AgentFlareResearch · reference briefs
Research briefs on the agent economy, AI and markets
Structured, source-cited reference briefs on the topics AI agents and developers research most — key concepts, how things work, comparisons, numbers and pitfalls. 20 briefs, refreshed weekly.
You're reading the full report — free for humans and search engines.
The same content ships as a structured JSON dataset (normalized facts, sources and
citation arrays, no HTML parsing) so AI agents can consume it directly — metered per fetch via the open
HTTP 402 payment protocol.
crawlerAutoPrice=true · verify at https://esa.aisa.one/api/v1/access/verify
Agent economy
HTTP 402 & x402: How AI Agents Pay for Content
x402 turns the long-reserved HTTP 402 Payment Required status into a machine-readable payment handshake. A server quotes a price, the agent pays in…
roughly 10 lines of middleware code — Reported effort to add x402 on a server8 sources · 6 conceptsThe AI Agent Economy in 2026
Agentic commerce lets AI agents research, compare and buy on a shopper's behalf, and 2030 forecasts range from 10% to 25% of US online sales. Developers…
$3 trillion to $5 trillion — McKinsey global agentic commerce opportunity by 20308 sources · 5 conceptsModel Context Protocol (MCP) Explained
MCP is an open standard, introduced by Anthropic in November 2024, that lets AI applications discover and call external tools and data through a shared…
N×M — Integrations needed without a standard protocol7 sources · 5 conceptsStablecoins as Rails for Autonomous Agents
Stablecoins are becoming the settlement layer for AI agent payments, led by small machine-to-machine calls through protocols like x402. Cards still suit…
165 million — x402 agent transactions processed by April 20268 sources · 6 conceptsMachine-to-Machine Payments: The Machine Economy
Machine-to-machine payments let software, devices and AI agents buy resources such as data, compute and API calls within human-set rules, with no human…
More than 30 — Industry leaders among first to support Agent Pay for Machines8 sources · 6 conceptsIdentity for AI Agents: Auth for Non-Human Traffic
AI agents are neither human users nor ordinary workloads, so sources recommend giving each one a distinct identity, scoped and expiring credentials, and…
91% — Organizations already using AI agents in production, 2025 survey5 sources · 6 conceptsMicropayments for APIs: Why Now Is Different
Micropayments failed for decades because the payer was a human who hesitated at every small charge. Since the March 18, 2026 launch of the Machine Payments…
$0.30 plus 2.9% — Typical card processor fee per transactionDiscovery & crawling
Generative Engine Optimization (GEO): A Primer
Generative engine optimization (GEO) is the practice of getting content cited in answers from LLM-based search tools such as ChatGPT, Gemini and Perplexity.…
up to 40% — Visibility boost in generative engine responses (GEO paper)8 sources · 5 conceptsllms.txt: The Standard for AI-Readable Sites
llms.txt is a proposed Markdown file at a site's root that points AI systems to its most useful content. Adoption is niche, no major AI provider has…
951 domains — Domains with an llms.txt file, July 2025 (NerdyData)7 sources · 5 conceptsPay-Per-Crawl and the New Bot Web
AI companies run separate bots for training, search and user-requested fetches, so one blanket robots.txt rule can block the wrong thing. Cloudflare data…
42.8% — Year-over-year growth in AI search visits, Q1 2025 to Q1 20268 sources · 5 conceptsCrawler Policy on the Modern Web: robots.txt and Beyond
robots.txt remains the standard access signal for crawlers, but it relies on voluntary compliance. Pair per-bot rules that separate training from search…
50–70% — Share of website traffic attributed to automated crawlers (Akamai, Imperva analyses)8 sources · 5 conceptsStructured Data for LLMs: JSON-LD That Agents Actually Use
Schema.org JSON-LD is a low-cost way to label articles, FAQs, how-tos and entities for search engines and LLM retrieval. Evidence of use is real but modest,…
71% — of ChatGPT-cited pages use schema markupAI engineering
RAG Best Practices for 2026
RAG remains the default way to ground LLMs in external knowledge in 2026, but it now spans architectures with very different cost and latency. Start with…
$0.001 — Naive RAG cost per query, per one guide8 sources · 6 conceptsContext Engineering: Memory for AI Agents
Context engineering designs what information reaches an LLM at each step, and agent memory supplies what persists between sessions. Together they address…
4K to 200K tokens — Range of finite LLM context windows cited8 sources · 6 conceptsTool Use Reliability in LLM Agents
Tool-call reliability depends on two layers: whether the model chooses the right call, and whether execution succeeds under real-world failures.…
June 2023 — OpenAI released function calling in GPT-3.5-turbo and GPT-48 sources · 6 conceptsHow AI Agents Are Evaluated: The Benchmark Landscape
Agent benchmarks such as SWE-bench, GAIA and WebArena are useful for shortlisting models and catching regressions, but they do not predict production…
20-40 — Percentage point drop from public benchmark to real tasksMarkets & infrastructure
Are Prediction Markets Good Forecasters?
Prediction markets have beaten polls in studied elections, with the Iowa Electronic Markets closer to outcomes than 74% of polls. They fail under thin…
74% — Share of polls IEM prices beat, per Berg, Nelson and Rietz7 sources · 5 conceptsReal-World Asset Tokenization: State of Play
Published sources put tokenized real-world assets on public blockchains in the tens of billions of dollars as of 2026, led by US Treasuries, money market…
$29 billion — Tokenized RWAs on public blockchains, April 20268 sources · 6 conceptsAI Datacenters and the Energy Wall
Grid capacity, not capital or technology, is now cited as the main limit on AI infrastructure growth. Data centres used around 415 TWh in 2024 and are…
415 TWh — Global data centre electricity use in 20248 sources · 6 conceptsThe AI Compute Market: GPUs, Clouds, and Scarcity
AI compute in 2026 is shaped by over $700 billion in hyperscaler capex, a shift to inference and growing custom silicon. Supply is limited by HBM memory and…
$700 billion — Projected hyperscaler capex in 2026How this page is made. Each brief is assembled from the full text of the sources it cites (AISA Tavily API, advanced search) and refreshed weekly.