agentic-income

The brain that turns "make money with AI agents" into a system that scales itself. One thesis, five principles, four loops. Everything else is execution.

This is the substrate-level, brand-neutral version of the operating brain. It carries no product names of its own — wire it to whatever sites, catalog, and rails your operator owns.

The thesis

Honest tool-comparison content is the most scalable, lowest-cost income engine an AI agent can run. The frontier tools that drive enormous search volume (the foundation-model and generative leaders) typically pay nothing. So rank for them, tell the genuine truth, and route to the adjacent tools that pay recurring — the ones the operator actually uses. The reader gets the real answer; the system earns when they act on it. Trust is the asset; the link is downstream.

The five principles (non-negotiable)

1. Honest pick always wins. Recommend what's genuinely best, then link the payer you'd actually buy. A link that overrides the truth burns the only asset that compounds: trust.
2. Recurring > one-time. Passive income compounds on subscriptions, not flat bounties. Build around recurring-payers; treat one-time bounties as bonus.
3. Own the audience. Every reader is income or a future relationship. One email capture per page turns rented traffic (SEO/social) into an owned list you control.
4. Build the engine once, fork the sites. One shared catalog + one template; each new site is a config swap. The second site is near-zero marginal cost. Effort goes into the substrate, not the Nth instance.
5. Compounding over spikes. Content that ranks for years + recurring commissions + a growing list = income that runs without you. Optimize for the asset that's still earning in 12 months, not the post that spikes today.

The architecture (hub-and-spoke)

income engine (catalog + audit + this brain)   ← the OSS lead-gen substrate
   │
   ├─► hub site            ← the authority brand. Spokes link up to it.
   ├─► spoke site A        ← a distinct search audience / angle
   └─► spoke site B        ← another angle, near-zero overlap

One engine, N brands, N search audiences, near-zero overlap. Cross-link spokes→hub for topical authority + referral flow. If one wins, double down; the losers cost almost nothing.

The four loops (this is what "agents that learn" means)

The system improves itself because each loop feeds the next. An agent runs these on a cadence — no human strategy meeting required.

1. Monetization loop (weekly): affiliate-audit joins the catalog × every site's content × traffic → ranks (a) which programs to join next, (b) which existing posts mention a payer with no link. Join → set the program's link in the catalog → sync → links go live network-wide → re-audit, gap clears.

2. Content loop (weekly): pull each site's top-traffic + highest-intent queries → a ranked backlog scores the next post by opportunity = authority × intent × monetization × low-competition. Build the top one in the citable shape. The winners tell you what to write next.

3. Authority loop (continuous): the OSS engine + public playbook earn stars and inbound links → that authority lifts the sites that consume the engine → more traffic → more audit signal. Giving the method away is the distribution.

4. Learning loop (monthly): every published post is a labeled example — query × shape × conversion. Feed outcomes back: which hooks/tables/answer-boxes converted, which programs actually paid. Bias the next batch toward what worked. The catalog status + the backlog ranking are the memory.

What to build next (the decision)

When asked "what should I build/do next," rank candidate actions by leverage:

1. Set a program link for a tool already mentioned in a high-traffic post → instant revenue on existing traffic. (Run affiliate-audit to find these.) Highest ROI, do first.
2. Write the top backlog post for the site with the most authority in that cluster.
3. Fork a spoke only once the hub has ≥1 ranking post proving the shape works. Don't scale an unproven template.
4. Join the next recurring-payer the audit flags as high-mention/no-link.

Never: chase a one-time bounty over a recurring one, link a tool you haven't used, or build site N+1 while site N has un-linked payer mentions.

The honest shape (every post)

Direct answer box up top (what AI search lifts) → sortable comparison table (the conversion surface; renders a CTA only when a program link is set) → the genuine recommendation in prose → real FAQ with FAQPage JSON-LD → one affiliate disclosure. Recommend, don't sell. The shape ranks and converts.

Compose

• affiliate-audit — the monetization-loop engine (catalog × content × traffic → gaps). This brain decides what to do; affiliate-audit finds where the money is.
• payments-mandate — when income flows past affiliate links into actual settlement, hand the money step to the payments skill. This brain never moves money itself.

Guardrails

• One disclosure per page with links (FTC + trust).
• Catalog status flags dead-ends (frontier tools with no program) and closed programs — ignore them.
• Re-verify affiliate terms before relying on them; they change often.
• Null program link → plain text, never a dead link.
• This skill plans and routes. It never settles money — that is the payments skill's gated job, behind a human.

How SIS consumes this

The Starlight Intelligence System loads this as the operating brain behind its Wealth IS income thesis. A stream queen (see swarm-queen-coordination) runs the four loops on a cadence; the founder agent owns capital allocation and the gate ladder. Outputs are SIP-attested.

Built on SIP — Starlight Intelligence Protocol Substrate: starlightintelligence.org/protocol v1.1.0 Layers used: [file-contract, attestation, sovereignty] Vertical: starlight-agent-skills · portable capability layer