CHAPTERS
- 0:00 – 0:44
Why “Little Tech” needed a seat in DC: the gap in tech advocacy
Collin and Matt set the context: policy conversations in Washington have long been dominated by large, institutional players, leaving startups underrepresented. They frame Little Tech as founders and small builders whose needs often diverge from Big Tech’s—and whose constraints make one-size-fits-all regulation especially damaging.
- •Startups and entrepreneurs lacked dedicated advocacy compared with entrenched DC players
- •Little Tech interests are not always aligned with Big Tech—and that mismatch matters in policymaking
- •Big Tech carries political “baggage” on both the left and right that startups don’t
- •The goal is differentiated policy that reflects the realities of small teams
- 0:44 – 6:10
The Little Tech Agenda: origins, pillars, and a VC time horizon
They explain how The Little Tech Agenda emerged at a16z and what it prioritizes across policy verticals. A core pillar is proportionality: five people in a garage cannot comply like a trillion-dollar company with a 1,000-person compliance function.
- •Agenda emerged to differentiate startup needs from large incumbents’ positions
- •a16z’s policy work is “verticalized” (AI, crypto, bio/health, fintech, defense procurement, etc.)
- •Key principle: compliance burdens must scale with company size/capacity
- •Venture incentives favor long-run healthy ecosystems, not short-term deregulation
- 6:10 – 10:16
“Regulate use, not development”: why they’re not arguing for zero regulation
Matt and Collin confront a recurring misunderstanding: their framework is often misread as anti-regulation. They argue for robust governance focused on harmful uses of AI, primarily enforced through existing consumer protection, civil rights, and criminal laws.
- •They rarely, if ever, advocate for “no regulation” across the portfolio
- •Core framework: regulate harmful use, not the act of developing models
- •Enforcement anchors: consumer protection, civil rights, and criminal law
- •The message gets lost because “don’t regulate development” makes an easy headline
- 10:16 – 14:29
How the AI policy debate escalated: Senate hearings, doomer narratives, and ‘safetyism’
They trace major inflection points starting in early 2023 and accelerating after high-profile Senate hearings. Collin argues that existential-risk narratives—amplified by well-funded networks over a decade—shifted policymakers toward rapid, heavy-handed regulation.
- •Fall 2023 hearings jump-started urgency: CEOs signaled both fear and desire for regulation
- •Capitol Hill reacted strongly to “Terminator”/existential framing
- •Biden executive actions and many state proposals followed, often poorly scoped
- •Effective Altruist networks and aligned institutions shaped debate with a long head start
- 14:29 – 17:12
Policy made without startups: voluntary commitments and an ‘only a few winners’ mindset
Matt describes how early AI governance was negotiated by a handful of frontier companies, excluding future entrants. Collin adds that some policymakers assumed only 2–7 firms could compete, which encouraged thinking about restrictive, quasi-permissioned regimes.
- •Voluntary commitments were negotiated by a small set of large companies
- •Little Tech and future startups were not represented at key negotiating tables
- •Some officials assumed the frontier would be dominated by only a few firms
- •This assumption fueled interest in restrictive frameworks that could entrench incumbents
- 17:12 – 20:00
Licensing regimes and open-source bans: why “nuclear-style” AI regulation alarms them
They recount proposals to license frontier model development and regulate it like nuclear energy—plus ongoing state-level debates about restricting open source. They argue such approaches would suppress innovation, reduce competition, and risk ceding leadership to China.
- •Licensing to build frontier AI was seriously proposed and treated like nuclear regulation
- •They argue licensing is unprecedented for software and typically anti-competitive
- •Open-source restrictions remain a live policy idea in some states
- •Heavy development regulation would slow breakthroughs (medical, scientific) and increase China risk
- 20:00 – 24:51
Motivations behind heavy regulation: consumer protection politics, incentives, and distrust of builders
Collin outlines political and institutional forces he believes drove aggressive postures (including in crypto): genuine consumer protection concerns mixed with fundraising incentives, personnel backgrounds, and skepticism toward profit-making builders. Matt adds that policymakers saw AI as a chance to “do over” social media governance.
- •Consumer safety concerns can be ‘weaponized’ by interest groups and fundraising dynamics
- •“Personnel is policy”: leadership backgrounds shape regulatory instincts
- •Some policymakers treat private enterprise and profit as inherently suspect
- •AI became a perceived ‘do-over’ after bipartisan dissatisfaction with social media governance
- 24:51 – 32:43
Existing law vs. new AI frameworks: shifting goalposts, First Amendment limits, and Colorado as a test
Matt argues that most concrete AI harms are already covered by existing laws and asks critics what’s actually missed by a ‘regulate use’ approach. They discuss why misinformation regulation is constrained by the First Amendment and critique Colorado’s high-risk/low-risk compliance regime as paperwork-heavy compared with direct enforcement of anti-discrimination law.
