Summary

a16z’s Little Tech thesis is that AI rules built for trillion-dollar incumbents could compound their existing capital, compute, and talent advantages. The operative comparison is “five people and you’re in a garage” versus “thousand-person compliance teams”; licensing, audits, impact assessments, and disclosure regimes could entrench a small number of frontier developers. For investors, policy design may determine whether startups remain credible challengers or the market consolidates by construction.
Matt Perault and Collin McCune reject zero regulation in favor of “regulate use, do not regulate development.” Existing consumer-protection, civil-rights, and criminal laws already reach many harmful AI uses, while development controls burden every builder before a violation occurs. Perault ties that position to venture’s 10-year fund horizon: unsafe, scammy products and public distrust cannot produce the durable ecosystem investors need.
The pair argues that the 2023 safety panic nearly normalized an unprecedented permission-to-build regime for software. CEO testimony, existential-risk advocacy, and a social-media-policy “do-over” produced proposals for frontier licenses, nuclear-style oversight, FLOPs-based disclosures, and open-source bans. Perault’s warning is concrete: U.S. nuclear policy yielded only “2 or 3 new nuclear power plants in a 50-year period”; applying that model to AI would suppress breakthroughs and, in his categorical view, mean “you lose to China.”
Colorado is their clearest case that compliance process can substitute for actual protection. Its framework asks resource-constrained startups to classify high-risk uses and conduct assessments or audits that might identify bias but will not “end racism in our society.” Perault prefers direct liability: codify that using AI to violate anti-discrimination law is illegal, then let the attorney general enforce against the observable harm.
They concede that existing law may not be the endpoint if AI creates genuinely incremental risk. Asked whether terrorism, cybercrime, or other capabilities could become “10,000x” more powerful, Perault calls that conceivable and endorses policy responsive to marginal risk. His resistance is to speculative ex-ante surveillance—predicting who might offend and intervening before conduct occurs—which may be both invasive and ineffective.
The National AI Action Plan marks a thesis-level shift from “safety with a splash of innovation” toward winning while keeping people safe. The speakers highlight support for open source, right-sized startup regulation, worker retraining, labor-market monitoring, and clearer federal-state roles. On China, McCune sees a hard tradeoff: restrictions must keep powerful technology from the PLA and CCP, but locking down U.S. open-source models invites Chinese products to become the world’s default platform.
The failed federal moratorium exposed political execution risk, not consensus for a 50-state AI patchwork. McCune says its perceived 10-year ban on all state AI law was a misreading, but “perception is reality”; a partisan reconciliation vehicle and one or two Republican senators were enough to defeat it. The next push is narrower federal preemption for model regulation, backed by a more organized coalition and Leading the Future PAC—even as Perault expects big and little tech may diverge again on the details.

Deep dive

1. Little Tech exists because startups had no seat at the policy table

McCune traces the agenda to a structural representation gap: major institutional technology players had operated in Washington and state capitals for years, but nobody specifically represented “the startups and entrepreneurs, the smaller builders in the space.” Those companies are not automatically aligned with Big Tech, even when their interests sometimes overlap.
Perault describes the agenda as his recruiting vehicle. After reading it, policy rooms acquired an obvious “empty seat”: officials would casually add another disclosure or compliance requirement without asking how it would work for companies absent from the meeting.
The agenda’s simplest test is organizational capacity. A five-person garage startup cannot absorb rules like a trillion-dollar company with hundreds of thousands of employees and “thousand-person compliance teams”; many portfolio companies lack even a general counsel, policy lead, or communications head.
That asymmetry compounds AI’s existing entry barriers. A startup trying to challenge Microsoft, OpenAI, Meta, or Google already needs scarce data, compute, and talent; McCune and Perault ask which frameworks protect people without making that competition “even more difficult than it already is.”

2. Ten-year capital needs governance, but it should attach to harm

Perault’s incentive argument starts with venture’s 10-year fund cycles: the firm is not trying to spike the AI market for six months or two years. Scammy products, democratic damage, or communities deciding AI is corrosive would undermine both public welfare and long-run financial returns.
McCune says “99.9% of people” they talk to assume the objective is no regulation, yet he cannot identify “a single example across the portfolio” where a16z advocates zero regulation. The framework’s actual sentence is “regulate use, do not regulate development,” but critics routinely omit its first half.
Regulating use includes enforcing consumer-protection, civil-rights, and criminal laws when AI facilitates violations. Perault’s diagnostic question—“what do we miss?” if policy begins with existing law—has produced few clear answers, because most concrete harms people cite involve conduct that existing law already covers.
Misinformation is a different category. Perault acknowledges problematic speech may demand a social response, but the First Amendment sharply limits government control over private platforms’ speech policies; regulation is therefore not automatically the appropriate instrument for every legitimate concern.

3. A 2023 safety panic almost normalized licensing software

McCune dates the policy starting gun to early 2023 and the decisive acceleration to Senate hearings that fall. AI executives said they wanted regulation while speculation implied, in his deliberately exaggerated rendering, “go hug your families because we’re going to all be dead in 5 years.” Capitol Hill predictably panicked.
He connects that atmosphere to the Biden executive order, restrictive state proposals, federal bills, and the EU AI Act. He also credits the effective-altruist community with a decade-long, well-funded head start influencing think tanks and nonprofits around existential safety; a16z’s policy operation is, in his words, “playing catch-up.”
Perault adds the post-2016 social-media backlash: companies had been accused of allowing products to outrun governance, so AI became a do-over. Three, five, or seven companies negotiated White House voluntary commitments for frontier development while every current and future startup outside that group went unrepresented.
The endpoint under discussion was extraordinary: government licenses to build frontier models, nuclear-style international oversight, FLOPs-threshold disclosures, and possible open-source bans. Perault invokes U.S. nuclear policy’s “2 or 3 new nuclear power plants in a 50-year period” as the cautionary mechanism—an agency’s effects can radically exceed its stated intent.

