日本語
The AI SDLC process at Solashi combines AI Copilot tools with professional engineers across 7 development stages. AI handles high-volume, repetitive tasks at each stage. Engineers review, adjust, and approve every output before the project moves forward. The result: 20-30% productivity gains, 100% quality governance, and locally air-gapped security across every project.
Most software delays do not happen in the code. They happen upstream – in unclear requirements, underestimated planning, and documentation written after delivery instead of alongside it. Solashi integrated AI into the development process to solve these upstream problems first.
If you are new to how Solashi structures software delivery, start with our standard software development process first. This article builds on that foundation – focusing specifically on how AI Copilot integrates into each stage and what that means for speed, quality, and governance.
What Is an AI SDLC Process?
An AI SDLC process is a software development lifecycle where AI tools assist engineers at each stage – from requirement analysis through release. AI does not replace engineers. Instead, it handles structured, repeatable tasks so engineers can focus on judgment, architecture, and client communication.
At Solashi, the AI SDLC process follows three rules at every stage. First, AI performs the first pass. Second, a named Solashi engineer reviews and approves the output. Third, nothing moves forward without human sign-off.
Why Solashi Built an AI SDLC Process
Traditional software development has three recurring failure points. Requirements get misunderstood early and corrected late. Planning takes longer than delivery. Documentation arrives after handover, when it is too late to be useful.
An AI SDLC process moves all three earlier in the timeline. Requirement gaps surface in Stage 1, not during sprint 3. Planning drafts are generated in minutes and reviewed by the PM and Tech Lead before any task is assigned. Documentation is generated from git commits and pull request summaries during development, not written separately at the end.
The productivity gains are measurable. Solashi’s AI SDLC process delivers 40–50% efficiency improvements. These come from automating WBS drafting, boilerplate generation, initial test authoring, and documentation tasks – work that previously consumed significant engineering time on every project.
The 7 Stages of Solashi’s AI SDLC Process
Stage 1 – Requirement Analysis
AI analyzes raw requirement documents and produces structured user stories, acceptance criteria, and an open questions log. The Tech Lead and BA review and approve the output before the project advances.
In Stage 1 of the AI SDLC process, AI reads raw client requirement documents. It extracts functional and non-functional requirements, proposes an initial module list, builds user stories with acceptance criteria, and compiles an open questions log for unresolved ambiguities.
The Solashi TL and BA then review everything AI has produced. They lead client meetings to resolve open questions. They define the final scope boundary based on real-world project constraints.
Deliverables: Approved Requirements Specification, Client Meeting Minutes.
Stage 2 – Architecture and Database Design
AI generates draft database schemas, ERD diagrams, and permission matrices. The Designer and Tech Lead refine and approve the final architecture before development begins.
In Stage 2, AI creates initial database schema drafts in SQL and DDL, draws ERD, sequence, and API flow diagrams in Mermaid syntax, and drafts the authorization permission matrix. For simple projects, Claude Design generates layout prototypes.
The Solashi Designer builds high-fidelity UI/UX mockups in Figma. The Tech Lead reviews the database design for indexing, security policies, and performance. Then the Tech Lead approves the final architecture blueprint.
Deliverables: Figma UI Design, Approved Database Design, System Architecture Specification.
Stage 3 – Project Planning
AI proposes a complete Work Breakdown Structure and pushes the approved plan to Jira or Redmine automatically. The PM and Tech Lead review and approve before any task is assigned.
In Stage 3 of the AI SDLC process, AI proposes a complete WBS with task estimations. It maps requirements to an Epic, Story, and Task hierarchy. It also prepares automation configurations to push approved issues directly to Redmine or Jira via API — with no manual ticket creation required.
The Tech Lead and PM review the proposed WBS. They adjust timelines, milestones, and task dependencies. After approval, the plan is pushed to the project tracker and tasks are assigned to the development team.
Deliverables: Approved Project WBS, Jira/Redmine Board Setup, Developer Task Allocation.
Stage 4 – Code Development and Testing
AI reads Figma designs, generates backend code stubs, and prepares initial unit tests. QA approves test cases before coding begins. Developers implement core logic with local AI checkers running throughout.
In Stage 4, AI reads UI blueprints from Figma via Figma MCP and implements frontend UI. It drafts OpenAPI Specification documents, generates backend stubs including DTOs, Repository, Service, Controllers, and Migrations, and prepares initial unit test files.
QA reviews and approves test case sheets before development starts. Developers implement core logic. Local AI checkers audit coding standards throughout the implementation.
Deliverables: OpenAPI Specification, API Code and UI Frame, Unit Test Suites, QA Approved Test Cases.
