Confluence for 3.23.25
The cybernetic teammate. Looking beyond efficiency. Claude gets internet access. A simple fix for Deep Research documents.
Welcome to Confluence. Here’s what has our attention this week at the intersection of generative AI and corporate communication:
The Cybernetic Teammate
Looking Beyond Efficiency
Claude Gets Internet Access
A Simple Fix for Deep Research Documents
The Cybernetic Teammate
A new paper hints at the future of teamwork in an AI-powered workplace
A new working paper titled “The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise” from Harvard Business School, Wharton, and Procter & Gamble offers a compelling study of how generative AI may fundamentally alter workplace collaboration1. Co-authored by Wharton professor Ethan Mollick, whose work and insights on AI we’ve frequently referenced, the study examined how AI transforms the core pillars of collaboration — performance, expertise sharing, and social engagement. The study involved 776 professionals at Procter & Gamble who tackled real product innovation tasks either individually or in teams, with or without access to generative AI (they used technology based on GPT-4). Through the study, the team set out to answer a straightforward question: can AI function not just as a tool, but as a true teammate?
The results were striking. Teams working without AI outperformed individuals working alone, confirming the traditional value of collaboration. When individuals were equipped with AI, however, they delivered work comparable in quality to what human-only teams produced, and did so in less time. In fact, the time it took to complete tasks was significantly reduced — by roughly 16% — while the length and depth of submissions increased. Let that sink in: AI elevated the output of individual contributors to that of groups working in teams. These findings suggest that AI doesn’t just make individuals more productive, but rather, it effectively replicates some of the key benefits that previously required human collaboration. Teams in the study that used AI achieved the best results overall, though the improvement over AI-equipped individuals wasn’t as dramatic as might be expected.
Improvements extended to expertise boundaries and emotional experience. Without using AI, R&D and Commercial professionals predictably favored solutions aligned with their professional backgrounds. But with AI, those distinctions faded—both groups produced more balanced ideas, effectively overcoming siloed thinking. This effect was particularly significant for less experienced employees, who performed much better with AI assistance. In essence, when assisted by AI, boundaries created by expertise dissolved.
The researchers also found that AI offered emotional benefits, with users reporting higher levels of enthusiasm and lower levels of frustration. Surprisingly, solo workers using AI felt just as energized and engaged as those working in teams. This flips a longstanding assumption on its head: rather than detracting from the social aspect of collaborative work, generative AI might actually reinforce it in new ways.
We think this is an important paper.
For some time we’ve known that generative AI is no longer just a writing tool or code generator, but has become a full-fledged collaborator, capable of enhancing performance, bridging gaps in expertise, and even providing the kind of emotional reinforcement that fuels motivation. With current models, it has become a profound skill enhancer and leveler. Professionals who can engage with AI effectively — asking the right questions based on their experience, refining outputs leveraging their expertise, and applying their distinctly human judgment — will be better at what they do, plain and simple. For those who wish to take it, it’s a profound increase in ability for a marginal expense of time, energy, and money.
Looking Beyond Efficiency
How we can use generative AI to do more, not just more of the same.
In certain circles, the idea of abundance is the hot discussion topic of the week. And while the literature and ideas driving this conversation are far beyond the scope of Confluence, the simple idea of doing and delivering more is one that we believe is critical for communication professionals to consider in the context of generative AI.
Our conversations with clients often turn to questions of what generative AI can do to make our work easier, faster, or more efficient. In these conversations, the focus is on how generative AI can help us do the work we do today. It’s about integrating generative AI into existing processes and workflows to deliver high-quality work in less time and often, less effort. These conversations are absolutely necessary, and our intention is not to undersell their importance or their value. The decisions and outcomes they produce have real, tangible effects on the work and lives of those who learn how to better integrate generative AI into their day-to-day.
But there is another conversation that we’re having amongst ourselves that we believe all communication professionals need to seriously consider. It’s a conversation not about how to use generative AI to do the work we do today, but instead what can we do with generative AI that we never would have been able to do before? In other words, how can we combine our expertise in communication and human interaction with the capabilities of generative AI to enter into entirely new types of work?
We don’t have the answers to this question, but we believe looking beyond efficiency gains with generative AI opens valuable opportunities for communication professionals to drive organizational outcomes. While many companies’ tools lag behind cutting-edge capabilities, forward-thinking teams should consider how this technology will influence organizational communication holistically and how they can best shape that dynamic. The most successful teams will be those who use the extra capacity that generative AI can create to devise entirely new ways of driving effective communication within their organizations.
Claude Gets Internet Access
Claude’s inability to search the web had previously been one of its biggest limitations.
We’ve noted before that Anthropic’s Claude — specifically, Claude Sonnet 3.7 — is our go-to model. This has been the case for just over a year, since the release of the Claude 3 family of models. One of Claude’s biggest limitations, however, has been its inability to access the internet and search the web. That changed this week, as Anthropic announced on Thursday that Claude can now search the web.
