TransformAI

Live Demo: https://ai-transformation.up.railway.app/

GitHub: https://github.com/chenyirui/ai-transformation-simulator

Pitch Deck: https://docs.google.com/presentation/d/13MMrgOvXWmsy4S0ag09Z8nPa9Y_dsnkE/edit?usp=sharing&ouid=102522715159586335966&rtpof=true&sd=true

Demo Video: https://drive.google.com/file/d/13cy1-JOeIhjb3o7rZMfBChlrkFVtGNpg/view?usp=sharing

TransformAI is a controllable multi-agent boardroom for traditional enterprises planning an AI transformation.

It is not a generic "AI transformation simulator" in the abstract. It is a productized decision game where the player acts as the internal transformation lead, assembles a stakeholder committee, chooses the first real workflow to change, survives board review, and ends with a concrete transformation report.

What It Does

Traditional companies do not usually fail at AI because they cannot buy a model. They fail because the first use case is vague, old systems are messy, employees do not trust the rollout, vendors create lock-in, ROI is unclear, and compliance arrives late.

TransformAI turns that ambiguity into a playable boardroom:

• choose the first AI transformation use case
• hear stakeholder agents argue from different incentives
• pick between SaaS, consultants, internal teams, process-level agents, or broad transformation
• handle budget, legacy systems, compliance, employee trust, and delivery risk
• pass monthly review stages
• generate a final AI transformation report and share card

中文简介

TransformAI 是一个面向传统企业 AI 转型的多 Agent 决策产品。

它不是泛泛的 "AI Transformation Simulator"。玩家扮演企业内部 AI 转型负责人，组建转型委员会，选择第一个真实业务流程，处理预算、旧系统、员工信任、合规和董事会压力，最后生成一份 AI 转型路径报告。

核心问题不是 "AI 能不能做"，而是：

• 先改哪个业务流程
• 买 SaaS、请咨询公司，还是自建 AI 能力
• 如何避免供应商锁定
• 如何让员工相信这是 Copilot，而不是替代通知
• 如何证明 ROI
• 如何留下审计日志和责任边界
• 如何避免转型变成 demo 剧场或 PPT 项目

Product Loop

The current build keeps the original game mechanics and flow intact:

1. Assemble the transformation committee.
2. Run a multi-agent roundtable.
3. Choose the first AI use case.
4. Survive weekly enterprise transformation incidents.
5. Pass monthly boss stages.
6. Finish with an AI transformation report.
7. Generate a vertical share image for the result.

Agent Roles

TransformAI uses stakeholder agents to expose enterprise tension:

• Board / CEO: mandate, narrative, and urgency
• CFO: budget, ROI, and cloud cost
• CTO / CIO: systems, integration, and maintainability
• HR: employee trust, training, and role anxiety
• Legal / Compliance: audit, data boundaries, and accountability
• Business Unit Lead: workflow reality and adoption
• AI Consultant: roadmap, readiness, and transformation theater risk
• AI Engineer: feasibility, evals, logs, and failure modes

The agents advise and argue. The user still makes the final decision.

Game Stages

Stage 1: First Use Case

Pick a real enterprise entry point, such as claims review, customer service, sales support, finance/legal automation, or internal knowledge workflows.

Stage 2: Strategy Tournament

Compare transformation routes:

• buy SaaS
• hire consultants
• build an internal AI team
• start with one process-level agent
• attempt broad AI transformation

Stage 3: Board Review

Defend the transformation plan in a multi-round board meeting. The board challenges scope, capability, adoption, team depth, and authorization terms.

Stage 4: Final Review

Resolve the accumulated risk and produce the final transformation outcome.

Outcomes

TransformAI can end in different enterprise futures:

• Controlled Transformation
• Internal Capability Built
• Human-in-the-loop Success
• Board Approval Won
• Demo Theater
• Consultant Black Hole
• Compliance Shutdown
• Vendor Lock-in
• Employee Rebellion
• AI Everywhere, Value Nowhere

Real-Case-Driven Events

TransformAI is inspired by public enterprise AI transformation cases and industry reports.

