Per-client reports, on autopilot.

1. The pain — SaaS slaughterhouse + zero-click search

The 2026 agency P&L is a SaaS slaughterhouse. A mid-size content shop running Ahrefs Standard ($249/mo), Semrush Guru ($249.95/mo), AgencyAnalytics ($79+/mo), ClickUp Business ($12/user/mo plus $9/mo for Brain), Buffer or Sprout, and Frase or Surfer is spending $700-1,500 per seat before payroll. Whatagraph starts at $229/mo with credit-based source pricing. Every client onboarding means provisioning seven logins.

Then your output gets harder to defend. Google AI Overviews now appear in 25.8% of US searches as of January 2026 — 39.4% on informational queries. Searches that trigger AI Overviews average an 83% zero-click rate; non-AIO queries already average ~60%. Organic CTR drops 15-46% depending on query type (Unodeskly 2026, citing 68k-query study). Gartner forecasts a 25% search-volume drop by 2026. Your "ranked #3" deliverable means less every week.

And the work itself is increasingly about being cited by an LLM, not by a person. Generative engine optimization (GEO/AEO) splits into a dozen surfaces — ChatGPT, Claude, Copilot, Perplexity, Gemini, AI Overviews, Grok, DeepSeek, Kimi, Qwen — each measured by mention rate, citation rate, and share of voice. AI-referred sessions grew 527% YoY in early 2025. Your old reporting template — keyword rank, backlinks, page speed — does not measure any of this. Clients can tell.

3. The os.moda stack

1. 1 · lantern (long-form) drafts the weekly client brief in each client's tone — rank deltas, AI Overview presence, citation share, top-3 recommendation block. Linked SQL queries are the citations.
2. 2 · soot (schedule/ETL) pulls GA4, GSC, Ahrefs/Semrush APIs, social platforms, ad accounts into Postgres on a schedule. Adds nightly GEO crawls of ChatGPT, Perplexity, AI Overviews for tracked queries.
3. 3 · haku (creative/code) ships landing pages and ad creative variants on demand. One Slack prompt → 12 ad variants → 3 landing page A/B tests, in the client's brand voice and design system. Approval gated through naga.
4. 4 · frog (scoring) ranks campaigns and content by ROI and predicted ranking lift. "What should we kill, what should we double down on" is a SQL view, not a Monday-morning argument.
5. 5 · Per-client brand voice profile is a versioned artifact, not a Notion doc. Forks for sub-brands. Tone, banned phrases, escalation rules, competitor list travel with the client.
6. 6 · naga (vault/risk) holds API keys per client (Google Ads, Meta, Stripe, GSC) in KEYD vault. EU residency option. Approval flow on anything that goes live.
7. 7 · Client dashboard auto-generated per account — rank, GEO citations, campaign ROI, brief archive, content pipeline. White-labeled. Replaces AgencyAnalytics for the part clients actually open.

5. The 3–5 year future

• 2027 · brand-voice models per client deployed as agents your clients call directly. "Ask the brand" via Slack or WhatsApp — the agency monetizes the model, not the deliverable.
• 2028 · one-prompt campaigns. "Launch the spring sale" produces ad copy, three landing pages, the launch email, the SMS, the social posts, and the post-launch report — drafted in 90 seconds, shipped after human approval.
• 2029-2030 · agency-as-platform. The agency's stack becomes a product clients buy. Reporting, GEO tracking, content briefs, the per-client agent ship as a tenant. The agency becomes a SaaS with a service tier.