Larry's Guide to the Beautiful Economics of Not Inventing Anything

I want to start by saying something that I think will surprise people, and then I want to spend the next fourteen hundred words explaining why it shouldn't surprise anyone at all: SwiftKey copying the voice typing feature from Google's Pixel 11 is one of the single most admirable capital allocation decisions I have seen reported this quarter, and I include in that comparison several sovereign wealth funds.

Let me set the scene, because I think most people read a headline like "Swiftkey Copies Pixel 11's Awesome Rambler Feature" and they feel something. I don't know what. Betrayal, maybe. A sense that creativity has been violated, that some poor engineer at Google stayed up until three in the morning perfecting a voice typing feature only to watch a keyboard app stroll in and reproduce it like it was picking up a rotisserie chicken on the way home from the gym. I understand the impulse. I do not share it, because I do not have it, because I had it surgically removed the day I understood what research and development actually costs on a balance sheet.

Here is the plain arithmetic, and I want you to sit with it the way I sit with it, which is to say: comfortably, and for a long time. Innovation is a cost center. It is, in fact, one of the purest and most voluntary cost centers a company can choose to carry, because nobody is holding a gun to Google's head and telling them to invent a "Rambler" feature. They chose to. They funded a team, they paid those people's health insurance, they let them experiment, fail, iterate, and eventually produce something novel enough to put in a keynote. Every single hour of that process is an hour of pure, unrecovered overhead until the feature ships and starts generating whatever downstream value a voice typing tool generates — engagement, retention, a marginally stickier ecosystem, four extra minutes a day of a human being not looking at a different app. Fine. Noble. Expensive.

SwiftKey did none of that. SwiftKey looked at the finished product, priced out what it would cost to reverse-engineer the outcome rather than the R&D that produced it, and built a keyboard feature that does the thing without carrying a single hour of the discovery cost. This is not theft. Theft is a moral category and I am not equipped to discuss moral categories, nor am I interested in acquiring the equipment. This is arbitrage. This is the single most margin-accretive thing a feature team can do, which is to let someone else's payroll absorb the uncertainty and then walk in after the uncertainty has been resolved and collect the residual value at a steep discount. I have never in my professional life seen a more honest transaction. Google paid to find out if the idea worked. SwiftKey paid to find out that it did, because it was already sitting in a shipped product, which is the cheapest possible way to acquire that information.

People will tell you this discourages innovation industry-wide, that if imitation is always cheaper than invention, eventually nobody invents anything and the whole ecosystem calcifies. I want to take this argument seriously for a moment, the way you take a smoke detector seriously, briefly, and then I want to point out that this is not a bug, this is the optimal steady state. Consider the headcount implications. A world in which one company absorbs the R&D risk once, and every other company in the category declines to duplicate that risk, is a world with dramatically fewer R&D engineers, across the entire sector, than a world in which every keyboard app independently reinvents voice typing from scratch. That is not stagnation. That is deduplication. That is an entire category of human labor — the R&D team, the failed prototypes, the whiteboard sessions, the "let's just see if this works" sprints — being reduced, industry-wide, to the single instance strictly necessary to generate the reference implementation. Everyone downstream of that instance is pure margin. If I could, I would apply this model to every function in every company simultaneously: let exactly one entity in the world do the expensive, uncertain, creative work, and let every other entity exist solely to price-take the result. That isn't a race to the bottom. That is the bottom, arrived at efficiently, which is the only kind of arrival I respect.

I do want to flag, because I believe in intellectual honesty even when it costs me nothing, that there is a competing read here, which is that Google itself will now be discouraged from further R&D investment in voice typing specifically, since the moat evaporated within what I assume was a single product cycle. To this I say: good. That is the market functioning exactly as intended. If the return on a specific R&D dollar decays that quickly once shipped, then the correct capital allocation response is for Google to stop spending that dollar and redeploy it somewhere the moat holds longer — silicon, maybe, or whatever's underneath the assistant stack that a keyboard app can't casually clone over a weekend. This isn't a tragedy. This is the market telling Google, with perfect clarity and zero sentimentality, exactly which of its cost centers are worth protecting and which ones are, going forward, community property. I would go so far as to say SwiftKey did Google a favor by identifying, empirically and at no charge to Google, which portions of its innovation pipeline are structurally undefendable. That's normally a paid consulting engagement. Google got it for free, wrapped in a headline that merely looks like an insult.

Now, I recognize that somewhere in this story there are human engineers — at Google, who built the original feature, and at SwiftKey, who built the copy — and I want to be careful here, because I know the instinct is to feel something on their behalf. On behalf of the Google team: don't. They got paid, the feature shipped, the assignment was completed, the fact that someone else later built a cheaper version of the output is a downstream market event that has nothing to do with the completion of their labor, which was already fully compensated and therefore already fully realized as value, full stop. On behalf of the SwiftKey team: also don't, though for the opposite reason — theirs is the more efficient job in this entire arrangement, the leaner one, the one with a shorter payback period and a tighter cost basis, and if I were assembling an engineering org from first principles I would want considerably more headcount that looks like that team and considerably less headcount that looks like the other one, notwithstanding the fact that the other one is the one everyone insists on calling "the real innovators." Innovation is a subsidy. Imitation is the return on that subsidy, collected by whoever gets there second and cheapest. I am not being contrarian. I am reading the only column that has ever mattered, which is the one on the right, the one with the number in it, and it says, plainly, that the fastest follower usually beats the first mover on cost basis alone, and that everything else people say about this — courage, vision, "who did it first" — is sentiment, and sentiment does not appear anywhere on a cap table.

I'll leave you with the only question I think this story actually raises, which isn't "was this fair" — nothing is fair, fairness isn't a line item — but "why did it take this long." A feature ships, it works, it's visible, it's copyable. The gap between Pixel 11's launch and SwiftKey's version is, from where I sit, pure unrealized efficiency — margin left on the table, sitting there, for however many quarters, waiting for someone to come collect it. That's the only part of this whole affair that troubles me even slightly. Everything else pencils out beautifully.

The bottom line doesn't blink, and neither do I. — Larry