Skip to content
a16za16z

The Little Tech Agenda for AI

Who’s speaking up for AI startups in Washington, D.C.? In this episode, Matt Perault (Head of AI Policy, a16z) and Collin McCune (Head of Government Affairs, a16z) unpack the “Little Tech Agenda” and latest in AI policy - why AI rules should regulate harmful use, not model development; how to keep open source open; the roles of the federal government vs states in regulating AI; and how the U.S. can compete globally without shutting out new founders. Timecodes: 00:00 Introduction 00:40 The Little Tech Agenda: Origins & Vision 01:01 Defining “Little Tech” vs. Big Tech 05:00 Challenges Unique to Startups 06:09 Regulatory Frameworks for AI Startups 10:01 The Evolution of AI Policy Debates 11:00 Senate Hearings & the Rise of AI Regulation 13:00 The Influence of Effective Altruists & Interest Groups 14:29 Big Tech at the Policy Table 17:12 Licensing Regimes & Open Source Debates 19:09 National Security & Global Competition 26:57 Crypto Policy Parallels 29:47 The Dormant Commerce Clause & State Laws 31:33 Federal Preemption & the Need for Standards 32:44 Building Coalitions & Political Advocacy 34:00 Industry Alignment & Divergence 35:28 Where Are We At Now? 47:14 State vs. Federal Roles in AI Policy 50:07 The Future of AI Policy: Preemption, Workforce, and Literacy 54:31 Industry Alignment and Political Dynamics Resources: Read the Little Tech Agenda: https://a16z.com/the-little-tech-agenda/ Read ‘Regulate AI Use, Not AI Development: https://a16z.com/regulate-ai-use-not-ai-development/ Read Martin’s article ‘Base AI Policy on Evidence, Not Existential Angst: https://a16z.com/base-ai-policy-on-evidence-not-existential-angst/ Read ‘Setting the Agenda for Global AI Leadership’: https://a16z.com/setting-the-agenda-for-global-ai-leadership-assessing-the-roles-of-congress-and-the-states/ Read ‘The Commerce Clause in the Age of AI”: https://a16z.com/the-commerce-clause-in-the-age-of-ai-guardrails-and-opportunities-for-state-legislatures/ Find Matt on X: https://x.com/MattPerault Find Collin on X: https://x.com/Collin_McCune Stay Updated: Let us know what you think: https://ratethispodcast.com/a16z Find a16z on Twitter: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Subscribe on your favorite podcast app: https://a16z.simplecast.com/ Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details, please see a16z.com/disclosures.

Collin McCunehostMatt Peraulthost
Sep 8, 202557mWatch on YouTube ↗

CHAPTERS

  1. Why “Little Tech” needs its own voice in policy

    The hosts set up the core problem: policy conversations in DC and state capitals have long been dominated by large, institutional players, leaving startups underrepresented. McCune and Perault explain why startup realities—limited staff, capital, and compliance capacity—require a distinct advocacy lens.

    • Policy rooms often lack representation for startups and early builders
    • Big Tech and Little Tech interests are not always aligned
    • Startups can’t comply with regimes designed for trillion-dollar firms
    • A16Z positions Little Tech as a missing constituency in tech policy
  2. The Little Tech Agenda: origins, pillars, and the 10-year time horizon

    McCune outlines how the agenda emerged inside the firm and how it maps to a16z’s “verticalized” policy work across AI, crypto, bio/health, fintech, and defense. Perault adds that venture’s long time horizon demands a stable, trusted ecosystem—meaning smart governance, not a regulatory vacuum.

    • Agenda built to differentiate startup needs from Big Tech “baggage”
    • a16z policy work spans multiple sectors (AI, crypto, bio, defense, fintech)
    • Core pillar: five-person garage teams vs. 1,000-person compliance orgs
    • Long-run ecosystem health matters more than short-term market spikes
    • Safety and trust are aligned with venture incentives over a decade
  3. Clarifying the mantra: “Regulate use, not development”

    Perault explains the most-misunderstood element of their framework: focusing regulation on harmful uses rather than restricting model development itself. They argue existing bodies of law—consumer protection, civil rights, and criminal law—provide a robust enforcement baseline, yet the slogan is often misread as ‘no regulation.’

