Who the hell am I to argue with Bill Gates?

I will still add my $0.02 worth to his recent article (gatesnotes.com/a-turbulent-ai-era-and-critical-choices-to-make).

Bill Gates has written a thoughtful and somewhat unsettling piece about what AI means for work, capitalism and the institutions that have developed around them. His central proposition is difficult to dispute: AI is different from previous technological revolutions because it has the potential to substitute not merely for physical labor, but for human cognition; and unlike previous transitions, which unfolded over generations, this one may occur within a decade.

His conclusion is that we need to start thinking now about the institutional consequences: what happens to employment, education, taxation, social safety nets and, ultimately, the relationship between work and economic security.

I agree with the diagnosis. I am less certain about the prescription, although perhaps that is because I think there is a question that precedes it.

Which institutions do we actually want to preserve?

To be clear, I am not arguing for dismantling capitalism or democracy. I am an avowed fan of both. I also happen to think that the rule of law is one of the great institutional achievements of human civilization.

But there is a difference between preserving an institution and preserving its historical form. I am not persuaded that we should regard existing institutional arrangements as immutable simply because they have worked reasonably well under the technological and economic conditions in which they were developed.

In fact, I would argue that doing so is antithetical to both capitalism and democracy.

I was reminded of this at Investors for a Better America, where I had a fascinating conversation with a brilliant young woman whose name I will leave out (there was some reluctance among attendees to be publicly associated with the conference, and I respect that). We were discussing nonprofits, and found ourselves in complete agreement on the premise: the functions performed by nonprofits in today’s society are important and, in many cases, indispensable. Yet we arrived at almost diametrically opposed conclusions. Her thesis was that the existing institutional form should be strengthened, through regulation designed to encourage additional private-sector funding. My instinct was almost the inverse: if the function is important, then the function should be preserved; the particular institutional form through which we currently perform it is a separate question, and one considerably less worthy of preservation in the abstract.

That distinction has become increasingly important to me as I think about AI.

Institutions are technologies

We tend to reserve the word “technology” for things that engineers build. But institutions are technologies too. Markets, corporations, law firms, universities and governments are all mechanisms for organizing human activity around particular constraints and objectives - in other words, solving the coordination problem.

The constraints change.

Technology changes them faster.

AI is therefore not simply another tool that we can bolt onto existing institutions. It changes the economics of what those institutions do.

Consider the legal profession, which is where I spend my entire career before making the leap to venture.

A law firm is, among other things, an institutional response to the scarcity of legal expertise. It aggregates human capital, reputation, relationships and responsibility and deploys them against problems that require judgment and cognition.

Now introduce a technology capable of performing an increasing proportion of that cognitive work at a fraction of the marginal cost.

The obvious question is: how many lawyers do we need?

I think that is the wrong question (as I have said time and time again in the course of the #TrojanHorseDebate with Quentin Solt).

The more interesting question for me is: what is the optimal institutional configuration for delivering legal services when cognition is no longer nearly as scarce as it once was?

The answer may involve fewer lawyers. It may involve very different lawyers. It may involve much larger volumes of legal work being performed because the economics now permit it. It will almost certainly involve law firms owning less of their technological infrastructure and relying much more heavily on specialist technology companies.

None of that necessarily implies the demise of the legal profession. It implies that the profession will evolve in response to a change in the production function.

This is why I find the “AI replaces lawyers” formulation so reductive. It treats the current allocation of tasks as though it were somehow intrinsic to the institution.

It is not.

The institutional equilibrium is not fixed

There is a useful connection here to George Soros’s concept of reflexivity.

The relationship between technology and institutions is bidirectional. Technology changes the constraints within which institutions operate; institutions, in turn, determine how technology is financed, developed, adopted, regulated and integrated into society. Each changes the conditions under which the other evolves.

That makes the future considerably less deterministic than the current AI discourse sometimes suggests.

AI capability does not map mechanically onto institutional extinction. It changes the feasible set from which institutional arrangements can be constructed.

The economist in me keeps asking whether a superior equilibrium (whether in the "Pareto optimal" or in the "Kalder-Hicks efficient" sense of the term) is possible given the technological shift.

Sometimes the answer will be no. Existing institutions may embody accumulated knowledge, trust and mechanisms of accountability that are difficult to replicate and easy to destroy.

Sometimes the answer will be yes.

And sometimes the answer will be more subtle: the institution remains indispensable, but its historical configuration becomes obsolete.

That is where I suspect much of the AI debate is going wrong.

The two poles tend to be technological determinism and institutional conservatism. One assumes that sufficiently capable AI necessarily destroys existing institutions; the other assumes that because those institutions perform valuable functions, their existing form must be protected from technological disruption.

Neither follows.

