For years, the canonical VC question was: "What if Google came after you?" And for years, the standard founder answer indexed on initial market size: "Google's got bigger fish to fry."
Historically, that held. Large companies could not effectively resource hundreds of teams to pursue small markets. This created space for startups.
As Clayton Christensen argued in The Innovator's Dilemma, startups attack from below. They enter at the edges with a new user, workflow, interface or distribution model that sits outside the incumbent's core profit pool. Over time, the wedge expands until the startup can compete directly.
Incumbents, meanwhile, defend themselves, among other ways, through large, measured bets outside their core. Think Meta with Reality Labs, Google with DeepMind or Amazon with AWS. They go sideways, a manoeuvre that has historically been a privilege of scale.
But AI changes things. In a world where any employee can direct a swarm of agents, large companies can now pursue opportunities that were previously too small to justify. A single employee, given enough compute, can increasingly do the work of an entire team and, soon enough, an entire company.
A small market is therefore no longer, by itself, a safe haven for startups. Attacking from below may still be how a company sets off, but it is no longer safe to assume that a narrow initial TAM will keep incumbents away.
To the great benefit of startups, AI does not make an incumbent nimble. Most still operate with legacy processes and workforces far too large for their own good, built for a prior world. While headcount scale, when coupled with agents, gives large organisations enormous output capability, output and organisational velocity are not the same thing. One need look no further than the quick death of tokenmaxxing and the slopification of the enterprise that it caused.
In an output-abundant world, deciding what not to build is as important as what you do build. When every employee can do the work of many, a 10,000-person org looks like a 100,000-person org, and a 100,000-person org looks like a 1,000,000-person org. All of this means more decisions, more dependencies, more risk of duplication. In the age of AI, coordination complexity grows exponentially with headcount. It is ultimately far easier to build up with agents than to scale down with them.
That is the value of being small today. You can operate at great velocity, coordinate effectively and, quicker than ever before, shift sideways to capture adjacent opportunities. We're entering the era of small giants.
These companies combine focus and speed with the desire and ability to take big swings into new territory. Like the startups that came before them, small giants begin with a narrow wedge, establish a strong core and gradually move from being single-product to multi-product.
@arlanr at @nozomioai, a LocalGlobe portfolio company, describes this as operating like a lab.
"Before AI, the consensus was to pick one niche and drill down as far as possible. Post AI, any company can become a product lab shipping adjacent software at scale."
Rather than directing all of Nozomio's energy towards Nia, its first product, Arlan has already moved on to the next adjacency with @try_folk.
@tryramp is formalising a similar model through @RampLabs, a group within its Applied AI team with no predetermined problem to solve. Its mandate is to explore new technologies and turn promising experiments into products. Nozomio shows the model at startup scale. Ramp shows what it can look like as an organisation matures.
@midjourney tests the edge of the definition. Best known for its image generation models, it is now using its cash flows to develop a full-body ultrasound scanner. While the leap from image models to medical hardware is a large one, the underlying pattern is no different to Nozomio and Ramp's. For what it's worth, I expect more small giants to follow a similar path. Given a successful, cash-generative core, there is little stopping a company from entering entirely new and unrelated markets. After all, this is what the giants of yesterday have done.
Constraints on the model naturally still exist: 1) rapid product judgement (as is true of any business today) and 2) compute. Small giants need to decide quickly which bets to pursue and which to kill, while keeping close tabs on token spend. More tokens spent, assuming good judgement, means more quality bets running in parallel.
Incumbents are desperate to reorganise around this way of working. They have distribution advantages, more capital and therefore access to far more compute. But they're now bottlenecked by the bloat of their organisational structures. They need to rebuild.
In the old playbook, startups attacked from below and hoped incumbents ignored them until it was too late. In the new playbook, winning startups will be the ones that coordinate humans and teams of agents to go sideways quickly, capturing increasing value along the way. It is no longer a privilege of scale.