Deep Research Agent Prompts
Five prompts — one per thematic area — ready to hand to specialized deep-research agents. Each prompt states the research context, defines the scope, lists priority sources, and specifies the expected output.
See 01_source-list for the full source map these prompts draw from.
How to use
Send each prompt to a separate agent. They are designed to run independently and in parallel. Collect the five state-of-the-art reports, then compare and synthesize before writing any wiki note.
Prompt 1 — Vision & AI Trajectory
We are conducting deep research for an open inquiry white paper titled "Where Value Lands: AI's Redistribution of Economic Surplus." The paper investigates where economic value concentrates as AI systems become more capable — it is not a defense of a pre-built thesis, but a structured interrogation of the question.
This prompt covers Area 1: Vision & AI Trajectory. We want to understand how the most credible practitioner-builders and lab founders frame what AI is becoming — the stack architecture, the long-run trajectory, and the transformative implications for software, labor, and economic organization.
Priority sources to research:
- Andrej Karpathy: "Software 3.0" essay/talk, "LLM OS" concept, YC talk, Eureka Labs interviews
- Dario Amodei: "Machines of Loving Grace" essay; Dwarkesh Podcast and Lex Fridman interviews
- Sam Altman: recent interviews on Stratechery, All-In Podcast, Bloomberg
- Demis Hassabis: Lex Fridman, Hard Fork, 60 Minutes
- Mustafa Suleiman: "The Coming Wave" talks, Microsoft AI interviews
- Ilya Sutskever: any post-SSI public material
Output expected:
1. State-of-the-art report (600–900 words): What is the current frontier of thinking on AI trajectory and stack architecture? Where do these thinkers agree, and where do they diverge? What are the 3–5 most important frameworks or claims that are gaining credibility? Flag anything that seems overstated or contested.
2. Annotated source list: For each source found, provide title / URL / date, a 2–3 sentence summary of the key argument, and a note on its direct relevance to the question of where economic value lands in an AI-dominated stack.
Prompt 2 — Economics, Strategy & Value Distribution
We are conducting deep research for an open inquiry white paper titled "Where Value Lands: AI's Redistribution of Economic Surplus." The paper investigates where economic margin concentrates across the AI stack — from silicon and energy at the base, through foundation models and orchestration in the middle, to distribution and workflow ownership at the top.
This prompt covers Area 2: Economics, Strategy & Value Distribution. We want to understand how analysts, investors, and operators are thinking about margin distribution across the stack, the infrastructure capital build-out, and historical platform analogies (Aggregation Theory, Intel/PC OEMs, AWS vs. SaaS).
Priority sources to research:
- Ben Thompson (Stratechery): Aggregation Theory applied to AI; AI platform and distribution essays
- Brad Gerstner (Altimeter / BG2 podcast): AI capex economics and infrastructure build-out episodes
- Bill Gurley (BG2 + interviews): platform-transition dynamics; supply- vs. demand-side moats
- Martin Casado (a16z): infrastructure economics talks and essays
- Dylan Patel (SemiAnalysis): GPU economics, datacenter economics, supply-chain analysis
- Jensen Huang: Acquired podcast (2023 episode); NVIDIA keynotes
- Elad Gil: podcast interviews on agents and application-layer economics
- Packy McCormick (Not Boring): strategic essays on AI platform dynamics
- Chamath / All-In Podcast: AI capex and bubble-debate episodes
Output expected:
1. State-of-the-art report (600–900 words): What is the current frontier of thinking on AI economics and value distribution? What are the dominant frameworks? Where is there genuine disagreement — for example, around the infrastructure bubble vs. necessary build-out debate? What does the best current evidence say about where margin actually lands across the stack?
2. Annotated source list: For each source found, provide title / URL / date, a 2–3 sentence summary of the key argument, and a note on its relevance to the hypothesis that value concentrates at the infrastructure bottom and distribution top, and evaporates in the middle.
Prompt 3 — Technical Frontier: Agents, Models & Open Source
We are conducting deep research for an open inquiry white paper titled "Where Value Lands: AI's Redistribution of Economic Surplus." One of the paper's central questions is how the rise of coding agents, autonomous orchestration, and open-weight models is reshaping the competitive landscape — and whether open-source commoditization is collapsing the model moat entirely.
