The termination shock: Where AI progress meets reality

To leave our solar system, a spacecraft must endure the termination shock, a region of space where the fiery solar winds of our Sun clash against the glacial currents of deep outer space. The termination shock can tear apart the most sophisticated and well-crafted probes and vessels, but overcoming it is the only way to explore the universe beyond our planets and Sun.

In October, I found myself at the second annual Progress Conference in Berkeley, California. Based on what I learned through its high-profile artificial intelligence (AI) track, AI progress, too, could be headed for a termination shock as it leaves the fast-paced environment of San Francisco and its tech industry and crosses the boundary into the real world of slow and thorny institutions. 

Crossing the threshold

Spirits at Progress Conference 2025 were generally high. Technological progress is moving fast, and many attendees arrived at the Lighthaven Campus eager to share stories of what they had built with the ambition sparked by the previous year’s event.

The conference was organized into tracks, each designed to go deep on a specific topic within the progress movement. The “AI Protopia” track exposed attendees to insights from leading AI researchers and development executives, as well as the United States’ top government official on AI — all of whom seemed united in their staunch belief in the prospect of progress through advanced, perhaps transformative, AI.

Despite this, much of the discussion on AI at the conference focused on the ice-cold peripheries of the technology, rather than the vanguard: the economy; the limits of atoms; legacy political institutions; the world of laws, regulations, and bureaucracy. All of these advance and adapt much more slowly than AI technology itself — and more slowly than conference attendees seemed to want.

This is perhaps the primary reason that many people believe AI will be merely a “normal technology,” rather than a world-changing one: No matter how quickly its capabilities advance, the frictions of the real world will slow tangible progress enough that AI isn’t able to have a truly profound impact for a long time. Those frictions — not AI models’ results on arcane benchmarks — might prove to be the primary obstacle preventing technical progress in AI from enabling economic and societal progress writ large, in preventing us from reaching the other side of the termination shock.

The government could be a critical bottleneck for maximizing progress in AI.

One particularly striking area where AI development is outpacing deployment is in the U.S. government. This was the theme of Progress Conference 2025’s “AI in the Machinery of Government” panel discussion, during which host Dean Ball, a senior fellow at the Foundation for American Innovation and a former senior policy advisor for the White House Office of Science and Technology Policy, hazarded a guess at government workers’ current AI adoption rate: “just short of double digits.” 

Study after study has found that AI is the fastest adopted consumer technology in history and that Americans are increasingly relying on it at work, with Gallup finding a doubling of adoption rates to 40% over the last year. Yet according to Ball, less than 10% of government workers take advantage of it. That gap suggests that the government could be a critical bottleneck for maximizing progress in AI — to do so, we need government officials to be as efficient as possible at everything from permitting and taxes to labor policy and military adoption.

The question in the back of my mind as I attended the many inspiring sessions at Progress Conference 2025 was this: Will the inability of important legacy institutions to keep up with AI progress prevent it from crossing the threshold between the lab and the real world? Or is the current pace and shape of AI progress soon headed for a termination shock?

Around the bottleneck

I still hadn’t been able to shake the question when I attended Tyler Cowen’s interview with OpenAI CEO Sam Altman. Altman spoke enthusiastically about new AI products, devices, models, and systems — all innovations now headed for the government bottleneck in the pipeline. 

I had the opportunity to speak with many researchers at the technical frontier of AI during the conference, too, and from these conversations, I got the sense that they genuinely believe transformative technological change is within grasp — that they can build AI systems that can cure disease and aging, mitigate poverty and climate change, and restore our abilities to build technological marvels at new heights. 

In the pursuit of technological progress, those at the frontier of AI development are doing their part — but are the rest of us?

When established service providers fail to adopt AI quickly, they might be supplanted by AI-first competitors.

Based on what I gleaned from the conference, I see two options for helping bring the pace of AI deployment in line with the pace of AI development. The first is to circumvent legacy institutions when they stand in our way — even if it leads to their downfall.

This is perhaps the most overtly “progress-coded” approach to the problem, and it’s reflected in the mindset of Blake Scholl, CEO of Boom Supersonic. In another Progress Conference 2025 conversation with Tyler Cowen, Scholl recounted his struggles with the American manufacturing establishment. When presented with the possibility of having to use a half dozen plants in different parts of the country to procure a single part — at a design-to-product time on the scale of months — Boom opted instead to bypass outside manufacturers and create the product in one fully integrated plant. This cut the procurement time down to just days.

During a panel discussion titled “AI in the Machinery of Government,” Samuel Hammond, chief economist at the Foundation for American Innovation, provided a regulatory example to go with Boom’s manufacturing one. 

Uber and other rideshare apps created a new business model for micromobility. Municipal taxi commissions — the local government organizations that regulate taxis — could have adopted the innovations, but because they didn’t, taxis have been rendered effectively irrelevant in many parts of the U.S. while ridesharing has thrived.

OpenAI CEO Sam Altman in conversation with Tyler Cowen at Progress Conference 2025. Credit: The Roots of Progress Institute

AI adoption might often work in similar ways. 

When established service providers — from lawyers and marketers to radiologists and therapists — fail to adopt AI quickly, they might be supplanted by AI-first competitors. When governments fail to adopt AI to more quickly and effectively deliver on their promises, citizens may use AI-mediated tools to coordinate more effectively outside the state’s purview — using data and modeling to find fairer, more efficient solutions to shared challenges. Housing, construction, and urban planning could come first, with infrastructure, welfare, and policing perhaps following.

Depending on where and how AI is deployed, it could render some legacy institutions obsolete — we might replace slow electoral institutions with fine-grained preference aggregation, overburdened courts with private arbitration, and outmatched law enforcement with more effective private hires, for example.

The hard path forward

But this is where things get tricky. Disrupting frustrating manufacturing plants or hidebound taxi commissions is different than disrupting — and rendering irrelevant — institutions that we may have fought long and hard for. Their institutional merits — still-high public trust, equitable structure, deep justification, and so on — are not easily reproduced. 

In some important areas, including government, we cannot rely on the disruptive path alone. We will need to opt for the second option: doing the hard work of reform. We must work with institutions to accelerate their adoption of AI. We need to actually walk the long and arduous path from a White House that is running one instance of the Gemma 7B model in a supply closet — as one panelist at the “AI in the Machinery of Government” discussion detailed — to a government that is an efficient machine powered by American-built frontier AI. 

Getting the future right will require the ability to build shortcuts at speed and the wisdom to realize when a shortcut is a mistake.

Progress Conference 2025 didn’t map out every turn along that path, but I still left it with a sense of optimism that the path does exist and that the nation, at large, is fit to identify and walk it.

Getting this right will require the ability to build shortcuts at speed and the wisdom to realize when a shortcut is a mistake, which is to say it will require thoughtful appreciation of the past and an unabashed willingness to steer the future in equal measure. That balance is what distinguishes the progress movement from both accelerationism and statism: It is deeply rooted in critical consideration of what society has done right in the past. In many conversations with fellow attendees, I could sense their desire to take seriously the challenge of building the best possible future without forgetting about the merits of past accomplishments. 

This propensity may be what distinguishes the progress advocate: a faith in progress and disruption, but also an appreciation for the fences they might tear down. I think that’s why the banner of progress has managed to rally a bigger and brighter crowd than most movements: It provides for thoughtful people of all stripes, united by a belief that the world on the other side of the termination shock is good enough to justify the challenge of navigating it.