The idea of “picks and shovels” comes from the California gold rush: the most reliable profits were not made by the prospectors chasing gold, but by those selling the tools. In the current wave of artificial intelligence, a similar dynamic is emerging. Companies like OpenAI and Anthropic are not just participants in the software market; they are building the infrastructure that may redefine its economics altogether.
At the center of this transformation is a simple but powerful force: competition. As more large players invest billions into training ever more capable models, the marginal cost of generating software begins to approach zero. Once a model is trained, producing another line of code, another feature, or even an entire application becomes almost free. When multiple firms compete on this basis, pricing pressure intensifies rapidly. What was once a labor-intensive, high-margin activity risks becoming commoditized.
This is where the pick-and-shovel analogy becomes particularly relevant. The real leverage lies not in writing individual applications, but in owning the platforms that make software creation trivial. Foundation model providers, cloud platforms, and tooling ecosystems effectively become the suppliers of “infinite developers.” Their business is not selling software in the traditional sense, but enabling its near-costless creation at scale.
For developers, this raises an uncomfortable question: if software becomes abundant, what happens to the value of those who produce it?
One plausible outcome is increasing market concentration. When marginal costs fall toward zero, economies of scale dominate. Large players can spread the immense fixed costs of training and operating models across vast user bases, while smaller players cannot. This dynamic has already played out in cloud computing and search, and there is little reason to believe AI will be different. A handful of dominant platforms could emerge, capturing a disproportionate share of value, while most developers see shrinking returns.
However, this does not necessarily mean that all developers “gain almost nothing.” Rather, it suggests a shift in where value is created. If code itself becomes cheap, differentiation moves elsewhere: domain expertise, distribution, trust, integration, and user experience. Writing code may no longer be the scarce skill; understanding problems deeply and embedding solutions into real-world contexts may be.
There are also areas of software development that are likely to resist full commoditization. Highly specialized systems, particularly those embedded in complex organizational processes or regulated environments, tend to have structural barriers. Enterprise resource planning, industrial control systems, and mission-critical infrastructure are not easily replaced by generative tools alone. Their value lies not just in code, but in accumulated knowledge, compliance, and deeply embedded workflows.
This brings us to companies like SAP. Firms of this kind occupy a unique position. Their products are not merely software; they are institutional backbones. Replacing them is costly, risky, and often politically complex within organizations. Even if AI dramatically reduces the cost of building competing systems, the switching costs remain high. Data migration, process reengineering, and organizational change create inertia that protects incumbents.
That said, incumbency is not immunity. Large enterprise vendors will face pressure to integrate AI deeply into their offerings. If they fail to do so, they risk being outflanked by more agile competitors who can leverage AI-native architectures. But if they succeed, they may actually strengthen their position, using AI to enhance their already entrenched ecosystems. In that sense, AI could reinforce existing monopolies rather than dismantle them.
For the single developer in a small company, the picture is more nuanced. On one hand, AI dramatically increases individual productivity. A single developer can now achieve what once required a team. This lowers barriers to entry and enables rapid experimentation. Niche products, micro-SaaS offerings, and highly tailored solutions become more feasible.
On the other hand, the same tools are available to everyone. What one developer can build, thousands of others can replicate almost instantly. This leads to a flood of similar products, intense competition, and downward pressure on prices. In such an environment, success depends less on technical execution and more on differentiation through insight, branding, or access to specific customer segments.
In effect, the role of the developer begins to resemble that of a designer or entrepreneur rather than a pure engineer. The technical challenge of building software recedes, while the strategic challenge of deciding what to build becomes paramount.
The pick-and-shovel providers sit above this dynamic. They benefit from every experiment, every failed startup, and every successful product built on their platforms. Whether a particular application succeeds or fails matters less than the aggregate demand for the underlying capability. This creates a powerful feedback loop: as software becomes cheaper to produce, more of it is created, which in turn increases demand for the infrastructure that enables it.
Yet even this layer is not immune to competition. The race among major AI companies is intense, and it is precisely this competition that drives marginal costs down. Ironically, the pick-and-shovel suppliers themselves may face commoditization pressures over time, particularly if open-source models and alternative infrastructures continue to improve.
The broader economic implication is a shift from scarcity to abundance in one of the core inputs of the digital economy. When software becomes effectively free at the margin, value migrates to other bottlenecks: attention, trust, proprietary data, and physical-world integration. Companies that control these bottlenecks may become the true winners of the AI era.
In this landscape, the future is unlikely to be a simple dichotomy of dominant giants and impoverished developers. Instead, it will be a stratified ecosystem. At the top, a small number of infrastructure providers capture large, stable returns. Around them, powerful incumbents adapt and integrate AI into their existing domains. Below that, a long tail of developers and small कंपनies compete in a highly dynamic, low-margin environment, where success is possible but far from guaranteed.
The pick-and-shovel theory suggests that the safest place to be is not necessarily where the gold is found, but where the tools are made. In the age of AI, those tools are models, platforms, and ecosystems. But as history shows, even the suppliers of tools must continuously innovate, because in a world of near-zero marginal cost, no advantage remains unchallenged for long.