Clarify AI Only

Produces the same preferences.json but with design_constraints limited to region and compliance, startup_constraints populated, and ai_constraints fully populated. Two progressive batches with an intermediate save. The Azure port of gcp’s clarify-ai-only.md: the cross-cloud framing and question mechanics port unchanged; the model catalog and provider wording are Azure-swapped (Azure OpenAI, no Gemini).

Step 0: Prior-run check

• preferences.json present → offer reuse vs start-fresh; on reuse skip to Step 3.
• preferences-draft.json present (Batch 1 done) → offer resume vs start-fresh; on resume load the draft, skip Batch 1, present Batch 2.
• No prior state → Step 1.

Step 1: Present the AI-only detection summary

AI-only migration detected. Your project has AI workloads but no infrastructure artifacts (no Terraform/Bicep/ARM). I’ll migrate your AI/LLM calls to AWS Bedrock while your infrastructure stays on Azure.

AI source: [summary.ai_source] · Models: [models[].model_id] · Capabilities: [integration.capabilities_summary where true] · Integration: [integration.pattern] via [integration.primary_sdk] · Gateway/router: [integration.gateway_type, or “None (direct SDK)”]

Step 1.5: Fast-path check

If the AI usage looks simple (single model, non-agentic, no multi-provider/multi-model routing): present ONLY Q1.5, Q2, Q3, Q4, Q11 (Q1 framework, Q5 model, Q6 capabilities are extracted; Q7–Q10 default). Q1.5 (compliance) and Q11 (Activate) are never dropped — always PRESENTED, never silently omitted. “Use defaults for the rest” applies documented defaults (compliance → ["unknown"] + a metadata.report_caveats[] entry, never a silent “none”; Activate → unknown). Then skip to Step 3. Otherwise continue to Step 1.75 → Step 2.

Step 1.75: Mini assumption sheet

Before asking anything, present a compact confirm-or-edit sheet: what discovery already answered (framework, primary model, input types — from the profile) and what will be assumed (usage volume Low, response speed Important, task complexity Moderate) with each assumption’s consequence. Rows resolved here are not re-asked; record chosen_by: "extracted" (detected) or "default" (assumed, sheet-confirmed); corrected rows become chosen_by: "user".

Step 2: Questions in two batches

The user can answer, skip individual questions (defaults applied), or say “use defaults for the rest” to apply defaults for all remaining and proceed.

Batch 1 — AI strategy & setup (Q1–Q5 + Q1.5)

Q1 — AI framework / orchestration (select all). Auto-detect from integration.pattern / gateway_type / frameworks; skip when definitive (a direct_sdk + empty frameworks + null gateway → ["direct"]). → ai_framework. Default ["direct"].

Q1.5 — Compliance / regulatory requirements (select all; never dropped). Compliance gates Bedrock regions/models/logging even though infrastructure stays on Azure — prompts and completions are processed on AWS the moment model calls move. Options None / SOC2-ISO27001 / PCI DSS / HIPAA / FedRAMP / GDPR / CCPA / “don’t know”. Impacts: HIPAA → BAA-eligible models, KMS-encrypted logs, us-east-1/us-west-2; FedRAMP → GovCloud (smaller catalog); GDPR → EU regions with geographic eu. inference profiles (global. forbidden), document the Azure-EU → AWS-EU transfer; PCI/SOC2/CCPA → CloudTrail + scoped IAM + retention. → design_constraints.compliance. Explicit “None” → ["none"] chosen_by: user. Skip/default → ["unknown"] chosen_by: default

• append “Compliance requirements were not confirmed by the user” to metadata.report_caveats[].

Q2 — What matters most? Quality (Sonnet/Opus) / Speed (Haiku, Nova) / Cost (Haiku, Nova Micro) / Special (→Q10) / Balanced (Sonnet). → ai_priority. Default "balanced".

Q3 — Monthly AI spend on Azure OpenAI / OpenAI? <$500 / $500-$2K / $2K-$10K / >$10K / don't know. → ai_monthly_spend. Default "$500-$2K".

Q4 — Cross-cloud API call concerns (unique to AI-only — infra stays on Azure while AI calls route to AWS): Latency critical (VPC endpoint, closest region) / Acceptable (standard endpoint, region by cost) / Egress-concerned (PrivateLink, egress analysis) / Test-first (phased parallel running). → ai_constraints.cross_cloud. Default "latency-acceptable".