Stage 5 – Code Review and PR Approval
AI reviews every pull request automatically for bugs, security vulnerabilities, and code quality. Developers fix all comments. Tech Lead reviews logic and approves the merge.
In Stage 5, AI code review triggers automatically on every new pull request. It checks for bugs, convention errors, duplicate code, complexity issues, SQL injection risks, and hardcoded secrets. Inline suggestions appear directly within the repository PR.
The developer resolves all AI review comments. The Tech Lead then reviews logic, patterns, and code architecture manually. Only the Tech Lead merges the pull request into the main branch. Therefore, no untrusted code enters production.
Deliverables: Refactored Codebase, Merge and PR Approval.
Stage 6 – QA and Security Scanning
AI runs OWASP security scans, generates automated end-to-end test scripts, and captures test evidence. QA validates results manually and covers paths that automation cannot reach.
In Stage 6, AI triggers automated dependency, secret, and static OWASP scans. It generates end-to-end test scripts using Playwright or Cypress, based on client-approved test case sheets. It executes tests via Chrome DevTools MCP and captures screenshots, DOM trees, and log files as evidence.
QA validates all auto-generated results. It then performs manual testing on paths that are difficult to automate — live payment APIs, SMS OTP flows, and external device integrations. QA logs all bugs into Jira or Redmine and signs off on regression testing.
Deliverables: Auto E2E Scripts, Test Evidence, Security Scan Reports, Manual Test Logs, Regression Signoff.
Stage 7 – Release and Handover
AI compiles release notes from git commits and generates changelogs and migration playbooks automatically. Engineers approve all documentation and manage the formal client handover.
In Stage 7, AI scans git commits and pull request summaries to compile draft release notes. It generates system changelogs automatically, drafts migration playbooks for database updates, and refreshes API documentation dynamically.
Solashi engineers review and approve all generated documentation. They manage the production deployment pipeline and initiate formal project delivery to the client. Every project closes with a Closure Report and dual sign-off, so the client receives a system their team can own and operate independently. For more on how Solashi structures the handover milestone, see our standard software development process.
Deliverables: Production Build Sign-off, Formal Documentation, Client Handover Package.
Key Outcomes of the AI SDLC Process
Solashi’s AI SDLC process delivers three measurable outcomes across every project.
Productivity increases by 20–30%. This comes from automating WBS drafting, boilerplate generation, initial test authoring, and documentation – tasks that previously required significant manual engineering time.
Quality governance stays at 100%. Test cases are pre-approved by QA before coding begins. The WBS is verified by the PM and Tech Lead before any task is assigned. Code is merged only by Tech Leads after manual review. No stage advances without a named human approver.
Security stays locally air-gapped. Automatic static analysis and secret detection run at local dev loops before any code leaves the machine. This is particularly important for enterprise clients in regulated industries, where data residency and code confidentiality are non-negotiable requirements.
Who Benefits from an AI SDLC Process
An AI SDLC process is most valuable for enterprise clients who need both speed and accountability. Specifically, it benefits teams that have experienced delays caused by unclear requirements, planning that changed mid-sprint, or handover packages that arrived incomplete.
It is also well-suited to Japanese enterprise clients and other regulated markets where governance documentation, audit trails, and formal sign-off processes are standard requirements. Solashi’s AI SDLC process produces documentation at every stage — not as an afterthought, but as a built-in output.
Frequently Asked Questions About the AI SDLC Process
How does the AI SDLC process protect code quality?
Every pull request triggers an AI code review. The developer fixes all comments. The Tech Lead reviews manually and is the only person authorized to merge code into the main branch.
AI review covers bugs, security vulnerabilities, coding conventions, and complexity. However, AI review is a first pass — not a final gate. The Tech Lead’s manual review of logic and architecture is what closes the quality loop. Code that does not meet Solashi’s standards does not merge, regardless of what AI has flagged or cleared.
What happens if AI generates incorrect outputs in the AI SDLC process?
AI output is treated as a draft. A named Solashi engineer reviews and approves every output before the project advances to the next stage.
From Stage 1, the TL and BA review AI-parsed requirements and lead client meetings to resolve open questions. Then at Stage 3, the PM and Tech Lead challenge and adjust the proposed WBS. Finally, at Stage 5, the Tech Lead reviews code manually before merging. Because humans approve every transition, incorrect AI outputs are caught before they reach downstream stages.
How does Solashi protect client data security in the AI SDLC process?
AI processing runs in locally air-gapped environments. Automatic static analysis and secret detection prevent data or credentials from leaving the machine at the local dev loop stage.
This architecture means client data, business logic, and credentials stay within controlled environments throughout development. For enterprise clients with strict data residency requirements, this is a critical factor when evaluating offshore development partners.
Ready to start your software development process? Book a 20-minute consultation with Solashi and let us show you how we work.
