The Specific Event

Andreessen Horowitz just led a $38 million Series A into Convey, a startup founded last year that builds what it calls AI "teammates" designed to absorb repetitive work from human employees. The pitch is straightforward: offload routine cognitive labor to AI agents so that workers can focus on higher-order tasks. The funding round is notable not just for its size but for its timing. It arrives precisely as a separate set of business news stories reports that AI token costs are forcing companies to rethink how they hire, budget, and manage compute usage. Companies that once encouraged unlimited AI use are now rationing it. These two stories are in direct tension with each other, and that tension points toward something that organizational theory has not yet adequately addressed.

The Coordination Problem Underneath the Pitch

When Convey frames its product as an AI "teammate," it is making an implicit claim about organizational coordination. The claim is that inserting an AI agent into a workflow is structurally similar to inserting a new human colleague. If that were true, the coordination problem would be familiar: onboard the agent, assign tasks, measure output. But the organizational evidence suggests this framing is misleading. Kellogg, Valentine, and Christin (2020) document how algorithmic systems at work do not behave like colleagues because they alter the conditions under which human workers develop competence. The algorithm does not just do work alongside humans; it reshapes what humans learn to do, what they stop doing, and which skills atrophy. A "teammate" framing obscures this dynamic entirely.

Token Budgets and the New Rationing Layer

The simultaneous news about AI token cost pressures makes this worse in a specific, underappreciated way. When compute is rationed, organizations must decide which tasks receive AI assistance and which do not. That decision is not neutral. It creates a new layer of internal inequality that mirrors what Schor et al. (2020) describe in the platform economy: access to the coordinating infrastructure itself becomes differentially distributed. Workers who receive AI assistance develop one set of competencies; workers who do not develop another. Over time, this divergence compounds. The power-law outcome distributions that my own research identifies on external platforms will begin to appear inside organizations as well, driven not by external algorithmic amplification but by internal resource allocation decisions.

Routine Versus Adaptive Expertise in the AI Teammate Model

There is a deeper problem with automating repetitive work specifically. Hatano and Inagaki (1986) draw a foundational distinction between routine expertise, which is procedural mastery of known task sequences, and adaptive expertise, which is the capacity to respond effectively to novel conditions. Routine expertise is precisely what AI agents like Convey's are designed to absorb. The organizational logic assumes this frees workers to develop adaptive expertise. But that assumption has no empirical basis in the current rollout. Workers do not automatically develop adaptive capacity just because routine tasks are removed. They develop it through deliberate practice, feedback, and schema formation. If organizations simply subtract routine work without investing in structured development of adaptive capacity, they are not upgrading their workforce. They are hollowing it out while billing it as augmentation.

The Folk Theory Risk Inside the Enterprise

What concerns me most about the Convey funding story is that it will accelerate a folk theory of AI collaboration that is already pervasive in management circles. That folk theory holds that AI handles the low-level work, humans handle the high-level work, and the division is clean. Rahman (2021) shows how algorithmic systems create invisible structural constraints that workers are systematically unaware of. The folk theory of clean human-AI task division functions the same way: it gives managers a confident but inaccurate schema for what is actually happening inside their workflows. Gentner's (1983) structure-mapping framework would predict that this folk theory will transfer across organizational contexts precisely because it appears structurally coherent, even when the underlying relational structure is wrong. The danger is not that Convey's product fails. The danger is that it partially succeeds while producing organizational dependencies that are not visible until a token budget gets cut or a model gets deprecated mid-project.

What the Research Suggests

The Convey investment is not simply a venture capital story. It is an early stress test of whether organizations can coordinate effectively when their coordination infrastructure is itself an AI agent with variable cost, uncertain continuity, and opaque decision logic. The ALC framework I am developing would predict that organizations which invest in structural schema training, helping workers understand why AI agents behave as they do, not just how to use them, will show better adaptive performance when those agents change or disappear. Organizations that treat the AI teammate as a black-box colleague will be left with workers who have lost routine competence but never built adaptive competence. That is not augmentation. That is a coordination failure dressed in Series A funding language.

References

Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science, 7(2), 155-170.

Hatano, G., & Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, & K. Hakuta (Eds.), Child development and education in Japan (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.

Rahman, K. S. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945-988.

Schor, J. B., Attwood-Charles, W., Cansoy, M., Ladegaard, I., & Wengronowitz, R. (2020). Dependence and precarity in the platform economy. Theory and Society, 49(5), 833-861.