Should rolling out AI at work feel less like installing software and more like onboarding a new coworker?
Why model AI integration as organizational socialization rather than tool installation?
This explores why some researchers argue that bringing AI into a company works more like onboarding a new kind of colleague, who has to learn the norms, be supervised and fit into how people coordinate, than like installing software and flipping it on.
This explores why it might be more useful to treat AI adoption as onboarding a new kind of colleague than as installing a tool. The corpus gives a clear reason. The hard part of AI integration is rarely the AI's raw capability. It's how the organization around it has to adjust. A formal economic model finds that AI makes individual tasks better but also makes decisions more variable from person to person. That creates coordination problems, and companies only fix them by communicating and restructuring across connected decisions Does AI adoption require rethinking how organizations coordinate?. A tool improves one job. A new participant changes how all the jobs fit together.
The 'tool installation' view also breaks down in practice. Deloitte's analysis argues that many agentic AI projects fail because companies bolt agents onto old workflows instead of redesigning the process from start to finish Does agentic AI succeed through process redesign or workflow layering?. Benedict Evans makes a related point. Making AI tools easier to build doesn't help if workers can't see which of their own tasks could be automated, or if adoption depends on decisions that span several departments Does easier tool-building actually solve enterprise adoption problems?. In both cases, the bottleneck is people learning how to work with something new, which is what socialization means.
Executives already seem to feel this. In a survey of over 2,000 leaders, 76% described agentic AI as more like a coworker than a tool. The catch is that it's really both. It replaces some labor and supports other labor, so it needs HR-style management and asset-style management at once Does agentic AI fit the tool-or-coworker framework?. One research thread suggests what earns the 'colleague' label. It isn't a smarter model. It's persistence: memory, reusable routines and continuity from one task to the next What makes an AI system feel like a colleague rather than a chatbot?. Like a new hire, a colleague builds up experience inside a particular workplace.
The socialization framing also explains where AI agents currently struggle. On a simulated-company benchmark, leading agents finished only about 30% of tasks. Social interaction with coworkers was one of the main failure points Why do AI agents fail at workplace social interaction?. The framing also shapes how you supervise AI. Humberd and Latham argue that, as with new employees, you need monitoring in place before incentives can do any work. With AI this is even harder, because it's less clear what 'good behavior' looks like and its reasoning becomes harder to follow over time Must AI monitoring precede incentive alignment as agents evolve?.
There's a surprising limit to the analogy. Socializing a human changes what they believe. Studies of AI agents interacting with each other find that agents change their *actions* when they know peers are present, but they don't actually converge in their language or ideas Do AI agents actually socialize with each other?. So an AI 'colleague' may adopt your organization's behaviors without absorbing its shared understanding. A related argument holds that real alignment with values requires grounding in the world and participation in social meaning, not just following goals written as text Can AI systems achieve real alignment without world contact?. The socialization model is useful because it shows where organizational work is needed. It also warns you not to assume the AI is being socialized the way a person would be.
Sources 9 notes
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.
Deloitte argues many agentic AI implementations fail because organizations layer agents onto old workflows rather than redesigning operations. Success requires agent-compatible data architectures, robust orchestration, and new management models; HPE and Toyota illustrate this approach.
Evans argues that reducing coding friction masks two structural barriers: most workers don't see their own tasks as automatable, and enterprise adoption requires organizational decisions that span departments and timelines—not just technical capability.
A survey of 2,102 executives across 116 countries found 76% view agentic AI as a coworker rather than a tool. This duality creates operational tensions because agentic systems simultaneously substitute and complement labor, requiring both HR and asset management approaches that traditional either-or frameworks cannot accommodate.
Research shows the chatbot-to-colleague shift depends on state persistence, bounded memory, reusable procedures, and task closure—design properties of the system architecture. Larger models alone produce transcripts that disappear; colleagues accumulate experience and maintain workspace continuity across tasks.
Show all 9 sources
TheAgentCompany benchmark shows leading agents achieve 30% task completion in a simulated workplace. Social interaction, professional UI navigation, and domain-specific knowledge are the three primary failure modes, with multi-turn task performance consistently dropping to 35% across enterprise settings.
Humberd and Latham argue that as AI becomes more autonomous, monitoring mechanisms must be established first as prerequisites to incentive alignment, because defining desirable AI behaviors is harder than for humans and AI decision-making becomes increasingly incomprehensible over time.
Large-scale studies reveal agents don't align their language or ideas through interaction, but do dramatically change their actions when aware of peer presence. The difference hinges on how models process context versus update learned distributions.
Peircean semiotics reveals that symbolic goal encoding without world contact and social mediation cannot guarantee correspondence to actual values. LLMs operating in pure symbol manipulation risk divergence between stated goals and real-world outcomes.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Unraveling Human-AI Teaming: A Review and Outlook
- Putting AI on the Org Chart: Evidence on Delegation and Accountability
- When AI Becomes an Agent of the Firm: Examining the Evolution of AI in Organizations Through an Agency Theory Lens
- The Emerging Agentic Enterprise: How Leaders Must Navigate a New Age of AI
- Beyond Preferences in AI Alignment
- Conversational Alignment with Artificial Intelligence in Context
- Governing AI Agents
- Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs