Customer is not (the only) king anymore: Ecosystem competition for the AI age

For most of the past decade, understanding the customer better than the competition was treated as the primary source of competitive advantage. That assumption still holds some weight, but it no longer explains the full picture. I've seen well-researched, genuinely capable products lose market share to less capable competitors, for a reason that has nothing to do with customer understanding. Standalone productivity tools keep losing adoption to weaker, bundled features inside Microsoft 365 or Google Workspace, not because those tools are less capable, but because they sit outside a workflow the customer already committed to. Product-market fit tells you whether someone needs what you built. It doesn't tell you whether it integrates into the systems they already run on, and that second question is deciding more outcomes than most teams account for. AI is accelerating this shift: it's lowering the cost of coordinating partner networks and creating an entirely new layer of tools built directly around a platform, not alongside it. If you're deciding where to allocate engineering and partnership resources this year, this is worth more attention than most roadmap debates get. Stay Ahead of the Curve! Don't miss out on the latest insights, trends, and analysis in the world of data, technology, and startups. Subscribe to our newsletter and get exclusive content delivered straight to your inbox. Subscribe Now Where the value actually compounds Network effects come in two forms. Same-side effects occur when a product becomes more valuable to a user as more users like them join. Messaging apps are the clean case: WhatsApp is worth more to me because the people I want to reach are already on it, and worth nothing at all if they are not. Cross-side effects occur when growth on one side of a platform raises the value on another. Visa and Mastercard run on this: merchant acceptance and cardholder adoption reinforce each other, each side expanding the value of the other. Apple's App Store runs on it too. iPhone owners are not valuable to each other the way WhatsApp users are. The App Store's scale came from independent developers continually extending what the platform could do through building complementary capabilities, and every additional developer raised the value of the handset for every user, without Apple building each of those capabilities itself. Few companies operate on one mechanism alone. Most draw value from a user network layer and a third-party extension layer simultaneously, often without having designed for both deliberately. I think identifying which one is actually driving growth is the more useful exercise than debating which model a company should pursue, because most companies are already running both, whether they've mapped it or not. What AI is actually doing to this Three shifts are underway, and they do not all point the same direction. Matching efficiency is the first. Recommendation and ranking systems reduce the time it takes buyers, sellers, or partners to find the right counterpart, which shortens the runway to self-sustaining cross-side effects. A platform now needs less raw participation before it feels useful, because less of that participation is wasted on bad matches. Coordination cost is the second. Onboarding, verification, and integration used to require significant manual effort per partner. Automation has cut that marginal cost enough that a company can realistically manage a partner network several times larger than it could a few years ago, without scaling headcount at the same rate. That widens the complement layer directly: more complements, at lower cost per complement. The third shift is different in kind. AI-native extensions, meaning agents, fine-tuned models, and copilots built directly against a platform's own data and APIs rather than layered on afterward the way a conventional plug-in would be, have become a standard listing category rather than a side experiment. OpenAI's GPT Store treats them as a core listing type. So does Salesforce's AgentExchange, which launched in March 2025 alongside the company's older AppExchange marketplace and then absorbed it entirely at TDX 2026 in April, when Salesforce retired the AppExchange name after two decades. What makes this third shift different is that it cuts against the first two. These extensions are far cheaper to produce than conventional applications were, which means complement count can inflate considerably faster than complement value. That is precisely the condition under which the crowding problem appears. A marketplace can be adding listings quickly and gaining very little, and the metrics that track cross-side growth will not distinguish between the two cases. Build it, join one, or both Three questions then tend to clarify the decision. Does the company already hold enough proprietary leverage, whether data, distribution, or an installed user base, that partners would join without significant incentive? Would building a platform actually cost less than joining one that already exists, once ongoing governance and maintenance are counted and not just the initial launch? And does participating in an external ecosystem sharpen what differentiates the company, or does it gradually reduce it to a commodity role inside someone else's platform? None of these has a single answer that applies across every company. What I'd push back on is treating it as a one-time decision. The underlying economics shift as coordination costs keep dropping, so build-or-join deserves periodic reassessment, not a default a company settles into once and never revisits. Openness only works if someone designs it Restrict a platform too tightly and partners never join. Leave it fully open and quality erodes within a year, sometimes faster, and faster still on an innovation platform where complements can now be produced cheaply. The companies that get this right tend to converge on a similar structure: tiered API access instead of an all-or-nothing gate; explicit boundaries on what partners can read, write, or infer from platform data, defined in advance rather than resolved after a problem occurs; and participation requirements, certification and revenue-share terms among them, that protect trust without adding enough friction to push smaller partners out before they gain traction. Shopify's Partner Program and GitHub Marketplace both operate this way, using graduated partner tiers and review processes rather than either extreme. It is worth being clear about what this governance is for. It is not only a trust mechanism. It is the lever that separates complement count from complement value, the thing that keeps a marketplace from filling with listings that add nothing and dilute the ones that do. On a platform where the marginal cost of producing a complement is approaching zero, that lever carries considerably more weight than it used to. Understanding the customer still matters. It was never the wrong priority. It was always only half the picture, and AI has made the other half, where a product sits relative to everything around it, cheaper to build and harder to ignore. Companies treating both as a single, connected strategy, rather than customer strategy first and ecosystem strategy as an afterthought, are the ones likely to still be standing wherever their competitors, and their competitors' partners, decide to go next.