The real revolution of generative AI may lie not in what it can do alone, but in how it will reshape human collaboration. Beyond headlines focusing on generative AI’s potential to replace jobs on the one hand, and the flood of frivolous “AI slop” clogging the consumer internet on the other, there is a quieter but more consequential opportunity: integrating AI into human team workflows in ways that increase collective intelligence and societal value creation.
With our colleagues at the Brookings Center for Sustainable Development (CSD), we have been exploring how generative AI tools can be integrated into collaborative research and insight generation for addressing the world’s toughest challenges—from extreme poverty to persistent inequalities and environmental degradation. As we have learned in the context of the 17 Rooms initiative, these tend to be problems that no single actor can solve alone. Progress often depends on small, time-bound teams coming together—across sectors, disciplines, and geographies—to share knowledge, align priorities, and chart action.
For such teams, the core opportunity of generative AI’s generalized reasoning and natural language capabilities is not merely in accelerating individual outputs, such as cleaner prose or faster coding, but in helping teams collaborate, think together, and execute strategies for systemic change. This raises a critical question of how to embed generative AI technologies within the collaborative fabric of teams themselves to enhance—rather than erode—human-to-human collaboration.
Introducing “vibe teaming”
To explore this question, we have been developing an approach (with active input from our CSD colleagues) that we call “vibe teaming”—a method of integrating generative AI into collaborative processes from the outset, not merely as a tool for individual productivity but as an active support to team-based problem-solving. We describe the observational findings in a new working paper, with the aim of eliciting feedback and prompting broader experimentation to test and validate the approach over time.
Vibe teaming is a play on the recent viral term “vibe coding,” coined by prominent software engineer Andrej Karpathy. Vibe coding describes a new paradigm of generative AI-enabled software development where developers outline the vibe of their desired outcomes in natural language, leaving the AI model to code the first draft. Freed from syntax-heavy tasks, software developers can iterate rapidly and emphasize strategy over implementation details.
As conversations about vibe coding soon became conversations about vibe working more generally, we began experimenting with vibe teaming. We envisioned a paradigm in which AI becomes a participant in a team’s upstream ideation and problem-solving—not just a post-hoc tool for synthesis or formatting.
By taking care of the more menial individual work of manually transcribing conversations, generating drafts, and iterative individual editing within word processing software, AI tools have allowed us to devote more time to interactive teamwork. The effect has been as much about greater efficiency as it has been about greater emphasis on collaborative synthesis and ideation that pushes the ‘edge’ of our team’s thinking and performance.
Summary of approach
To evaluate the practical value of vibe teaming for real-world outputs, we selected a challenge that was both ambitious and urgent: the eradication of extreme poverty. We organized a virtual session with Homi Kharas, senior fellow at Brookings and a leading authority on global poverty, to collaboratively develop a high-level strategy for achieving Sustainable Development Goal (SDG) 1.1—ending extreme poverty globally by 2030.
The meeting involved what we have identified—at least for now—as four consistent steps of vibe teaming (Table 1):
|
Step |
Human-AI configuration |
Details |
|---|---|---|
|
1. Structured team conversation, transcribed by AI |
Team (Homi, Jacob, Kershlin)+AI |
A semi-structured team discussion with the domain expert (Homi in this instance) focused on problem diagnosis, constraint identification, and framing of strategic levers. The discussion was recorded and transcribed using AI tools, enabling a real-time capture of insights (30 minutes). |
|
2. First draft via AI |
Individual (Kershlin)+AI |
A custom language model—primed with both the transcript and a five-part strategic framework—generated an initial draft strategy reflecting the conversation’s core themes (5 minutes). |
|
3. Human-AI drafting |
Team (Jacob, Kershlin)+AI |
We engaged in rapid iteration with the AI model, probing the draft for feasibility, political nuance, operational logic, and communication strategy. This was a collaborative thinking process, where AI helped us test and stretch emerging insights (10 minutes). |
|
4. Structured team review, transcribed by AI |
Team (Homi, Jacob, Kershlin)+AI |
A second discussion with the domain expert (Homi) provided both validation and further ideation. The transcript of this exchange informed revisions to the strategy and supported the development of a draft Brookings-style commentary (15 minutes). |
Following these four steps, we spent an additional 30 minutes using the team review transcript to refine the strategy and draft a 1000-word Brookings-style commentary based on that strategy (forthcoming). The entire vibe teaming demonstration totalled roughly 90 minutes.
Despite this highly compressed timeframe, vibe teaming produced outputs of notable quality. Similar experiments conducted with other Brookings scholars—on topics ranging from gender equality to state fragility and community-led development—have yielded similarly promising results. These outputs demonstrate the potential of a rapid “human-human-AI” workflow for accelerating and enhancing knowledge work. Importantly, with vibe teaming we spend more time collaborating—brainstorming and discussing—and less time on individual tasks like transcription and drafting, compared to our conventional workflows.
Lessons learned
The primary breakthrough was not simply the speed or polish of the outputs, but also the shift in how the team worked. Unlike conventional prompting-based AI workflows, which often begin with sparse or decontextualized inputs, vibe teaming starts with a rich human-to-human dialogue as the first input to an AI model. This move appears to give the AI model a deeper foundation to build upon. Throughout the process, the AI played a catalytic role in accelerating synthesis and surfacing patterns—but the core insights emerged through collaboration: live triangulation and convergence of domain expertise and iterative reflection among the team.
While the method’s generalizability remains untested, three key lessons from vibe teaming are emerging:
- Start with rich human context: Transcripts of live team conversations of at least two team members are more effective than abstract, individually generated or templated prompts.
- Customize and coach: Tailoring AI models and prompts to the specific domain or problem can mitigate generic or sycophantic responses and increase the specificity of insights.
- Human review remains essential: Expert review of outputs is critical to address errors and refine strategic framing beyond what AI can produce alone.
Managing the risks
As with any innovation in workflow or technology, vibe teaming introduces new risks that must be deliberately managed:
- Data security and privacy: Large volumes of transcribed interactions require strong data management and privacy measures—often beyond what teams are used to.
- AIs as people pleasers: Because models converge rapidly on conventional answers, there is a need for mechanisms that encourage divergent thinking and human expertise and authority as oversight.
- Cognitive atrophy: Heavy dependence on AI for drafting and editing can undermine a team’s ability to organize arguments and refine their writing—particularly among junior staff or within low- or medium-performing teams.
Well-designed team interventions could help mitigate risks. Bias audits, data stewardship roles, and team rituals of interrogation and dissent can all help maintain rigor. And while we recognize the common view that writing is thinking, vibe teaming invites new forms of human-AI collaborative authorship—a paradigm shift that warrants its own kind of mastery.
Conclusion
As different types of organizations adapt to an AI-driven era, the challenge will extend beyond deploying new tools to reimagining teamwork itself. Vibe teaming offers a glimpse of what a future of thoughtfully designed human-AI collaboration could look like: AI embedded from the outset—not to replace human insight, but to unlock its potential.
Related Content
The Brookings Institution is committed to quality, independence, and impact.
We are supported by a diverse array of funders. In line with our values and policies, each Brookings publication represents the sole views of its author(s).
Commentary
Introducing vibe teaming: How AI can enhance collaborative problem-solving
June 11, 2025