We often confuse scale with progress.
Scale is expanding what already works: more markets, more distribution, more copies.
Progress is increasing what's possible.
A useful lens is to separate progress into two modes:
- Horizontal progress: copying a proven system and spreading it wider.
- Vertical progress: creating something fundamentally new, new capabilities, new primitives, new leverage.
Horizontal progress distributes value. Vertical progress creates new categories of value.
The Ceiling of Replication
At a macro level, horizontal progress resembles what globalization does: taking systems that work in one place and deploying them elsewhere. That pattern has generated massive wealth.
But replication has a ceiling. Distribution scales benefits and costs. If the base system is inefficient, wasting energy, attention, capital, or operational complexity, scaling it multiplies the waste. You get more output, but also more strain.
Vertical progress is what raises that ceiling.
The Illusion of Automatic Progress
There's a quiet assumption many of us inherit: "the future gets better by default."
The 20th century makes that belief feel true. But its prosperity came from breakthroughs that changed society's input/output ratio:
- electricity
- computing
- aviation
- antibiotics and modern medicine
- networks and the internet
These weren't better copies. They were new primitives. Replication helped distribute the gains, but the gains themselves originated with the invention.
What This Means for Builders (Especially in AI)
In software, horizontal progress often looks like:
- another CRUD SaaS with slightly better UX
- "AI wrappers" with thin differentiation
- niche clones of existing startups
- the same workflow repackaged for a new audience
These products can be useful and profitable. But they usually don't create new capabilities, only new distribution.
Vertical progress looks like work that changes what's feasible:
- new architectures and abstractions that simplify whole classes of problems
- infrastructure layers that make systems meaningfully more reliable, secure, or efficient
- tools that make complex workflows predictable and debuggable
- primitives that turn demos into dependable systems
In AI, vertical progress shows up less in prettier dashboards and more in hard engineering:
- evaluation systems (measuring quality beyond vibes)
- reliability engineering (handling failures, drift, uncertainty)
- cost/performance orchestration (routing, caching, model choice)
- memory and state that's inspectable and controllable
- tooling that makes "intelligence" more programmable
A lot of today's AI products are horizontal progress at API speed. That's not an insult, it's a phase. The enduring leverage comes from the vertical layer: the primitives that make AI systems reliable at scale.
Constraints Are the Point
Replication assumes we can scale indefinitely. In practice, we hit constraints:
- energy and compute
- attention and cognitive bandwidth
- capital efficiency
- infrastructure and operational complexity
If you scale an inefficient system, you scale the inefficiency.
Technology, when it's truly vertical, does the opposite: it increases output per unit of input. It bends constraints instead of multiplying them.
A Practical Test: Horizontal or Vertical?
Ask:
- If competitors copy your UI in a weekend, do you still have an advantage?
- Are you building a new capability, or just new packaging?
- Does your product get structurally better via feedback loops/evals, or is it static?
- Are you reducing a fundamental constraint (time, cost, error rate), or just moving it around?
- Can you explain differentiation without mentioning your niche?
If the answers point mainly to "distribution" and "packaging," you're playing a horizontal game. That's fine, just be honest about it.
The Builder's Responsibility
Horizontal progress distributes gains. Vertical progress creates them.
Both matter. But only one raises the ceiling.
What are you building that moves things forward, not just outward?
Top comments (0)