In our limited experimentation so far, we’ve found Claude’s web search functionality useful if not quite as sophisticated as ChatGPT’s or Perplexity’s. In the side-by-side example below, ChatGPT’s initial response is significantly better:
While this initial response from Claude was not what we were looking for, we were able to get the information we wanted with a single follow-up prompt. Again: useful, if not necessarily state of the art. Claude was already our go-to model for the majority of our day-to-day tasks and continues to be. The web search functionality seems good enough, for most cases, to eliminate one more reason to need to switch between tools. If we need to perform a particularly complex web search, we’ll likely continue to use ChatGPT or Perplexity (which is purpose-built for search). But on the whole, the ability for Claude to search the web is a helpful addition that further cements its position as our tool of choice.
A Simple Fix for Deep Research Documents
A helpful website makes OpenAI’s powerful research tool play better with your workflow.
Those of us using OpenAI’s Deep Research know the power of having an AI research analyst that can synthesize hundreds of sources into comprehensive reports (and we have an example of that work here). We also know the frustration of trying to share those reports in other formats — the hyperlinked citations, headings, subheadings, and other formatting that make the original so easy to read and use online render poorly when you try to copy and paste them elsewhere.
Enter DeepResearchDocs.com, a simple utility that solves this formatting problem. Paste in your Deep Research output, and it reformats the document for clean export to Word or other text-based formats, or as a PDF for download. For those who’ve found themselves caught between Deep Research’s impressive capabilities and its export limitations, this tool removes a small but meaningful friction point in an otherwise transformative workflow.
Here’s a quick video of the site at work:
We’ve been using this tool for a few weeks now, and while it’s a modest innovation, it’s been a Godsend as we try to do more with Deep Research’s impressive output. The site is free to use, but has a donation link that’s worth seeing, too.
We’ll leave you with something cool: A doctor from the University of Pennsylvania is saving lives by using AI to find new uses for old drugs.
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.ChatGPT 4.5’s summary of the paper:
Summary of "The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise"
Overview Dell’Acqua et al. (2025) conducted a comprehensive field experiment at Procter & Gamble (P&G) involving 776 professionals, examining the impact of generative AI (GenAI) on team collaboration, knowledge sharing, and emotional engagement. Using P&G’s real-world new product development challenges, the study explored how GenAI alters traditional collaboration dynamics within cross-functional teams comprising Commercial and R&D professionals.
Methodology Participants were randomly assigned into one of four experimental conditions using a 2x2 factorial design: (1) individuals working alone without AI, (2) individuals with AI, (3) teams without AI, and (4) teams with AI. Each team included representatives from R&D and Commercial departments, simulating realistic organizational structures. Tasks involved developing innovative solutions addressing genuine business challenges at P&G, ensuring practical relevance. Performance metrics included solution quality, novelty, feasibility, technical orientation, productivity (time spent), and emotional experiences.
Key Findings
1. Performance Enhancement:
The use of AI significantly enhanced performance, with individuals assisted by AI producing results comparable to traditional two-person teams without AI. Teams aided by AI achieved the highest quality outcomes, showcasing a notable improvement in both average performance and the likelihood of producing exceptional, top-tier ideas.
2. Democratization of Expertise:
GenAI facilitated the breakdown of traditional functional silos, enabling participants to propose balanced, interdisciplinary solutions regardless of their original expertise (commercial vs. technical). Individuals less experienced in product development significantly benefited from AI assistance, performing at levels comparable to more experienced colleagues, thus democratizing knowledge and reducing expertise-based disparities within teams.
3. Emotional and Social Impact:
Contrary to concerns about AI's negative impact on workplace social dynamics, the study found that AI integration fostered significantly more positive emotional experiences—such as increased excitement and reduced anxiety. This implies that GenAI can partly replicate the emotional and motivational roles typically provided by human teammates.
Critique and Implications While the study provides compelling evidence for GenAI's positive effects on team productivity, knowledge sharing, and emotional engagement, it acknowledges several limitations. The relatively short duration of the experimental setup (a one-day virtual workshop) may not fully capture long-term complexities, iterative coordination challenges, or deeper relational dynamics typical of ongoing organizational interactions. Additionally, the study’s participants were relatively inexperienced with AI tools, suggesting that the observed benefits could either represent an initial novelty effect or an underestimate, given potential skill development over time.
The study importantly highlights AI’s transformative potential not merely as a productivity tool but as an active collaborative participant—a “cybernetic teammate.” It challenges traditional views on organizational design, suggesting that companies may need to reconsider optimal team sizes, composition, and role definitions as GenAI integration becomes commonplace. Further research is necessary to explore long-term impacts, specifically how continued reliance on AI might affect human skill development, professional identities, and organizational culture.
Conclusion Dell’Acqua et al.’s experiment offers robust initial evidence that generative AI can fundamentally reshape teamwork and expertise boundaries, enhancing collaborative performance and emotional engagement. As GenAI becomes increasingly integrated into organizational processes, its potential to democratize expertise, enhance innovation quality, and positively influence workplace culture merits thoughtful exploration and strategic consideration.
(Dell’Acqua et al., 2025)