Generate Artifacts AI

The Azure port of gcp’s generate-artifacts-ai.md + the essential migration-plan output of gcp’s generate-ai.md, folded into one fragment (azure keeps Generate lean — generate-artifacts-* fragments, no separate plan file). Reads aws-design-ai.json, estimation-ai.json (for the budget seed), ai-workload-profile.json, preferences.json. Missing a REQUIRED input → STOP (“Missing required artifact: <file>. Complete the prior phase that produces it.”).

Step 0: Determine the artifact path

Read aws-design-ai.json → ai_architecture.code_migration.migration_path and preferences.json → ai_constraints.ai_framework. Language for the adapter comes from ai-workload-profile.json → integration.languages[0] (python→.py, javascript/typescript→ .js, go→.go, else .py).

Always emit (every path): ai-migration/setup_bedrock.sh, ai-migration/test_comparison.py (always Python), ai-migration/bedrock_monitoring.tf.

Mantle is the PRIMARY path for Azure. Azure OpenAI apps use the OpenAI SDK, so a same-model move keeps the SDK and changes only base URL, credential, model ID, and IAM — migrate_to_mantle.sh, not a rewritten adapter. The direct adapter path is the exception (cross-family swap, or a Converse feature the mantle models cannot serve), not the default.

Step 1M: migrate_to_mantle.sh (mantle path)

Shell script, dry-run by default with an --execute flag. Sets OPENAI_BASE_URL=https://bedrock-mantle.{region}.api.aws/openai/v1 (the /openai/v1 segment is load-bearing — bare /v1 404s), a Bedrock API key / token provider (NOT an OpenAI key), the openai.gpt-* model ID, and IAM needing bedrock-mantle:* (not bedrock:InvokeModel). Per-workload MAX_TOKENS from a lookup table, default 1024. No prompt changes when the source already uses responses.create; a Chat Completions source needs a reshape (flag it in the script comments).

Step 1: provider_adapter.{py,js,go} (direct / gpt-oss path)

Feature-flagged on AI_PROVIDER (values azure_openai | bedrock | shadow; default azure_openai). Emits methods gated on integration.capabilities_summary: text_generation→ generate, streaming→generate_stream, embeddings→embed. Bedrock side is boto3 Converse (BEDROCK_INFERENCE_PROFILE_ARN env, model-id fallback). Azure source side: the azure_openai branch calls the AzureOpenAI client (azure_endpoint, api_version) for Python / @azure/openai (or the openai SDK with azure config) for Node — NOT Vertex/@google-cloud/vertexai.

Step 2: test_comparison.py

Always Python. Runs the same prompts through the source (Azure OpenAI) and the Bedrock target and diffs outputs, so the user validates behavior before cutover.

Step 3: setup_bedrock.sh

Enables the chosen Bedrock model(s), sets up IAM, and (mantle path) the Bedrock API key.

Step 3B: gateway_config.{yaml,py,json} (gateway users)

Format by framework: llm_router → gateway_config.yaml (LiteLLM); framework → gateway_config.py; voice_platform → gateway_config.json; api_gateway → gateway_config.yaml. OpenRouter branches on code_migration.openrouter_path (same_model_mantle → run Step 1M; direct → adapter; litellm → yaml; keep_openrouter → json; absent → litellm).

Step 3C / 3E: agentic artifacts

Harness (migration_approach == harness): harness.json, deploy_harness.sh, incremental_migration.sh (only if harness_config.incremental_migration). Strands (== strands): strands_agents.py, deploy_strands.sh, bridge_retarget.py (only if strands_config.bridge_phase). Both port from gcp unchanged (AgentCore is the target, agnostic of source cloud).

Step 3D: eval artifacts (opt-in)

eval-prompts.jsonl, run-evaluation.sh.

Step 3F: bedrock_monitoring.tf (all paths)

Budget limit_amount = ceil(projected_bedrock_monthly * 1.5), min 10, emitted as a computed integer (seed projected_bedrock_monthly from estimation-ai.json → cost_comparison.projected_bedrock_monthly, default 50). Cost-anomaly monitor monitor_type = "DIMENSIONAL", monitor_dimension = "SERVICE". aws_bedrock_inference_profile requires the AWS provider >= 5.66.

Write generation-ai.json

rollback_plan.mechanism MUST be "feature_flag"; flag_name AI_PROVIDER; default_value azure_openai (the pre-cutover provider — flip to bedrock to cut over, back to azure_openai to roll back).

Write STARTUP_PROGRAMS.md

AWS Activate tiers, branching on preferences.json → startup_program_status (eligible_founders/eligible_portfolio/has_credits/unknown). Include the Generative AI Accelerator section ONLY when ai_monthly_spend is high AND agentic_profile.is_agentic. Content is AWS-only and ports from gcp unchanged.

Step: Self-Check

• Not done: AI_PROVIDER default is azure_openai everywhere (adapter flag + rollback default).
• Not done: The mantle path emitted migrate_to_mantle.sh and NO provider adapter; the direct/gpt-oss path emitted the adapter and NO mantle script.
• Not done: bedrock_monitoring.tf present on every path; budget is a computed integer ≥ 10.
• Not done: No secret VALUE in any emitted file; credentials are env-var references.
• Not done: No proprietary openai.gpt-* model ID paired with a Converse/bedrock-runtime path (mantle-only).
• Not done: Plugin attribution strings say “Azure-to-AWS”.
• Not done: generation-ai.json validates; rollback_plan.mechanism == "feature_flag".

Phase Completion

Do NOT update .phase-status.json — the parent generate.md handles phase completion after the main-window validation step.

_assert has no teeth. The self-check is prose the rw worker both produces and evaluates; CI binds it but never runs it. Judge correctness by the step-5 AI fixture oracle and a fresh-context run, never by “the artifacts were emitted.”

Status — build step 3 (AI route)

The AI Generate fragment. Wired into generate.md _fragments (_when aws-design-ai.json exists AND run_mode == decide_and_execute). Its inputs come from design-ai.md (step 2) and estimate-ai.md (step 3). Reachable end-to-end only once discover-app-code.md (step 4) produces ai-workload-profile.json for a non-IaC estate.