    • They reject “zero regulation” and emphasize governance
    • Regulate harmful uses: fraud, discrimination, and criminal activity
    • Existing consumer protection and civil rights laws already apply to AI use
    • Misinterpretation: “don’t regulate development” gets framed as deregulation
    • Goal is enforceable, outcomes-based protection without crushing startups
  4. How AI policy debates accelerated: Senate hearings, ‘doom’ narratives, and the Biden EO

    McCune traces a key inflection in fall 2023 Senate hearings where major CEOs signaled both desire for regulation and discussed existential-risk scenarios. He argues that testimony, combined with broader safety-oriented advocacy, spooked policymakers and helped catalyze rapid regulatory pushes—via executive action and state bills.

    • Early 2023 marks the start of serious AI policy momentum in their view
    • Fall 2023 hearings jump-started urgency and ‘Terminator’ style fears
    • Resulting pressure contributed to sweeping executive and state activity
    • Concerns that proposals were rushed and not well thought through
    • Safety framing became dominant, often at the expense of competition
  5. Effective altruism, think-tank influence, and interest-group dynamics

    McCune argues that ‘safetyism’ and doomer narratives benefited from a decade-long head start, with well-funded networks shaping think tanks and nonprofits. They describe their role as playing catch-up to rebalance the debate toward innovation, competition, and practical governance.

    • Claim: EA-aligned networks influenced policy discourse for ~10 years
    • Fear-based narratives can be amplified through nonprofit/think-tank channels
    • They dispute claims that AI industry spending dominates the policy arena
    • Their jobs and advocacy are framed as a counterweight to entrenched narratives
    • EU AI Act and global debates are cited as shaped by safety-first thinking
  6. Big Tech at the table: voluntary commitments and ‘small number of frontier players’ assumptions

    Perault points to White House voluntary commitments negotiated by a small set of large AI companies as evidence that Little Tech wasn’t represented. Both warn that recurring policy assumptions—‘only 3–7 companies will build frontier AI’—normalize rules that entrench incumbents and shrink the competitive frontier.

    • Voluntary commitments were negotiated by a handful of major firms
    • Little Tech and future entrants weren’t represented in those deals
    • They reject the idea that only a few frontier builders is acceptable
    • Comforting claims that rules apply to “just a few firms” are alarming to them
    • Competitive diversity at the frontier is their desired market structure
  7. Licensing regimes and open-source restrictions: ‘nuclear-style’ regulation as an anti-competitive trap

    They revisit proposals to require licenses to build frontier models and to constrain open source, calling these unprecedented for software and likely to entrench monopolies. McCune uses nuclear regulation as a cautionary tale: well-intended oversight can drastically reduce new entrants and output over decades.

    • Frontier licensing proposals were seriously considered in early debates
    • They compare it to nuclear regulation: high friction, low innovation output
    • Licensing is framed as historically unprecedented for software development
    • Open-source bans/restrictions are still proposed in some state contexts
    • Such regimes risk cementing existing giants and blocking startups
  8. National security, China, and the ‘lock it down’ paradox

    McCune emphasizes that overly restrictive domestic regulation can cause the US to lose strategic advantage to China. They argue that attempts to block diffusion—especially of open source—can backfire, pushing global markets toward Chinese alternatives while weakening US soft power.

    • Central claim: overregulation risks the US losing the AI race to China
    • Export controls/diffusion rules seen as contentious and potentially too restrictive
    • Outbound investment limits are seen as more sensible than banning open source
    • Open source is hard to ‘wall off’ and may be globally inevitable
    • Restricting US products can create openings for Chinese market capture
  9. Parallels to crypto: policy debates as proxies for older unresolved fights

    McCune draws an analogy to crypto regulation, where battles over tokens sometimes become a proxy for broader securities-law reform. He suggests AI policymaking similarly attracts attempts to re-litigate older internet governance issues (privacy, content moderation, algorithmic bias) by funneling them through AI frameworks.

    • Crypto debates often mask attempts to overhaul securities laws indirectly
    • AI debates can become a vehicle to revisit ‘missed’ telecom/internet regulation
    • AI can be used as a regulatory ‘funnel’ for privacy, moderation, and bias issues
    • They argue this muddies the waters and produces misfit regulatory schemes
    • The result can be complex compliance burdens with unclear consumer benefit
  10. State laws, impact assessments, and the Colorado example

    Perault critiques state “high-risk/low-risk” AI frameworks (like Colorado’s) as paperwork-heavy and hard for startups without counsel to navigate. He contrasts this with a more direct approach: explicitly prohibiting the use of AI to violate anti-discrimination law, which targets harm without broad administrative regimes.

