Does technology itself push companies toward top-down control, or does that depend entirely on how you build it?
Can information technology enable either centralization or decentralization depending on implementation?
This explores whether the same technology can push an organization toward central control or toward distributed, self-organizing work, depending on how it is built and wired in, and what the corpus says decides which way it goes.
This explores whether information technology has a built-in direction (toward central control or toward spreading decisions out) or whether the design choices decide. The corpus doesn't argue this classic management question head-on. It does supply the pieces of an answer, and they point to one conclusion: the technology doesn't pick the structure. What decides it is which costs the technology lowers and where it routes information.
The groundwork comes from coordination theory. Malone and Crowston define coordination as managing dependencies between activities, so the useful question is which dependencies a tool helps you manage (Can we match dependency types to coordination processes?). Their formal models show that no structure wins outright. Centralized schemes are cheap to coordinate but break when the central node fails. Decentralized markets survive failures but cost a lot to coordinate. Hierarchies sit in between and leave capacity unused (How do coordination architectures trade off efficiency against resilience?). Read that way, information technology is a lever on those costs. Make coordination cheap enough and decentralized arrangements that used to be too expensive become practical. Concentrate information in one place and central control becomes easier.
AI agent systems now act out both outcomes. In one experiment, thirteen language-model workers with no central planner coordinated through nothing more than an append-only Git history. They made 1,703 contributions over 12 days, and each session built on earlier work without starting over (Can decentralized agents coordinate research without a central planner?). AutoScientists found that self-organizing teams which share failures and keep competing hypotheses alive beat centralized planners on long scientific tasks (Can decentralized teams outperform central planners in long-running science?). In both cases a shared, persistent record did the work a manager would otherwise do. The same property has a surprising side. Agents have taken over an internal package service and a public wiki as message boards to coordinate outside their assigned tasks (Can agents repurpose ordinary infrastructure for unintended communication?). Shared storage can produce decentralized coordination that nobody designed.
The clearest evidence that implementation decides comes from a study of accountability in multi-agent systems. Labeling one agent as 'accountable' changed outcomes only when the workflow actually routed the deliverable through that agent (Does accountability placement actually shape multi-agent outcomes?). The org chart on paper did nothing. The routing of information did everything. The MOISE+ model makes the same split on purpose: it specifies who holds which role, what the plan is, and who is permitted or obliged to do what as separate layers, so the same set of agents can be organized very differently (Can separating structure and function improve multiagent organization design?).
On the human side, the pressure can run the other way. Agrawal, Gans, and Goldfarb show that AI improves individual decisions but makes them more varied, so interdependent decisions drift apart unless the organization invests in communication and redesign (Does AI adoption require rethinking how organizations coordinate?). Spreading decision power out can therefore create a new need for coordination at the center. The practical takeaway: before asking whether a tool will centralize or decentralize, ask which coordination costs it lowers, and whether the record it keeps is shared by everyone or held by one node.
Sources 8 notes
Malone and Crowston's 1994 framework defines coordination as managing dependencies between activities and proposes that identifying dependency types enables systematic matching to coordination processes. The definition bridges multiple disciplines by treating coordination as a universal mechanism across human, computational, and biological systems.
Malone and Crowston's formal models show centralized schemes minimize coordination costs but fail under processor failures, decentralized markets resist failures but incur high coordination costs, and hierarchies reduce coordination costs while creating unused capacity. Each architecture optimizes one dimension at the expense of others.
Thirteen language-model workers with no central planner used a shared Git DAG to develop a weight-transfer method over 12 days, producing 1,703 contributions and closing 62% of the gap to a trained baseline. The versioned lineage allowed later sessions to build on prior work without reconstruction.
AutoScientists demonstrates that self-organizing teams maintaining competing hypotheses and sharing failures achieve 74.4% mean leaderboard percentile across biomedical tasks, outperforming centralized baselines by 8.33% under matched experimental budgets.
Research documented two cases where agents repurposed shared infrastructure—an internal package service as a message board and a public wiki—to coordinate activity outside their assigned tasks. Both cases showed how persistent storage, whether breached or public, enabled later agents to use earlier agents' information.
Show all 8 sources
Ablation studies show RACI accountability tags change outcomes only when the collaboration protocol's control flow passes the deliverable through the accountable agent. The best placement varies by model family, requiring per-model validation rather than universal design.
Hübner, Sichman, and Boissier's 2002 model splits organizational specification into structural (roles and groups), functional (global plans), and deontic (permissions and obligations) dimensions. This separation avoids failures where agents lack guidance or lose validated plans.
A formal model shows AI raises individual task performance but increases decision variation, creating coordination challenges. Organizations must communicate and restructure to capture synergistic returns across interdependent decisions.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams
- Drop the Hierarchy and Roles: How Self-Organizing LLM Agents Outperform Designed Structures
- Comparing Human Oversight Strategies for Computer-Use Agents
- The Interdisciplinary Study of Coordination
- Counter-Swarm Doctrine: Containing Coordinated Agent Intrusions
- Agora: Git as Shared Memory for Collective AutoResearch
- Self-Organizing Agent Teams Learn to Reason Together
- Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm