There is a metaphor that has been turning over in my head since the Anthropic legal launch two weeks ago. I have not been able to stop thinking about it. So I want to write it down.
Imagine an ant and an elephant on the same path. The ant starts walking ten days before the elephant does. By any human-scale measure, the ant has an enormous head start. Ten days is a lot. The ant has covered ground, formed strategy, learned the terrain. Surely this matters.
The elephant takes one step. The ant’s ten days are over.
This, I have come to think, is the most accurate description I can offer of what is happening to small companies and early innovators in the AI era. The traditional logic of business—that early movers are rewarded, that pioneers earn their lead, that being first creates compounding advantages—has reversed. In the AI era, going first does not produce reward. It produces a free market validation service that the big platforms then absorb.
I want to write about why this is happening, what it means, and why I find the conclusion genuinely disturbing.
The Old Logic of First-Movers
For most of business history, going first was rewarded. The first company to identify an opportunity, build the product, find the customers, and establish the brand had a structural advantage that competitors found difficult to overcome. Network effects compounded. Brand recognition compounded. Operational expertise compounded. Customers built habits around the pioneer’s product. By the time competitors arrived, the pioneer had a moat made of all the things that take time to build.
This logic was the basis for venture capital. It was the basis for “move fast and break things.” It was the basis for the entire startup ecosystem of the past four decades. Pioneers were rewarded because being first was, in itself, an advantage that was hard to replicate.
The AI era has inverted this logic. The advantages that took years to build for previous generations of pioneers can now be absorbed by a large platform in a single product release. The structural reason this is true is worth examining, because it is not accidental.
What the AI Platforms Have That Pioneers Don’t
A large AI platform—Anthropic, OpenAI, Google, Microsoft—has four things that a vertical-specific startup pioneer does not.
The first is user base at scale. When Anthropic releases a legal product, it lands in front of millions of existing Claude users. A legal startup’s product lands in front of the few thousand lawyers it has managed to acquire. The asymmetry of reach is roughly a thousand-to-one. When the platform adds a feature, every existing user automatically sees it. The startup has to fight for every new user.
The second is media attention. When a platform releases a feature, the technology press covers it. The industry press covers it. The general press covers it. When a startup releases the same feature, almost no one covers it. The asymmetry of attention is even larger than the asymmetry of reach. A platform feature launch is heard around the world. A startup feature launch echoes in a few specialty publications.
The third is capital. Large platforms can spend a hundred million dollars to build a feature that a startup raised a hundred million dollars to specialize in. The platform’s hundred million is a small fraction of its overall capital allocation. The startup’s hundred million is its entire existence. The platform can be wrong and survive. The startup cannot.
The fourth is integration. This is the most important one. The platform can build the feature directly into existing user workflows—Microsoft Office, Google Workspace, the user’s email client, the user’s existing AI tool. The startup has to build a separate application that asks users to leave their existing workflow. This is the friction problem I have written about before. It is the structural disadvantage that all vertical startups face against general platforms.
The combination of these four factors means that any feature a startup builds, that proves to be valuable, becomes a target for absorption by the platform. The startup has done the hard work of proving the feature matters. The platform then builds it, distributes it through its existing user base, gets media coverage for it, and integrates it into existing workflows. The startup’s “moat” of being first dissolves.
The Anatomy of the Punishment
The mechanism by which early innovators get punished follows a remarkably consistent pattern.
Phase one: A founder identifies a specific opportunity that general AI platforms have not yet addressed. They build a product around it. They raise money based on their thesis. They acquire early customers, generate revenue, demonstrate that the opportunity is real.
Phase two: The platforms watch. They do not have to be the first to identify opportunities; they can wait for startups to identify them. They observe which products are getting traction, which use cases are generating real revenue, which workflows are getting adopted. They are running a free market research operation, paid for by the startups themselves.
Phase three: The platform builds the feature. It does not have to be perfect; it just has to be good enough, distributed widely enough, and integrated deeply enough into existing workflows. The platform announcement is covered by every relevant publication. Customers who had been considering the startup’s product now see the platform’s version inside the tool they already use. They do not switch wholesale; they just slowly stop adopting the startup’s tool for new use cases.
