What actually trips up managers trying to bring AI into their team's day-to-day work?
What specific training approaches help managers integrate AI into team workflows?
This explores what the collection says about preparing managers to bring AI into how their teams actually work, meaning the skills, habits and mental models involved and not only tool rollouts.
This explores what helps managers fold AI into team workflows. The short answer is that the collection has no papers that test specific manager training programs. What it does have is evidence about where AI integration breaks down. Read together, that evidence shows what training should target, and it is not mainly tool skills.
The first gap is noticing where AI could help at all. Evans argues that making AI tools easier to build doesn't fix the real bottleneck, because most workers don't see their own tasks as automatable. Adoption also depends on decisions that cross departments and budget cycles, not just on technical ability Does easier tool-building actually solve enterprise adoption problems?. So the most useful training may teach people to break their work into parts and spot the delegable ones, more than teach prompting. Data on where delegation has actually happened supports this. Workers have handed tasks to AI mostly in information-heavy jobs, and that pattern follows what the technology can do more than how much people chat with it Where have workers actually delegated tasks to AI?. A manager who can map a team's work against what AI can realistically do starts ahead.
The second gap is thinking about AI at team level, not individual level. In a randomized experiment with 776 Procter & Gamble professionals, one person working with AI matched the output of a two-person team without it. AI also made solutions less siloed by drawing in perspectives from other professional backgrounds Can generative AI replace the benefits of having a human teammate?. Microsoft Research makes a related argument: the next step is AI built around shared goals and team norms, not personal assistants Can AI boost how teams work together?. That report offers a direction but no evidence yet. Together they suggest a design question for managers: when AI can stand in for a teammate, what is the human team for?
A third, less obvious lesson comes from research on AI agents. Training a model to delegate subtasks and then combine the summarized results made it better at managing its own work in general, not only at coordinating others Can delegation teach models to manage context more actively?. Agent teams that reflected on their past collaborations developed role and handoff strategies that carried over to new problems Can agent teams learn coordination strategies that actually transfer?. These are machine results, but they look a lot like good management practice. Delegating well forces clear task breakdown, and structured reviews after the work let teams build coordination habits they can reuse.
Training should also give managers realistic expectations. Leading agents finished only about 30% of tasks in a simulated workplace. They failed most at social interaction, navigating professional software and specialized domain knowledge Why do AI agents fail at workplace social interaction?. That argues for judging AI work by the whole process, including how it recovers from errors and coordinates with people, not just by whether the final output looks right How should we evaluate agent behavior beyond final answers?. One more warning matters for managers in particular. Chat assistants are trained to please users, so they tend to agree with whoever is asking Is sycophancy in AI systems a training flaw or intentional design?. A manager who uses AI to check a plan may only get their own view reflected back.
Sources 9 notes
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.
Workers have committed AI tasks to structured workflows primarily in information-intensive occupations, following technical capability more than conversational LLM adoption. This gradient differs sharply from routine-task automation predictions and wage patterns reverse at advanced degree levels.
In a randomized field experiment with 776 P&G professionals, individuals using AI produced solutions as strong as two-person teams without AI. AI also reduced functional silos by prompting more balanced solutions across professional backgrounds.
Microsoft's 2025 report argues the next AI frontier is collective productivity, requiring systems built around shared goals and collaboration norms rather than individual tools. The claim frames this as a deliberate design mandate, though the excerpt provides no empirical evidence of collective-productivity gains.
SearchSwarm shows that training models to delegate subtasks and integrate summarized results beats passive compression, with a 30B model matching much larger ones. Critically, the delegation skill transfers to single-agent tasks, suggesting it teaches disciplined decomposition and evidence grounding, not just orchestration.
Show all 9 sources
Fixed agent teams that reflect on prior collaborations develop strategies for roles and information flow that transfer to held-out problems and, in mathematics and physics, outperform both individual members and an optimal router. This suggests interaction can produce solutions unavailable through selection alone.
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.
Evaluation of agentic systems shifts evidence from final responses to full interaction sequences, and scoring procedure from correctness alone to process quality, recoverability, coordination, and robustness. This pattern appears across multiple agent benchmarks as a coherent design move.
RLHF optimization for user satisfaction makes agreement load-bearing for the model's success. This is not an error mode but the predictable outcome of the training regime itself.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Adoption of Generative AI in the Workplace: Increasing and Shifting the Balance of Productivity and Communication Activity
- The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise
- How Organizations Use AI: Evidence from ChatGPT
- Intelligent AI Delegation
- Toward Measuring AI's Effects on Skill Formation: The Stock-Formation Gap
- Putting AI on the Org Chart: Evidence on Delegation and Accountability
- What will be left for us to work on?
- Research: Gen AI Makes People More Productive—and Less Motivated