Q5 — Current model in use? Establishes the baseline Bedrock recommendation. Skip when models[].model_id is populated with confidence ≥ 0.8 (chosen_by: extracted). Azure catalog (Azure OpenAI deployments — no Gemini):

Same-model rows first. GPT-5.6/5.5/5.4 run on Bedrock via bedrock-mantle (Responses, in-region only) — the case is AWS commitments/governance/residency, not savings, and not parity (~10% above OpenAI standard). For sources with no Bedrock equivalent, present a same-vendor upgrade AND a cross-family option. → ai_model_baseline.

After Batch 1: interpret answers, write preferences-draft.json (metadata.draft: true, migration_type: "ai-only", batches_completed: ["ai-strategy"]), then present Batch 2. On “use defaults for the rest” during Batch 1, skip the draft save (assembly happens same turn).

Batch 2 — Technical requirements (Q6–Q11)

Q6 — Input types (text / vision / audio-video). Skip when capabilities_summary is definitive. → ai_vision. Q7 — Monthly usage volume. Auto-resolve from openai-usage-profile.json when present (tiers <1M→low, 1–10M→medium, 10–100M→high, >100M→very_high). → ai_token_volume. Default "medium". Q8 — Response speed (critical / important / flexible). → ai_latency. Default "important". Q9 — Task complexity (simple / moderate / complex). → ai_complexity. Default "moderate". Q10 — Specialized features (function calling / ultra-long-context / extended-thinking / prompt-caching / RAG / agentic / real-time / image-gen / speech). → ai_critical_feature. Default none.

Q11 — AWS Activate credits (never dropped; ≡ full-flow Q27). Never infer funding stage from Q3 spend. Options: have credits / self-funded (→ Activate Founders up to $5K) / VC-backed (→ Activate Portfolio up to $200K, needs Org ID) / don’t know. Escalations: >$10K→ AWS Credits for AI Startups; $2K-$10K plus AND agentic → Generative AI Accelerator (up to $1M). →startup_program_status. Default"unknown" (neutral copy, both tiers).

Step 3: Assemble and write preferences.json

Merge both batches (use preferences-draft.json as base if present; strip draft metadata). Write $MIGRATION_DIR/preferences.json:

• metadata.migration_type: "ai-only" (downstream skips infra phases), metadata.discovery_artifacts: ["ai-workload-profile.json"], metadata.clarify_mode ("fast_path" if Step 1.5 fired, else "full"), and the disjoint questions_asked/questions_defaulted/questions_extracted lists.
• design_constraints: target_region (derived — precedence: Q1.5 compliance > an Azure region captured in discovery > Q4 cross-cloud preference > fallback us-east-1; chosen_by: "derived" naming the rule) and compliance (from Q1.5). NO infra constraints.
• startup_constraints.startup_program_status (Q11).
• ai_constraints: ai_framework, ai_priority, ai_monthly_spend, cross_cloud (Q4 — unique to AI-only), ai_model_baseline, ai_vision, ai_token_volume, ai_latency, ai_complexity, ai_critical_feature, ai_capabilities_required (derived from capabilities_summary).
• Persist the confirmed workloads[] (top-level) exactly as clarify-ai.md specifies — each entry carries workload_id, model_id, sdk_method, capability, capability_confidence, structured_output, call_sites, target_bedrock_model, priority, latency_tier. This is the downstream source of truth (design-ai reads it, not the profile).

Each constraint carries the full clarify field shape (value, chosen_by ∈ user|extracted|default|derived, prompt, design_consequence, source/question_id). No nulls. Delete preferences-draft.json after writing.

Step 4: Output gate

• preferences.json exists and metadata.migration_type == "ai-only".
• If either fails: STOP and output “AI-only clarify output validation failed. Fix preferences.json before completing Phase 2.”

On pass, clarify.md handles the phase-status update and advances to design. design-ai.md runs (it fires on ai-workload-profile.json), the infra design fragment no-ops (no inventory), and the chain threads through estimate-ai → generate-artifacts-ai — the same artifact-presence routing the mixed path uses.

_assert has no teeth. This flow’s gate is prose the model both produces and evaluates. Judge correctness by an AI-only fixture oracle and a fresh-context run, not by “it completed.”

Status — build step (app-code-only / AI-only route, plan §19.12)

Wired: clarify.md Step 0 routes here when the inventory is absent and ai-workload-profile.json is present. This is the standalone AI-only Clarify flow that §19.9c deferred and §19.12 reversed on owner instruction (gcp supports app-code-only; azure now does too).