Full architecture for producing, routing, and delivering intelligent email digests across all 15 AaaS surfaces. Covers model selection per digest type, Vertex AI integration, the military execution roadmap, and every email product from daily AI picks to monthly ecosystem updates.
Emails are not static templates. Each digest is dynamically generated by a Cloud Function that: (1) queries Firestore for new content since last send, (2) fetches structured data from each surface's backend, (3) calls a Claude model on Vertex AI to write the narrative, and (4) renders a React Email template with the AI output. The result is sent via Resend in batch.
Different email types use different generation strategies. The Today daily digest is almost mechanical — GravityClaw already scored and ranked the 5-7 items, the function just formats them. The Weekly Roundup needs Claude to write a cohesive narrative from 6 sources. The nurture sequence needs Claude to personalise tone based on account type and onboarding state.
Claude Code is a coding assistant (this CLI tool). It is not an API endpoint you can call from server code. It has no programmatic interface for backend services. What you want is the Claude API — which is available two ways:
Option A — Anthropic API directly: Separate API key, billed directly to Anthropic. Works fine, but requires managing a separate credential outside your GCP account.
Option B — Claude on Vertex AI: Claude models (Haiku, Sonnet, Opus) available in Google Cloud's Model Garden via the Anthropic ↔ Google partnership. Same GCP billing account, same IAM, same service accounts as Firebase and Cloud Functions. This is the recommended path.
All existing infrastructure runs on GCP: Firebase, Cloud Functions, Cloud Scheduler, Firestore, Cloud Run, GravityClaw (already uses Vertex AI Gemini). Adding Claude on Vertex AI means zero new billing relationships, zero new secrets to rotate, and sub-50ms latency from Cloud Functions to the model. The GravityClaw scanner already uses @google-cloud/vertexai — the same SDK adds Claude access with one additional model identifier.
Additionally, Gemini continues to run scoring/ranking (it excels at structured classification). Claude handles narrative writing (it excels at coherent prose). Both run on the same Vertex AI endpoint.
| Email Product | Frequency | AI Task | Model | Rationale |
|---|---|---|---|---|
| Today — Daily AI Picks | Daily | Format pre-scored items into email HTML | Gemini Flash | GravityClaw already ranks items. Just templating — no prose generation needed. |
| Academy — Daily Pack Nudge | Daily | 1-line personalised nudge + streak status | Claude Haiku | Short output, high volume, low stakes. Haiku is 3× cheaper than Sonnet. |
| Blog — Weekly Channel Digest | Weekly | Summarise top 10 new entities per subscribed channel | Claude Haiku | Structured summaries of existing content. No original thinking needed. |
| Platform — Weekly Usage Digest | Weekly | Write task accomplishment narrative from structured metrics | Claude Sonnet | Needs coherent "what your agents did this week" narrative — not just data dump. |
| Quest — Daily Job Alerts | Daily | Match jobs to user profile, write 1-line relevance note per role | Claude Haiku | Mechanical matching with short output. Fast, cheap, high volume. |
| Academy — Weekly Challenge Digest | Weekly | Write challenge summary + leaderboard narrative | Claude Haiku | Structured challenge data → short formatted summary. |
| Forum — Weekly Digest | Weekly | Summarise top threads, best prompts, tool reviews | Claude Sonnet | Community content summarisation needs nuance and tone. |
| Weekly Roundup (cross-surface) | Weekly | Write cohesive narrative linking content from 6 surfaces | Claude Sonnet | Multi-source synthesis with consistent voice. Core flagship email — invest here. |
| Boutique — Monthly Case Study | Monthly | Narrativise implementation case from structured brief | Claude Sonnet | Story-driven content, low volume — Sonnet quality warranted. |
| Monthly Ecosystem Update | Monthly | Write strategic product narrative + roadmap teaser | Claude Sonnet | Brand voice, low volume, high visibility. Sonnet delivers with review. |
| Lifecycle Nurture — Welcome | Triggered | Personalised welcome by account type + interests | Claude Sonnet | First impression. Personalisation matters. One-time per user. |
| Lifecycle Nurture — Day 7–30 | Triggered | Contextual nudges based on behaviour gaps | Claude Haiku | Short, targeted messages. Pattern-driven not creative. Haiku sufficient. |
Use for: high-volume mechanical tasks, short outputs, structured formatting, daily sends with 1,000+ recipients. ~3× cheaper than Sonnet, ~15× cheaper than Opus.
Use for: narrative writing, multi-source synthesis, personalisation, brand voice, weekly and monthly digests. Best quality-to-cost ratio for prose generation.