We should preserve the functions that matter and remain agnostic about the form through which they are delivered.

This is not an argument against capitalism

If anything, it is an argument for taking capitalism more seriously.

Capitalism is extraordinarily effective precisely because it permits institutional experimentation. Businesses are created, capital is allocated, organizational forms evolve and inefficient arrangements disappear. We do not regard the corporation as a sacred object; we regard it as an extraordinarily useful mechanism whose form has itself evolved repeatedly.

There is no reason to suspend that logic simply because the institution in question happens to be a law firm, a university or a nonprofit.

Nor should we assume that every new technology requires an entirely new institutional architecture.

Sometimes the optimal outcome will be an existing institution incorporating new technology. Sometimes it will be a new intermediary emerging alongside it. Sometimes vertical integration will give way to specialization. Sometimes functions that were previously bundled together will be unbundled and recombined.

That is how markets work.

And that is, in a sense, how institutions have always evolved.

Back to Gates

I think Gates is right that AI forces us to think seriously about institutional consequences. But I would broaden the inquiry.

Before asking what new institutions we need, we should ask what existing institutions are actually doing, why they exist, and whether the technological conditions that justified their current configuration still prevail. (If you permit a slight digression - this is precisely the analysis that lawyers are involved in when challenging long-established precedent: does the rule still make sense, or has it become obsolete due to a change in the facts and circumstances?)

That is a very different exercise from dismantling capitalism or democracy. It is an attempt to understand whether we can move to a better equilibrium while retaining the things that make those systems valuable.

And that is how I think about the future of legal tech.

I do not believe the objective is to replace lawyers. Nor do I believe the objective is to protect lawyers from technology.

The objective is to discover what the legal services ecosystem should look like when the marginal cost of cognition is dramatically lower.

That may mean that some things that are currently expensive become ubiquitous. It may mean that some things currently performed by lawyers migrate to technology companies. It may mean that lawyers spend less time producing information and more time exercising judgment, assuming responsibility and navigating relationships. It may mean that entirely new categories of legal work become economically viable (that is the Legal Capability Frontier concept comes in).

I am not prepared to say that I saw the map of the future in a dream, like Mendeleev saw the periodic table of elements.

But that is precisely why I am reluctant to make categorical predictions about institutional extinction.

The more interesting question is not what AI will do to our institutions. It is what our institutions will become in response to AI, and what kind of AI ecosystem those institutions will, in turn, create.

That is the reflexive part of this.

Technology changes institutions. Institutions change technology. And the equilibrium that emerges is a product of both.

The stakes are considerably larger than the legal profession

I was also impressed by another conversation at the Investors for a Better America conference (thank you once again for the invitation, David Teten). I spoke with a gentleman who exuded both intelligence and confidence; he was clearly thinking several moves ahead of the conventional debate, and, rather than broadcasting his views, intends quietly to sponsor research supporting a policy change that he believes will be essential over the coming decades.

His objective is ambitious but straightforward: in a world dominated by AI, dramatically increase the number of “haves” while reducing the number of “have-nots.”

I found that conversation particularly encouraging because it framed the policy question correctly. The objective should not be to preserve the economic arrangements of 2026 against technological change. Nor should it be to allow technological change to determine the distributional outcome by default.

It should be to conduct enough scenario analyses to influence what the new equilibrium is likely to be, and to proceed about it in the most intentional manner.

That is a very different proposition from trying to stop technological progress.

And perhaps this is where Gates’s article ultimately points us: not toward a choice between capitalism and its alternatives, or democracy and its alternatives, but toward a much more difficult question of institutional design.

If AI fundamentally alters the economics of cognition, then the distribution of economic opportunity will depend, to a considerable degree, on the institutions through which that new abundance is mediated.

We should therefore be experimenting now.

Not because we know what the answer is, but because we do not.

So, to come full circle: who the hell am I to argue with Bill Gates?

Here is my best imitation of Jaqen H’ghar: “A man is no one.”

But here goes, regardless:

Gates is right that AI is going to force us to rethink institutions. I would simply add that we should be careful about assuming that the answer is always to build new institutions around the old ones.

Sometimes the better answer will be to redesign the old ones.

Sometimes it will be to let them die.

And sometimes it will be to preserve the institution while completely changing the way it operates.

The objective shouldn’t be institutional preservation; nor should it be institutional destruction.

It should be a better equilibrium.

One in which technology and institutions reinforce each other rather than one simply overwhelming the other.

And, lest I am accused of being overly theoretical about this:

I am completely immersed in legaltech investing. This is my full-time job, and I think about it 24/7. But if there is someone out there seriously working on the policy and institutional dimensions of this transition, I will find the 25th hour in the day to contribute what I have learned from inside the legaltech ecosystem.