As AI tools evolve from single products to labour markets, experts argue that the real value lies in managing contextual continuity, prompting proposals for a two-layer system to preserve user history amid increasing model competition.
The argument that artificial intelligence is becoming less like a single product and more like a labour market is gaining traction, but the sharper point is that the value in AI may lie less in the model itself than in the context around it. In the essay “Models Are Labour. Context Is Capital.”, Jagadeesh K says many users now juggle several tools for different jobs: a fast, low-cost model for summaries, a more capable system for reasoning and drafting and, in some cases, a local model for sensitive tasks. That workflow, he argues, is efficient in theory but costly in practice because every switch forces the user to rebuild the surrounding context from scratch.
The piece frames that problem as an economic one. Models are treated as labour because they are bought per token, compete with one another and can be replaced as capabilities shift. Context, by contrast, is described as capital because it accumulates over time in the form of project history, preferences and prior decisions. Once that information is moved from one system to another, the user loses time and often loses continuity as well. The result, the author says, is a switching tax that can erase the savings from using a cheaper model.
That distinction aligns with a broader body of research on AI and the distribution of economic value. A recent Axios analysis argued that the current AI investment boom is delivering more benefits to capital than to workers, with data-centre spending and related infrastructure producing relatively few jobs while labour’s share of national income continues to fall. Academic work has also reached similar conclusions from different angles. One study in PLOS One found that AI-enabled enterprises can raise efficiency and support innovation, but the distributional effects depend on how technology is combined with labour and capital. Another paper on arXiv suggested that cheaper AI may either weaken formal labour demand or expand it, depending on how easily AI capital substitutes for human work.
The essay also places today’s multi-model behaviour in a historical frame. Attempts to optimise across cloud providers promised flexibility and lower costs, but often ran into the practical realities of data gravity, fees and integration friction. The author argues that AI is different because calls are largely stateless and interfaces are converging, making switching between models far easier. Yet a new form of lock-in is emerging one layer higher, in the memory and conversation histories that sit inside individual vendors’ systems. In that sense, the bottleneck is no longer moving data, but preserving context.
The proposed answer is a two-layer design: one layer for routing tasks to the cheapest or most capable model, another for storing a user-owned context layer that travels with the query. The article presents that as a way to preserve continuity while allowing vendors to compete on labour. A separate commentary on AI pricing makes a related point, arguing that the market is moving from selling tokens and model access towards selling economically useful work. If that shift continues, the decisive advantage may not be which model answers a prompt, but whether the system already understands the work being asked of it. That is the article’s central claim: the best model may still lose to the best context.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