Phase four: The startup discovers, six to twelve months later, that its growth has stalled. Existing customers stay (for now), but new customer acquisition has dried up. Investors who would have funded the next round are now asking whether the company has a long-term defensible position against the platform. The valuation that seemed reasonable a year ago is now suddenly aggressive. The company has to choose: pivot, sell, or continue with diminishing returns.
I have watched this play out, in compressed form, in the legal AI space over the past eighteen months. I am watching it play out, in real time, in the wake of the Anthropic launch. The pattern is structural, not coincidental. It will keep repeating.
”Nothing Grows Under the Big Tree”
There is a Chinese expression I have heard from colleagues in similar conversations: nothing grows under the big tree. The shade of a large tree prevents anything else from growing. The big tree gets all the sunlight; the smaller plants under it get nothing.
This is what is happening in AI. The big platforms—each one is a tree—are absorbing the sunlight that smaller companies need to survive. The ground under them is not empty by choice; it is empty because nothing can grow there.
The deeper problem with this metaphor is what it implies about innovation itself. If pioneers are systematically punished for going first, then over time, fewer people will be willing to go first. The supply of new ideas that platforms can absorb will diminish. The platforms themselves, which depend on this supply for their continued evolution, will slow down. The whole ecosystem becomes less innovative.
This is not a hypothetical concern. I am already hearing it in conversations with founders in the legal-tech space. People who would have started companies a year ago are now hesitating, asking themselves the question: “what is the path by which my company survives the platform absorbing my idea?” Most of them cannot find a good answer. Many of them are deciding not to start.
The big tree is not just blocking sunlight from current plants. It is preventing seeds from sprouting in the first place.
What This Means for Founders
If I were advising a founder in any AI-adjacent space today—and as a lawyer, I do talk to founders—I would offer three uncomfortable observations.
First, your moat probably does not exist. What you think is your defensible competitive advantage is likely a temporary lead. Test this honestly: would a major platform, with three months of focused work, be able to replicate your core capability? If yes, you have a temporary lead, not a moat. Plan accordingly.
Second, your exit strategy may need to be planned from day one. The traditional path of “build, scale, IPO” is harder to access when platforms can absorb your innovation before you reach scale. The realistic path may be to build a product that is attractive specifically as an acquisition target—either by the platforms that would otherwise absorb you, or by larger players who want to own the specific niche you are creating.
Third, consider whether the smallest niches are the safest. Counterintuitively, the markets that are too small for the platforms to bother with are the markets where you can survive. If your total addressable market is two billion dollars, you are interesting to the platforms. If your total addressable market is twenty million dollars, you are beneath their attention. The fortress is not in being big; it is in being small enough not to attract attention.
This is, frankly, dispiriting advice. Building a small, defensible niche business is not what most founders sign up for. The startup mythology is about being the next Google, not the next sustainable mid-sized company that the platforms ignore. But the mythology was built in an era where the platforms did not absorb everything they touched. That era is ending.
What This Means for Lawyers
I write about this on a site for lawyers, so let me close with the lawyerly version.
If you are advising a founder who is considering an AI-adjacent business, the questions you should be asking them have changed. The traditional questions—about market size, competition, fundraising—still matter, but they are now overlaid by a meta-question: what is your relationship to the dominant AI platforms in your space?
If the answer is “we compete with them,” the next question is: how long can you compete before they absorb you?
If the answer is “we complement them,” the next question is: how dependent are you on their continued goodwill?
If the answer is “we ignore them,” the next question is: are you sure the space is small enough for them to ignore you back?
These are not legal questions, technically. But they are the questions that determine whether your client’s investment, your client’s career, your client’s company will exist in three years. They are the questions a good lawyer would help a founder think through before they sign the term sheet.
The ant is not going to outrun the elephant. The question is what the ant should do, knowing that.
This is the fourth and final piece in a series of reactions to Anthropic’s Claude For Legal launch and what it reveals about the legal-tech industry. Earlier pieces examined the announcement itself, the role of connectors, and the question of vertical moats.
Email [email protected] if you are a founder facing this situation or a lawyer advising someone who is. I will not publish without permission, but I read everything.