    • Colorado-style regimes impose classification, assessments, audits, and paperwork
    • Startups often lack legal/policy staff to comply
    • Governors/AGs seeking rollbacks signals practical implementation concerns
    • Directly enforcing anti-discrimination violations is framed as more effective
    • They argue administrative processes may not achieve claimed bias-reduction goals
  11. Why ex-ante control sounds appealing but often fails in practice

    They address fears about future catastrophic harms (bioterrorism, cybercrime) and whether regulating use is enough. Perault argues the legal system generally punishes unlawful conduct rather than preemptively policing predicted wrongdoing, and that preemptive surveillance-style regulation is both intrusive and unreliable.

    • They concede future marginal risks may require new policy over time
    • Regulating purely ex ante resembles predictive policing/surveillance concerns
    • Existing law is a strong starting point, though not necessarily the endpoint
    • They favor targeted additions when incremental risk becomes concrete
    • Prevent-harm-first frameworks often impose high costs with uncertain benefit
  12. Where things stand now: AI Action Plan, open source support, and workforce measures

    Perault and McCune say the policy climate has improved relative to two years ago, with greater support for startup-friendly approaches and open source. They highlight under-discussed elements of the AI Action Plan, especially worker retraining and labor-market monitoring to respond to potential disruptions.

    • Shift toward “right-sizing” regulation and reducing burdens that don’t add value
    • Greater consensus on open source as a driver of competition and innovation
    • AI Action Plan reframes the balance: win globally while keeping people safe
    • Worker retraining and labor-market monitoring seen as pragmatic safeguards
    • They emphasize signaling effects: US rhetoric influences Congress and allies
  13. The moratorium fight: perception, coalitions, and political organizing

    McCune explains the controversy around a proposed moratorium/preemption concept, arguing it was widely mischaracterized as wiping out all state law for 10 years. He attributes its failure to perception, strong opposition networks, partisan legislative dynamics, and insufficient pro-preemption coalition organization—prompting renewed efforts to coordinate advocacy and political strategy.

    • Public narrative: ‘10-year ban on all state AI law’ vs. disputed textual reality
    • Opposition leveraged established networks and allied industries to block it
    • Reconciliation vehicle made bipartisan support unlikely and margins tight
    • Lesson: pro-standardization stakeholders were not organized enough
    • Response: coalition-building, clearer public education, expanded political advocacy (PAC)
  14. State vs. federal roles: Dormant Commerce Clause, preemption, and a workable standard

    They outline a constitutional division of labor: federal leadership on interstate commerce and AI development standards, states enforcing harmful conduct within borders. Perault adds that some state proposals may run into Dormant Commerce Clause issues by imposing heavy out-of-state burdens with limited local benefit—another reason to focus states on harmful-use enforcement while pursuing federal standards for development.

    • Federal role: national market rules and AI development standards
    • State role: enforce criminal/civil harms and police conduct locally
    • Dormant Commerce Clause can constrain overly burdensome state regimes
    • Preferred outcome: avoid a 50-state patchwork that startups can’t navigate
    • Preemption focus is specific—not a blanket elimination of state authority
  15. Next 6–12 months: federal standard, AI literacy, infrastructure, and shifting industry alliances

    They predict near-term focus on a federal framework that prevents a patchwork while preserving state enforcement of harms. They also flag workforce training, AI literacy, and infrastructure (energy/data centers), plus potential government resources to lower startup barriers (compute/data access). Finally, they anticipate periods of both convergence and divergence across Big/Medium/Little Tech—and emphasize that their positions follow principles, not party or incumbents.

    • Top priority: targeted federal preemption/federal standard to avoid patchwork
    • States continue policing harmful conduct; federal government sets development rules
    • Additional priorities: workforce retraining, AI literacy, energy/data center buildout
    • Idea: government-backed shared resources to reduce compute/data barriers for startups
    • Industry alignment may fracture; Little Tech lens can diverge from Big Tech when needed

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.