Regulate AI Use, Not Development: Believes that regulation should target the harmful application of AI technologies rather than imposing licensing or other restrictions on the development of the models themselves [45].
Enterprise AI Requires External Expertise: Argues that "out-of-the-box" AI solutions are largely ineffective and that successful enterprise deployment requires specialized, externally-driven teams for integration and fine-tuning [1, 2, 32].
Pro-Competition via Open Source: Strongly supports open-source AI and government aid for startups (e.g., compute access) as essential tools to foster competition and prevent market domination by a few large technology companies [17, 24, 42].
Human Feedback is Irreplaceable: Rejects the idea that synthetic data will soon eliminate the need for human feedback in training and refining AI models, viewing human-in-the-loop systems as critical for success [10, 9].
In-Person Work Drives Higher Productivity: Posits that transitioning from a fully remote to a primarily in-person work model can lead to an "exponential increase in productivity," based on his experience at Invisible [6].
Pre-2023 (approx.)
Invisible, the company Matt represents, operated for nine years having raised only $7 million, indicating a period of slow, capital-efficient growth before the recent AI boom [11].
Recent Past
Matt describes a period where the dominant AI policy discussion, even among industry players, involved proposals for strict licensing of frontier models, akin to nuclear energy regulation [19].
Most Recent Legislative Session
Observes that despite significant debate, major state-level AI bills failed to pass, with Colorado's controversial risk-based framework being the sole exception [13, 14].
Current (Policy Shift)
The policy discourse has shifted. Matt identifies a growing bipartisan consensus favoring open-source AI to foster competition [42] and notes significant political pushback against Colorado's new AI law from within the state's own government [16].
Current (Business Growth)
Invisible has entered a rapid growth phase, raising $130 million [11], securing 12 enterprise deals in 45 days [8], and shifting from a remote to an in-person model to boost productivity [3, 6].
▶The Enterprise AI Implementation Gap
Matt consistently highlights the significant chasm between the hype surrounding Generative AI and its actual successful deployment in enterprises. He uses data to show low operational rates (5% of deployments are working), poor accuracy of out-of-the-box agents (as low as 33%), and high project cancellation forecasts (40%) [30, 32, 37].
Investors should be wary of companies promising seamless, off-the-shelf AI solutions and instead focus on those providing deep, human-assisted integration, which Matt claims are twice as effective and address this clear market failure [1].
▶Regulating Application, Not DevelopmentApr 2026
A core tenet of Matt's policy stance is that AI regulation should focus on penalizing harmful uses rather than restricting the creation of the technology itself [45]. He views proposals to license frontier models like nuclear energy as misguided and believes existing laws already cover many potential harms [19, 43].
This theme reflects a strategic venture capital perspective aimed at preventing regulatory capture by large incumbents and keeping the barriers to entry for model development low, thereby fostering a competitive environment for startups.
▶The Pro-Competition AI AgendaApr 2026
Matt advocates for policies that actively support competition against the handful of large tech companies dominating the AI landscape. He champions open-source AI as a key driver of innovation and supports federal initiatives to provide startups with access to crucial resources like compute and data [24, 42].
This focus on competition signals a clear strategy to create a fertile ground for new market entrants, directly challenging the technological and resource-based moats of companies like Google, Microsoft, and OpenAI [17].
▶The Human-in-the-Loop Imperative
Matt consistently argues against the idea that AI will soon operate autonomously, particularly in complex enterprise settings. He is skeptical that synthetic data will replace human feedback [10] and his company, Invisible, is built on integrating a massive human expert marketplace into AI workflows to ensure quality and effectiveness [9, 38].
This perspective indicates a belief in a significant and durable business opportunity in data labeling, verification, and AI fine-tuning, suggesting the need for human intelligence to guide and correct AI systems will grow, not shrink.