4. The social-media do-over collided with competition policy

Asked why the previous administration took this path, McCune refuses to claim certainty. He points to consumer-safety constituencies, fundraising built around messages such as “AI is coming for your jobs,” and the principle that “personnel is policy,” while still saying the motivating concerns may have been held in good faith.
Perault’s more sympathetic reading is that policymakers believed they had slept through social media’s harms and wanted to get AI right early. His objection is the contradiction: politicians who had condemned concentrated social-media markets then supported licensing, a tool that typically reinforces concentration in an AI market already predisposed toward high barriers.
McCune sees AI as a venue for policy fights that cannot move independently. Just as crypto legislation attracts people seeking broader securities-law reform, AI bills can become vehicles for privacy, content-moderation, and algorithmic-bias rules—placing a “mesh screen” over everything as more economic activity flows through AI.

5. Colorado shows why paperwork can miss the harm

Colorado’s enacted framework requires companies to determine whether an AI use is high-risk and, if so, complete obligations such as impact assessments or audits. Perault stresses the mismatch for startups without lawyers: the process might reveal some bias, but it probably will not eliminate bias and “certainly isn’t going to end racism in our society.”
A proposed alternative during Colorado’s reconsideration went directly to the conduct: codify that using AI to violate the state’s anti-discrimination statute is illegal and empower the attorney general to enforce it. Perault cannot see why that is less compelling than an “amorphous” paperwork regime that may or may not change outcomes.
The host’s hardest pushback is prospective: AI does not currently appear to make terrorism or crime “10,000x” more capable, but could a breakthrough make use-based regulation insufficient? Perault’s answer is explicitly provisional—existing law is “a good place to start” and “probably not where we end”; genuinely incremental risk may justify additional policy.
Prevention nevertheless carries its own danger. Perault’s analogy is government gathering information about a person, predicting a probability of future illegality, and intervening before any act; that kind of ex-ante surveillance feels invasive and may not predict conduct reliably enough to prevent the feared harm.

6. Washington’s center of gravity moved from safety-first to winning

Perault sees markedly better federal conditions for Little Tech: efforts to right-size burdens, broader recognition of open source’s competitive value, and a National AI Action Plan that assigns development primarily to Washington while leaving harmful in-state conduct to state enforcement.
The less-publicized labor provisions matter because the speakers do not claim certainty about disruption. Productivity gains have historically benefited labor overall and remain their expected “direction of travel,” but the plan’s retraining programs and labor-market monitoring preserve the ability to respond if AI displacement proves severe.
McCune calls the rhetorical reset as consequential as individual rules. The prior posture was “only focus on safety with a splash of innovation”; the new one places greater emphasis on ensuring that America wins in AI while keeping people safe. That signal influences agencies, Congress, and foreign governments.
China policy creates a boundary problem for open source. McCune supports constraining private U.S. investment into Chinese companies and preventing the PLA or CCP from exploiting powerful U.S. technology, but warns against rules that effectively prohibit globally distributed open models. DeepSeek, in the speakers’ telling, already undermined the premise that capabilities could simply be locked away.

7. The moratorium failed because its coalition was not organized

The proposed moratorium became understood as prohibiting every state AI law for 10 years, a reading McCune disputes while conceding that “perception is reality.” Its reconciliation vehicle guaranteed a partisan contest, leaving margins narrow enough for one or two Republican senators to sink it.
His larger postmortem is organizational: supporters of some moratorium or preemption were not coordinated enough to explain the text, counter “FUD,” or align big, medium, and little technology interests. The intervening three or four months were spent building that coalition for the next contest.
The political component is now explicit. a16z donated to Leading the Future PAC, envisioned as a federal, state, and local “center of gravity” for advocacy aimed at sustaining U.S. AI leadership; McCune expects additional participants to join the effort.

8. Federal rules should govern models; states should police conduct

Perault grounds the division in interstate commerce: Congress should lead governance of the national AI market and model development, while states retain an “incredibly important role” policing criminal and other harmful conduct within their borders. Federal leadership does not mean state inactivity.
State rules that control developers elsewhere may encounter the dormant Commerce Clause. Perault describes a balancing test between burdens on out-of-state commerce and local benefits; the aim is not to disable legislatures, but to guide them toward enforceable harmful-use laws instead of exporting costly development mandates nationwide.
McCune’s priority for the next six to 12 months is targeted federal preemption, “not preempting all state law”: one national framework for model regulation and, ideally, model use rather than a 50-state patchwork. Adjacent priorities include workforce training, AI literacy, data centers, and energy.
The affirmative Little Tech program would increase enforcement capacity, clarify that AI is no defense under existing civil or criminal law, train officials technically, and create public resources that reduce startup barriers to compute and data. Perault cautions that alignment with Big Tech is conditional: shared support for federal standards may give way to renewed divergence when people invoke “industry agrees” without representing smaller builders.

The Hidden Truth About the US Energy Grid: Subsidies, Demand & Rising Prices - Chris Wright Yummy Century Egg's Guowei Zhang Echostar follow up $SATS