ARIS VIETNAM today announces the Hybrid Delivery Model (HDM), a software delivery model that combines engineers and AI agents, with three levels to weigh together with each client on a per-project basis.
HDM is not a default applied across the board. The level is agreed with the client during the analysis and contracting phases and written into the statement of work. Where a project requires no AI at all, ARIS VIETNAM delivers entirely by engineers (HDM-0).
Why an opt-in model
Over the past two years, trials of AI coding tools have moved from line-by-line suggestions to AI agents: tools that can read a codebase, edit multiple files and run tests on their own, with the goal of raising productivity for clients.
But the right degree differs from project to project. An internal system that needs to ship quickly and a device-control application with safety, security and performance constraints are different problems. Applying one approach to both gets it wrong at both ends: too slow where speed matters, too loose where rigor matters.
So ARIS VIETNAM, with its business analysts, senior engineers and its AI Agent / local AI research group, built a model that combines AI agents and people, named the Hybrid Delivery Model (HDM). Depending on the goals of each project, ARIS VIETNAM proposes the level that fits.
The three levels of the Hybrid Delivery Model
| Item | HDM-0 · Manual | HDM-1 · Assisted | HDM-2 · Accelerated |
|---|---|---|---|
| AI agent involvement | None | Limited scope | Extended scope with control gates |
| Scaffolding and boilerplate | Engineer | AI drafts, engineer reviews | AI drafts, engineer reviews |
| Core business logic | Engineer | Engineer | Engineer |
| Test case authoring | Engineer | AI drafts, engineer reviews | AI drafts, engineer reviews |
| Mechanical refactoring, version upgrades | Engineer | Engineer | AI within a closed scope, engineer signs off |
| Technical documentation | Engineer | AI drafts, engineer edits | AI drafts, engineer edits |
| Architecture design | Engineer | Engineer | Engineer |
| Merge approval | Engineer | Engineer | Engineer |
| Best suited for | Device-control software, safety-constrained systems, projects with a no-AI policy | Most enterprise projects | System transitions, platform upgrades, projects with a large mechanical workload |
Four controls applied at every level
1. Approved tool list. Each project has a list of permitted AI tools, agreed at the start. Tools outside the list are not used on the project's source code.
2. Data configuration. Tools are configured so that project data is not used to train models. For clients with stricter requirements, ARIS agrees dedicated terms in the contract.
3. Named approver. Every change to the main branch is approved and signed off by a named ARIS engineer, regardless of which parts were drafted by AI. There is no automatic merge.
4. Audit log. The scope and timing of AI agent use are recorded for periodic review and made available to the client on request for due diligence.
One principle runs through all three levels: AI shortens drafting time, not review time. ARIS's quality commitments and SLAs do not change with the HDM level a project chooses.
How the level is chosen
- Consulting: ARIS presents the three levels against the client's system characteristics and security requirements.
- Contracting: the HDM level and the approved tool list are written into the statement of work.
- During the project: the chosen level does not change without written agreement from both sides.
- Acceptance: the client may request the audit log as part of the handover package.
Clients do not have to pick one level for an entire contract. A single project can apply levels flexibly, work package by work package.
Background
On 31 March 2026, Japan's Ministry of Internal Affairs and Communications (MIC) and Ministry of Economy, Trade and Industry (METI) published version 1.2 of the AI Guidelines for Business. This version adds definitions of AI agents and physical AI to the main text and expands the appendix with agent risk scenarios, such as unintended actions (placing orders, deleting files) or being manipulated through connections to external systems.
Against those risks, the guidelines set out three measures worth noting: restrict the tools and systems an AI is connected to through an allow-list; build a mechanism for human judgment to intervene, with decisions organized by their level of importance; and record and keep logs of the processing so that it can be checked and traced periodically.
These three measures map directly to HDM's approved tool list, named-approver control and audit log. ARIS used the guidelines as the reference framework when designing the model.
Availability
The Hybrid Delivery Model applies to projects starting from September 2026. For projects already in progress, ARIS discusses with the client before applying any change.
Not sure which level fits your project? ARIS VIETNAM offers a working session to review your system characteristics and security requirements and agree on the right HDM level. Contact ARIS VIETNAM
About ARIS VIETNAM
ARIS VIETNAM is an offshore software development company and Odoo partner, with teams in Ho Chi Minh City and Hanoi and an office in Tokyo. Services: DX consulting, custom software development, Odoo ERP, QA and test automation, cloud and DevSecOps, security and OT security, and SLA-based operations.
References
- AI Guidelines for Business (AI事業者ガイドライン) Ver1.2: publication page, Ministry of Economy, Trade and Industry (METI)
- AI Guidelines for Business: listing page, Ministry of Internal Affairs and Communications (MIC)
- AI事業者ガイドライン（第1.2版）main document, PDF, Japanese, 31 March 2026
- AI Guidelines for Business Ver1.2, English provisional translation, PDF
