewing roles and repeat flagging and fixing.

Hooks we built control the steps, and the points a person must review are set in the workflow. Work runs in parallel on Git worktrees, and changes verified by CI are merged.

Security by design

Security
as the design baseline.

OWASP ASVSSecurity requirements and verification standard
  1. Define as requirements
  2. Reflect in design
  3. Verify the build

Referring to OWASP ASVS, we turn the data and permissions involved into requirements for design and verification. Our experience of third-party penetration tests goes into the build.

How we handle security

We set out authentication, access control and data protection as requirements and check them in design and implementation review. Drawing on penetration tests by firms such as NRI Secure, LAC and SHIFT SECURITY, we also verify against expected attacks. Third-party penetration tests are planned to fit each project’s requirements.

CI/CD and serverless

The pipeline that keeps it updated,
built with the system.

  1. Update PRDependabot etc.
  2. CIAutomated tests
  3. AI + peopleReview & approve
  4. CDRelease

Updates are verified automatically, then checked by AI and people before release. We design serverless-first, to cut the burden of OS maintenance.

How updates and operations work

Dependabot and similar tools turn dependency updates into pull requests, and CI runs the automated tests. AI and engineers review the results and the change, and approved changes are released by CD. A change with a problem goes back to fixing and re-verification.

With serverless, maintenance of the underlying OS is left to the cloud provider, while we keep managing application dependencies and settings. The architecture and the extent of automation are decided to fit the constraints and operating conditions of the existing system.

A rebuild for referenceDigital credential issuing platform

An unstable beta was rebuilt by two people in about three months. A proprietary format was revised to W3C-compliant DID/VC, and the native app was replaced with web and a LINE mini app.

This rebuild included revising the existing implementation. The team size and period are the actual figures for this project.

How the service
comes in

A consultation starts with a look at the current system.
We sort out what needs doing now and what a rebuild will solve, then proceed in stages.

  1. Assessment and first measures

    We look at the current system and set out the security concerns and their priority. We check accounts, open ports, source addresses, logs and backups, and propose measures you can take right away, such as adding a WAF.

    Decided hereImmediate measures and priorities
  2. Research and requirements

    We study the existing code and operations, and sort features into those to carry over and those to retire. We confirm integrations and data migration conditions, and make the scope, architecture and cost concrete.

    Decided hereRequirements, migration plan, estimate
  3. Build, verify, migrate

    We implement with spec-driven AI development, CI and cross-review. Where needed, a third party runs a penetration test. We build through to the CD that verifies updates and releases them, then migrate.

    Confirmed hereAcceptance conditions and the go-live decision
  4. Operation and continuous updates

    Update pull requests are tested automatically and released after review by AI and people. We also handle issues and change requests that come up in operation.

    OngoingVerifying, approving and releasing updates

Frequently asked questions

Can we consult you without a specification?

We research from the existing code, screens, architecture information and how the work is actually done. Even if documents are missing, tell us the issues you know of. We agree together on the information and access needed.

Does everything have to be rebuilt?

Research sorts features into those to keep, to rebuild and to retire. We also consider integration with existing services and migration in stages, then decide the scope.

How are current operations and data migrated?

In requirements we confirm integrations, data volume and how long the system may stop, then plan the migration method and cutover steps. Data migration is verified against the current operating conditions before going live.

How is the cost decided?

Research and requirements run under a time-and-materials contract with a set period and cap. Development is estimated with the scope and effort shown, and the cost of running AI is itemized at cost. Operation is a monthly fee. The scope of assessment and third-party tests is confirmed before work starts.

Can you arrange third-party security testing?

Depending on the data handled and the standard required, we plan a penetration test by a third party. We agree the scope, timing and cost, and build the handling and re-check of findings into the development plan.

How do you handle source code and confidential information?

We use AI tools set not to train on input data, with a mechanism that controls what confidential information is sent out. Before contracting we confirm your security standards and state the handling of information and the terms of AI use in the master agreement.

Can you handle maintenance after release?

After go-live we set up operations that continue dependency updates, automated tests, checks by AI and engineers, and releases. The scope and contact method are agreed before operation begins.

First, tell us about
the issues you have now.

It can’t be updated. Nobody knows the spec. The budget doesn’t fit.
Starting from the current state, we will work out how to rebuild together.

Consult us about your system