AI Systems Diagnostic | The Architect Lab
Systems Diagnostic·The Architect Lab

Find the AI system
your business can
actually survive.

Most companies don't need more AI tools. They need a controlled operating layer: the right agents, the right data boundaries, and a human approval system that stops the business becoming a sentient spreadsheet with legal exposure.

Select your business type, what is breaking, your data sensitivity and current tools. Your recommended agent stack updates on the right as you go.

0 / 4 inputs·Stack pending
01 — Business type
02 — What is breaking

Select business type first

03 — Data sensitivity
04 — Current tools
Recommended stack
0 / 4

Select a business type

Your agent stack, data controls and first build recommendation will appear here as you make selections on the left.

Model routing logic

Architecture, not fandom.

We don't name specific models. Models version, deprecate, and get replaced. What matters is matching task class to capability tier — and routing sensitive data away from shared inference endpoints.

Task type
Model class
Example providers
Strategic reasoning
Complex synthesis, research, long-form analysis, decision support
→Frontier reasoning model
OpenAIAnthropic
Internal knowledge search
Intranet, document retrieval, team memory, policy Q&A
→Enterprise workspace model
Gemini EnterpriseMicrosoft Copilot
High-volume automation
Classification, routing, extraction, tagging at scale
→Fast task model
OpenAIAnthropicGroq
Sensitive EU workflows
GDPR-bound data, legal, financial, health — no US routing
→Private / EU-controlled route
MistralPrivate cloudLocal modelEU
Coding / technical build
Software, scripts, APIs, automation pipelines
→Code-optimised model
AnthropicOpenAI
Agentic web & real-time tasks
Browser control, live data, scheduling, tool-use chains
→Tool-augmented agent model
OpenAIAnthropicCustom MCP
Local / air-gapped processing
Nothing leaves the device or network — sovereign requirement
→On-device or self-hosted model
OllamaLM StudioOpenClawEU
Privacy architecture

European data privacy
is not an add-on.
It is the foundation.

01
Standard SaaS AI
Low-risk marketing, brainstorming, public research, non-sensitive drafts.

Control

Low–Medium

You have AI tools. You do not have an AI system.
02
Enterprise AI with controls
Internal knowledge, approved team workflows, Microsoft or Google environments.

Control

Medium–High

Most companies have not configured their tenant settings, checked their DPA, or audited what connectors are accessing. The tool is on. The governance is not.
03
Private workflow stack
Client operations, campaign data, recruitment, sales, internal reporting.

Control

High

You control what data moves where. Requires someone who can build and maintain workflows. Audit trail is your responsibility.
04
Sovereign / local-first
HR, legal, finance, healthcare, IP-heavy, EU AI Act compliance, air-gapped requirements.

Control

Very High

EU-hosted does not mean no US subprocessor. Check the DPA and full subprocessor list — not just where the servers are.