- •Many feared harms map to existing consumer, civil rights, and criminal statutes
- •Government regulation of speech-related harms faces major First Amendment constraints
- •Colorado’s high-risk/low-risk regime relies on assessments/audits that may not solve bias
- •A cleaner alternative: explicitly enforce that using AI to violate anti-discrimination law is illegal
- 32:43 – 35:40
The case against ex ante control: marginal risk, legal norms, and ‘pre-crime’ fears
Pressed on whether ‘use’ regulation is always sufficient, Matt concedes future capabilities may require incremental policy—guided by “marginal risk.” But he cautions that preemptive enforcement against hypothetical future misuse resembles invasive, often ineffective ‘pre-crime’ surveillance that contradicts how U.S. legal systems typically work.
- •They see existing law as the starting point, not necessarily the endpoint
- •“Marginal risk” framing: add policy only when incremental risks become concrete
- •Ex ante enforcement can become invasive and unreliable (predicting harm before it occurs)
- •Overly preventive development regulation can impose high costs without preventing harm
- 35:40 – 39:31
Where AI policy stands now: the National AI Action Plan, open source, and worker-focused measures
They describe a shift toward frameworks more friendly to startups: right-sizing burden, growing support for open source, and clearer federal/state roles. Matt highlights under-covered Action Plan elements like worker retraining and labor-market monitoring, while Collin emphasizes the rhetorical pivot to ‘win while keeping people safe.’
- •Federal rhetoric and policy have moved from ‘safety-first’ toward ‘win + safety’
- •Greater consensus has emerged on the value of open source for innovation and competition
- •Action Plan includes worker retraining and labor market monitoring for potential disruption
- •They advocate federal leadership on development standards, with states policing harmful conduct
- 39:31 – 43:05
National security and China: export controls, outbound investment, and the open-source dilemma
Collin and Matt discuss competing priorities: preventing U.S. tech from aiding adversaries while avoiding overreach that undermines open source’s global influence. They argue that locking down U.S. products too tightly can create openings for Chinese alternatives, weakening U.S. soft power and market leadership.
- •Strategic goal: ensure the U.S. remains the leading AI ecosystem vis-à-vis China
- •Concerns about model-level export restrictions and outbound investment proposals
- •Open source can’t be ‘walled off’ by design; policy must account for that reality
- •Over-restricting U.S. products risks ceding global adoption and influence to China
- 43:05 – 47:12
The moratorium fight: perception, procedural politics, and coalition-building lessons
They unpack why a proposed moratorium/preemption concept failed: public perception that it banned all state AI laws for 10 years, organized opposition from ‘safety’ networks, and the constraints of a partisan reconciliation vehicle. Collin argues the loss revealed that supportive stakeholders weren’t coordinated enough, prompting new coalition and political advocacy efforts.
- •Moratorium was widely perceived as sweeping 10-year ban on state AI laws (they dispute that reading)
- •Opposition leveraged longstanding networks and allied industries to kill the effort
- •Reconciliation dynamics made bipartisan support unlikely and margins extremely thin
- •Aftermath: focus on clearer messaging, alignment across the ecosystem, and new political infrastructure (PAC)
- 47:12 – 50:07
State vs. federal roles: dormant Commerce Clause, patchwork risk, and targeted state enforcement
They propose a constitutional division of labor: Congress should govern a national AI market and development standards, while states focus on harmful conduct within their borders (especially criminal enforcement). Matt adds that some state bills may violate dormant Commerce Clause principles by imposing heavy burdens on out-of-state developers with limited local benefit.
- •Federal government: lead on interstate commerce and AI development standards
- •States: enforce and update laws against harmful conduct (criminal, civil rights, consumer harms)
- •Dormant Commerce Clause balancing test can constrain overly burdensome state regimes
- •Preferred state approach: regulate harmful use directly rather than broad compliance schemes
- 50:07 – 57:01
What’s next: federal preemption, standards, workforce/literacy, and shifting industry alignment
Looking ahead, they prioritize a federal framework that prevents a 50-state patchwork while preserving states’ ability to police harm. They anticipate more proactive state engagement, potential federal resources to lower compute/data barriers, and evolving alignment—sometimes convergent, sometimes divergent—between Big Tech and Little Tech on specific proposals.
- •Top priority: targeted federal preemption/standards to avoid a 50-state compliance maze
- •Complementary agenda: enforcement capacity, AI literacy, worker retraining, and infrastructure/energy
- •Idea gaining traction: government-backed resources to reduce compute/data barriers for startups
- •Alignment is issue-specific: Little Tech may agree with Big Tech on some items and oppose on others