The game does not reproduce any single company one-to-one. Instead, it abstracts real case patterns into playable news, weekly events, agent conflicts, trade-offs, endings, and final report labels.

Real case patterns used in the current build include:

• Gartner's prediction that many GenAI projects will be abandoned after proof of concept because of poor data quality, inadequate risk controls, rising costs, or unclear business value.
• RAND's discussion of high AI project failure rates.
• Gartner and Reuters coverage of agentic AI project cancellation risk and agent washing.
• Air Canada's chatbot liability case.
• NYC MyCity chatbot giving incorrect or unlawful advice.
• McDonald's ending its IBM AI drive-thru ordering test.
• DPD disabling part of its AI chatbot after brand-safety issues.
• Samsung's reported ChatGPT data leakage concerns.
• CNET correcting many AI-written articles.
• Morgan Stanley's controlled internal knowledge AI.
• Klarna's customer service AI assistant.
• Telstra's employee augmentation through AskTelstra.
• Oracle Fusion Agentic Applications.
• OutSystems Agent Workbench enterprise workflow cases.
• Pega Blueprint for legacy modernization.
• IBM COBOL modernization and semantic equivalence testing.

The goal is to simulate the enterprise AI problems behind these cases: workflow integration, ROI, auditability, employee trust, legal liability, vendor lock-in, data governance, legacy modernization, and human-in-the-loop control.

真实案例驱动

TransformAI 不是直接复刻某一家公司的故事，而是把公开企业 AI 新闻、官方案例和行业报告中的共性问题，抽象成可玩的新闻、周事件、Agent 冲突和结局。

当前版本参考的真实模式包括：

• GenAI POC 后被放弃
• AI 项目失败率高
• Agentic AI 项目因为 ROI 不清、成本上升或风险控制不足被取消
• 客服 chatbot 造成法律责任
• 政府 chatbot 给出错误甚至违法建议
• AI 在真实现场环境中失败
• Shadow AI 带来数据泄露风险
• AI 自动生成内容需要人工审校
• 内部知识 AI 在受控边界内成功
• 客服 AI 带来的效率叙事和员工信任压力
• 员工增强比全自动替代更容易落地
• 企业软件正在变成 agentic apps
• 老系统现代化需要语义等价测试和人工验证

这些真实模式会变成游戏里的选择题：要不要追求速度？要不要保留人工确认？要不要先证明 ROI？要不要买供应商平台？要不要先做治理？每个选择都有代价。

Tech Surface

• Landing page: ai-manager/index.html
• Main game runtime: ai-manager/AI Manager.dc.html
• Lightweight runtime shell: ai-manager/app.js
• Core content: ai-manager/content.js
• Enterprise transformation override data: ai-manager/transform-db.js
• Agent server: ai-manager/agent_server.py
• Local agent fallback runtime: ai-manager/agent_runtime.py

Run Locally

./start.sh

The server defaults to port 8000.

Optional environment variables:

PORT=8000
OPENAI_MODEL=gpt-4.1-mini
OPENAI_API_KEY=<your-api-key>

If OPENAI_API_KEY is not set, Agent Mode still works through the bundled local structured agent runtime.

Open-Source Package

This repository includes the playable web prototype, bundled runtime data, local agent fallback, and share-image report flow. Internal planning notes, private todo files, source design documents, local logs, and API keys are intentionally excluded.

Demo Path

Recommended demo flow:

1. Start TransformAI.
2. Assemble the transformation committee.
3. Let the agents brainstorm transformation entry points.
4. Choose one use case.
5. Play through weekly incidents and monthly review stages.
6. Reach the final report.
7. Click "Create share image" / "生成分享图".

One-Line Pitch

TransformAI helps traditional enterprises preview AI transformation decisions in a multi-agent boardroom before spending real budget on systems, consultants, teams, and organizational change.

Current Status

This repository contains the open-source playable build of TransformAI.