Model choice + agents + prompt flow + evaluation + deployment. Data + AI + Digital + App Innovation Solutions Partner designations.
Last updated by Errin O'Connor, Founder & Chief AI Architect, EPC Group
Azure AI Foundry is Microsoft's enterprise developer platform for custom AI applications with model choice (GPT-4 + Llama + Mistral + Claude + open-source) + Prompt Flow + evaluation + Custom Engine Agents + Content Safety + fine-tuning. Foundry vs Copilot Studio: full-code + model choice + any channel vs low-code + Copilot models + M365 primary. Cost: Azure consumption per token + compute + storage (Microsoft consumption pricing; a mid-size agent example is in the FAQ). Six use cases: customer-facing products, complex internal agents, custom domain models, multi-modal, compliance-required, high-volume. EPC Group Foundry portfolio: Discovery, Custom Agent, Foundation, Fine-Tuning and a monthly Retainer — each a fixed fee set at scoping.
Azure AI Foundry is Microsoft's enterprise developer platform for building custom AI applications with model choice + tools + evaluation + deployment. Formerly Azure AI Studio + Azure OpenAI Service, unified in 2024. Six capabilities: (1) Model catalog — GPT-4/GPT-4o/o1 via Azure OpenAI + Meta Llama + Mistral + Claude via Bedrock + open-source. (2) Prompt Flow — visual workflow designer for LLM apps. (3) Evaluation — response quality + safety + performance metrics. (4) Agent orchestration — Custom Engine Agents. (5) Content Safety — moderation + content filters. (6) Fine-tuning + distillation. Deployment: any channel (websites, mobile, custom SaaS, embedded).
Two build-your-own-AI platforms with different orientations. Copilot Studio: (1) Low-code visual builder. (2) Tightly integrated with M365 + Teams. (3) Deployment primarily to Teams + Microsoft 365 apps. (4) Best for citizen developers + business analysts. (5) Fixed message-consumption pricing. Azure AI Foundry: (1) Full-code developer platform. (2) Model choice (Azure OpenAI + Meta + Mistral + Claude + open-source). (3) Any deployment target. (4) Best for developer teams. (5) Consumption pricing (compute + tokens + storage). Use both: Copilot Studio for M365 internal agents, Azure AI Foundry for customer-facing AI products or complex custom logic.
Custom Engine Agents are full-code agents built on Azure AI Foundry with any underlying model, deployable to any channel including Microsoft 365 (via declarative agent + Custom Engine backend). Comparison to Declarative Agents: (1) Declarative — YAML manifest, Microsoft Copilot models, M365 deployment. (2) Custom Engine — full code, any model, any channel. Use Custom Engine when: (1) Need model choice (not just Copilot). (2) Complex agent logic beyond Copilot Studio capabilities. (3) Deployment outside M365. (4) Advanced tool integration (multi-step reasoning, complex state management). (5) Need response streaming or custom telemetry.
Azure AI Foundry cost model — consumption across: (1) Model inference — per-token pricing varies by model (GPT-4o ~$2.50/M input tokens, $10/M output; Claude Sonnet 4 via Bedrock similar). (2) Fine-tuning — GPU compute + storage during training + hosting fee. (3) Prompt Flow compute — serverless or dedicated compute for agent runtime. (4) Storage — training data + prompt logs + evaluation results. (5) Content Safety — per-request pricing. Enterprise budgeting: typical Azure AI Foundry deployment for mid-size agent (1M user interactions/month) costs $10K-$50K/month total. Larger deployments scale linearly.
Prompt Flow is Azure AI Foundry's visual workflow designer for LLM apps. Capabilities: (1) Drag-drop pipeline builder — connect LLM calls + tools + data + logic. (2) Prompt engineering + versioning. (3) Batch evaluation — run test suites + score responses. (4) A/B testing between prompt variants. (5) Deployment to production endpoints. (6) Monitoring in production. Use cases: (1) Complex multi-step agents. (2) RAG pipelines. (3) Multi-model routing. (4) Structured output pipelines. Prompt Flow reduces custom code for common LLM app patterns.
Six proven enterprise use cases: (1) Customer-facing AI products — chatbots on public websites, in mobile apps, embedded in SaaS. (2) Complex internal agents — beyond Copilot Studio complexity (multi-source RAG, complex state, high-volume). (3) Custom domain models — fine-tuned Azure OpenAI models for legal, medical, financial verticals. (4) Multi-modal AI — vision + audio + text combined workflows. (5) Compliance-required custom AI — where Copilot governance model does not fit. (6) High-volume programmatic AI — bulk document processing, translation pipelines, content moderation.
EPC Group Azure AI Foundry portfolio: (1) Foundry Discovery + Design (4 weeks) — use case validation + model selection + architecture + fixed-fee proposal. (2) Custom Agent Build (12-24 weeks) — full-code Custom Engine Agent + multi-source RAG + evaluation + deployment. (3) Enterprise Foundry Foundation (12 weeks) — Foundry hub + projects + RBAC + governance + monitoring baseline. (4) Fine-Tuning Program (12-20 weeks) — domain fine-tuning + evaluation + production deployment. (5) Ongoing Foundry Engineering Retainer — monthly, scoped to the estate. Each engagement is a fixed fee quoted after a scoping call. All led by senior AI engineering architect. Solutions Partner: Data + AI (Azure) + Digital + App Innovation (Azure).
4-week fixed fee: use case + model selection + architecture. Call (888) 381-9725.
Monday-Friday, 8 AM - 7 PM CT
We respond to all inquiries within one business day
AI assistant — not human