Not used for bulk email. Reserved for one-time tasks: designing new email templates, reviewing nurture strategy, generating onboarding copy. Never in scheduled functions.
GravityClaw uses @google-cloud/vertexai with Gemini Flash for: scoring news items by novelty × impact × clarity, classifying AI entities into the 16 blog types, ranking content for the Today digest.
Keep Gemini for: classification, scoring, structured JSON extraction, ranking, high-volume low-latency tasks.
Via Anthropic ↔ Google partnership, Claude Haiku, Sonnet, and Opus are available in Vertex AI Model Garden. Same SDK, same IAM service account, same GCP billing.
Use Claude for: prose generation, narrative synthesis, personalisation, email copy, nuanced summarisation, brand-consistent writing.
Running Anthropic API separately would mean: a separate API key in Secret Manager, separate billing account, separate rate limit management, and higher latency from us-central1 to api.anthropic.com. Vertex AI keeps everything in the same GCP project with unified observability, IAM, and billing — which is already set up.
| Type | Platform | Academy | Blog | DIY | Toolbox | Boutique | Business | Quest |
|---|---|---|---|---|---|---|---|---|
| founder | ★★★ | ★★ | ★★ | ★★ | ★★★ | ★★ | ★★★ | — |
| builder | ★★ | ★★★ | ★★★ | ★★★ | ★★ | — | — | — |
| creator | ★★ | ★★ | ★★ | ★★★ | ★★ | — | — | — |
| business | ★★★ | — | ★ | — | ★★★ | ★★★ | ★★ | — |
| consultant | ★★ | ★★ | ★★ | ★★ | ★★ | ★★★ | — | — |
| investor | ★★ | — | ★★ | — | — | — | ★★★ | — |
| learner | — | ★★★ | ★★ | ★★ | ★★ | — | — | ★★★ |
| researcher | — | ★★ | ★★★ | ★★★ | — | — | — | — |
| Surface | URL | Free Digest Offer | Frequency | Status | Model |
|---|---|---|---|---|---|
| Today | aaas.today | 5-7 curated AI picks, GravityClaw-scored | Daily | Infra built | Gemini Flash |
| Blog | aaas.blog | Top 10 new entities per subscribed channel | Weekly | Live | Claude Haiku |
| Academy | aaas.academy | Daily pack nudge + weekly challenge | Daily + Weekly | Live | Claude Haiku |
| Platform | agents-as-a-service.com | Agent task accomplishments + credits used | Weekly | Live | Claude Sonnet |
| Toolbox | aaas.blog/toolbox | New templates + featured pick | Monthly | Live | Claude Haiku |
| Quest | aaas.quest | AI-adjacent job alerts matching profile | Daily or Weekly | Coming Q2 | Claude Haiku |
| Forum | aaas.forum | Top threads, best prompts, tool reviews | Weekly | Q3 2026 | Claude Sonnet |
| Boutique | aaas.boutique | One implementation case study | Monthly | Coming Q2 | Claude Sonnet |
| Builders | aaas.builders | New skill slams + project showcases | Weekly | Coming Q2 | Claude Haiku |
| Business | aaas.business | Investor intel + startup funding news | Weekly | Coming Q2 | Claude Haiku |
| Weekly Roundup | All surfaces | Best of everything above in one email | Sunday | Build First | Claude Sonnet |
| Monthly Update | All surfaces | Ecosystem news, new surfaces, roadmap | 1st of month | Build First | Claude Sonnet |
| Day | Founder | Builder | Business | Researcher | Learner |
|---|---|---|---|---|---|
| D+0 | Context generating | Dev context ready | Business context ready | Research context ready | Learning path starts |
| D+1 | First agent task CTA | DIY intro — GitHub repos | Select email task CTA | Blog intro — 2,847 entities | Day 1 Academy pack |
| D+3 | Vault highlight — founder skills | Vault — top 5 technical | Toolbox — top 10 ops templates | Channel setup — personalised feed | Streak nudge (Day 3!) |
| D+7 | Toolbox — 10 founder templates | Context Engineering crash course | Case study — [company] automated [process] | Paper highlight — top 3 this week | Weekly challenge announce |
| D+14 | Case study — 3 hires replaced | Builders — first skill slam | Boutique CTA — full workforce | DIY — research papers to study packs | Quest — AI jobs now hiring |
| D+21 | Platform new features | Forum waitlist | Weekly digest sample | Benchmark update | Progress — X packs in 3 weeks |
| D+30 | Strategy call CTA | Skill review — what are you building? | Upgrade to boutique path | API intro — build your pipeline | Forum waitlist — Q3 launch |