Information Intermediary Evolution

An information intermediary is a business whose core value is connecting buyers to sellers of information — Google (search ads), Amazon (product discovery), Yelp (reviews), Zillow (real estate data). The agent era rewrites this value chain fundamentally.

The Agentic Disruption

Traditional chain: User → Intermediary (Google, marketplace) → Supplier → User consumes info → User decides

Agentic chain: Agent → Direct API/database → Agent evaluates all options → Agent decides

Agents bypass the intermediary because they:

Query sources directly via API
Aggregate across sources programmatically
Evaluate without attention limits
Execute transactions without intermediary capture

"Organic search" and "discovery ranking" become irrelevant when agents pull structured data directly.

Bot Information Search: Layer 0-4 Framework

How bots and agents access information forms a hierarchy from raw to synthesized:

Layer	Description	Technology	Cost Profile	Status
Layer 0	Raw web scraping	HTTP requests, HTML parsing, headless browsers	Very low per-request, high infra cost	Increasingly blocked/paywalled
Layer 1	Structured API access	REST/GraphQL APIs, MCP tools, SDKs	Low-Medium	Growing with API-first services
Layer 2	Aggregated search APIs	tavily, Exa, Brave Search API	$0.001-0.01/query	Current battleground — providers compete on quality/latency/price
Layer 3	AI-synthesized answers	perplexity Sonar, ChatGPT Browse, Google AI Overviews	$0.01-0.05/query (LLM inference)	High-value but expensive
Layer 4	Agent-mediated research	Multi-turn, multi-source autonomous research agents	$0.05-0.50+/task	Emerging — agents that plan, cross-reference, verify

Key dynamics:

Layers 0-1 are commoditizing and facing increasing access controls
Layer 2 is the current battleground — search API providers differentiate on latency, extraction quality, and agent-native output formats
Layer 3 faces the economic challenge of LLM inference cost per query — viable only when answer value exceeds cost
Layer 4 represents the frontier: agents that can conduct multi-step research, cross-reference sources, and verify claims autonomously

Intermediary opportunity at Layer 2.5: Aggregating across multiple Layer 2 search APIs (similar to how openrouter aggregates LLM APIs) with value-add through verification, synthesis, and routing optimization. Locus (YC F25) is an early example: buying APIs wholesale and re-selling at markup.

Bot Cost Economics by Content Type

The cost of serving automated traffic varies dramatically by content type, creating different economic incentives for bot management:

Content Type	Cost per Request	Bot Management Approach	Examples
Static (CDN-cached)	<$0.0001	Rate limiting + CDN edge blocking sufficient	Images, articles, cached pages
Dynamic (database-backed)	~$0.001-0.01	Multi-layer detection (IP, ASN, TLS fingerprint)	Real-time inventory, user-specific pages
Intelligent (LLM-processed)	~$0.01-0.10+	Pay-per-call (x402) or strict access control	AI agent responses, real-time analysis, personalized recommendations

The economic threshold: For static/dynamic content, bot traffic is an annoyance with manageable cost. For intelligent/LLM-processed content, bot traffic is economically significant — a single AI-generated response can cost 100-1000x more to serve than a cached page. This creates a structural need for either:

Pay-per-call monetization (x402-protocol, MPP) — every request generates revenue
Strict bot detection — multi-layer fingerprinting (JA3/JA4 TLS, ASN analysis, behavioral ML) to distinguish legitimate agents from scrapers
Tiered access — free cached content, paid intelligent content

OpenEvidence: Information Platform + Agent Model

OpenEvidence (medical information platform) demonstrates a viable monetization pattern for agent-mediated information access:

Product: AI agent interface that surfaces curated medical evidence to clinicians at decision-making moments
Users: High-value — clinicians making treatment decisions with real consequences
Revenue model: Advertising from pharmaceutical/medical device companies targeting clinicians at the point of clinical decision-making
Pattern: High-intent user base + targeted advertising = sustainable revenue without charging users directly

Why this matters for info intermediaries: OpenEvidence proves that agent interfaces can attract high-value users who engage deeply (not just click-and-leave), creating advertising inventory that commands premium CPMs. It's a counterpoint to the "agents kill advertising" thesis — advertising adapts when user intent is strong and verifiable.

Contrast with pure search APIs: tavily monetizes through API subscriptions (per-query fees), while OpenEvidence monetizes through advertising (free to users, advertisers pay). Both are viable, but the OpenEvidence model works only when user intent is monetizably high (healthcare, legal, finance, enterprise procurement).

Key Dynamics

1. Discovery Value Erodes

Agents find products/services directly via API. Search engine ranking stops mattering. "Organic search" becomes irrelevant for agent traffic.

2. Comparison Value Shifts

Agents compare at scale (thousands of options). Human-friendly comparison sites lose relevance. Machine-readable data (structured specs, real-time prices) gains enormous value.

3. Review/Reputation Changes

Agents verify claims directly (see verify-not-trust). Star ratings designed for humans lose meaning. Verification databases and cryptographic attestations gain value.

4. Ad Targeting Transforms

Agents don't see display ads. Ad networks dependent on human attention face existential crisis. New models emerge: incentive-based discovery, procurement-priced placement, verified-claim advertising.

5. The Aggregation Opportunity

As Layer 2 search APIs proliferate, a Layer 2.5 aggregator that routes queries to the optimal provider (by quality, latency, cost) and adds verification/synthesis can capture margin — analogous to openrouter in the LLM space.

Who Survives?

Information intermediaries that:

Provide exclusive data — APIs agents can't self-serve from public sources
Add verification value — Cryptographic attestations, ground-truth verification
Enable transactions — Fulfill, not just discover (closing the loop)
Aggregate with intelligence — Route across providers with quality/cost optimization

Pure discovery intermediaries without exclusive data or verification capability are structurally at risk.

agentic-commerce — Commerce implications of agent-mediated transactions
verify-not-trust — Trust shift from reputation to verification
no-app-commerce — Text-to-buy as precursor to agent commerce
tavily — Leading agent-native search API, now part of Nebius
perplexity — AI-synthesized search at Layer 3
openrouter — LLM API aggregation model for Layer 2.5 analogy