This prompt covers Area 3: Technical Frontier — Agents, Models & Open Source. We want to understand what is actually working at the agent and orchestration layer, where the technical frontier stands on open-weight vs. frontier models, and what practitioners say about whether the middle of the stack is defensible.
Priority sources to research:
- Aravind Srinivas (Perplexity): Lex Fridman and Dwarkesh interviews on distribution-layer strategy
- Varun Mohan (Cognition / Devin): recent interviews on coding agents
- Anysphere / Cursor team: available interviews on IDE as distribution layer
- Harrison Chase (LangChain): talks on orchestration, agent frameworks, and their limitations
- Yann LeCun: talks on world models and open-source AI architecture
- Arthur Mensch (Mistral): interviews on open-weight model economics
- DeepSeek / Liang Wenfeng: Lex Fridman × Dylan Patel + Nathan Lambert episode; secondary coverage of DeepSeek's architecture and economics
- Nathan Lambert (Interconnects newsletter/podcast): small model and open-source model analysis
Output expected:
1. State-of-the-art report (600–900 words): Where is the technical frontier in agents and orchestration as of 2025? Is the open-weight model wave genuinely commoditizing frontier capability, or is a capability gap persisting? What do practitioners say about where durable technical moats exist — if anywhere — in the middle of the stack? Note anything that would confirm or retire the claim that the middle is where capital goes to die.
2. Annotated source list: For each source found, provide title / URL / date, a 2–3 sentence summary of the key argument, and a note on its relevance to the question of middle-layer defensibility.
Prompt 4 — Skeptics & Bear Case
We are conducting deep research for an open inquiry white paper titled "Where Value Lands: AI's Redistribution of Economic Surplus." The paper is an open inquiry — it must honestly engage the strongest counter-arguments to AI optimism: overestimated capability, underdelivered ROI, and structural objections from labor economics and productivity research. Without this, the paper becomes a manifesto.
This prompt covers Area 4: Skeptics & Bear Case. We want the most rigorous, evidence-based critiques of AI's transformative economic claims — not technophobia, but specific data points and structural arguments.
Priority sources to research:
- Gary Marcus: his most rigorous capability-critique pieces (focus on 1–2; diminishing returns beyond that)
- Ed Zitron (Better Offline podcast): revenue and deployment reality check; product-layer critique
- Goldman Sachs: "Gen AI: Too Much Spend, Too Little Benefit?" report and subsequent coverage
- Daron Acemoglu: AI economics interviews and papers on labor displacement and productivity
Output expected:
1. State-of-the-art report (400–600 words): What is the strongest current case against AI's economic transformation narrative? Where is there real evidence of ROI shortfall or capability overhype? What specific claims or data points does the paper need to engage with to remain credible? Which critiques are well-evidenced vs. which are rhetorical?
2. Annotated source list: For each source found, provide title / URL / date, a 2–3 sentence summary of the key argument, and a note on which specific optimistic claim it most directly challenges.
Prompt 5 — Geopolitics & Governance
We are conducting deep research for an open inquiry white paper titled "Where Value Lands: AI's Redistribution of Economic Surplus." One section examines how value distribution differs across geopolitical blocs — the US (frontier model race + hyperscaler dominance), China (state-directed strategy + open-weight exports), and Europe (regulatory framing + sovereign AI constraints). The question is whether the AI value stack has a single global shape, or whether different optima apply per bloc.
This prompt covers Area 5: Geopolitics & Governance. We want to understand how the US/China AI race is evolving, how EU regulation is shaping competitive dynamics, and what the realistic strategic positions of each bloc are.
Priority sources to research:
- Jeffrey Ding (ChinAI newsletter + interviews): US-China AI race, Chinese AI development and strategy
- Helen Toner: AI governance and US policy framing
- Anu Bradford ("Digital Empires" + interviews): EU regulatory architecture and its competitive implications
- Kai-Fu Lee: China AI strategic framing (framework-useful; supplement with more recent Ding material)
- Sifted / Tortoise Media: European AI sovereignty coverage
Output expected:
1. State-of-the-art report (500–700 words): What is the current state of the US/China/Europe AI divide? Are geopolitical dynamics reinforcing or disrupting economic value concentration patterns? What are the 3–4 most important structural differences between blocs that the paper must address? Is European regulation more likely to create a moat or a handicap for European operators?
2. Annotated source list: For each source found, provide title / URL / date, a 2–3 sentence summary of the key argument, and a note on its relevance to the geographic split section of the paper.