The Collaboration Gap: AI Has a Teamwork Problem
AI made individuals faster, but teams are still stuck with tools built for a pre-AI world
AI can help one person turn an idea into a draft, prototype, or presentation faster than ever before. The challenge starts the moment they try to share it. Most AI tools are built for individual use—but work isn’t. This mismatch is introducing a new kind of friction for teams trying to collaborate in an AI-first world.
To explore how AI is changing the way teams work, we spoke with US-based knowledge workers who regularly use AI and work primarily with others. Participants included enterprise employees, consultants, and small business owners. Specifically, we wanted to understand how they’re adapting their workflows and where their current tools fall short.
The collaboration gap showed up in several ways. One product manager described building an HTML presentation with Claude Code, then struggling to find a way for a coworker to contribute. An operations professional said their team turned to GitHub as a workaround to share AI-generated artifacts and build on them together.
Almost all of the participants said AI helped them accomplish more individually but didn’t meaningfully improve how they worked together. In fact, some said it made collaboration harder: work moved out of shared documents, teams started building before aligning, and colleagues had more to review with less visibility into how it was created.
For these participants, AI worked well in a one-to-one exchange. The gap emerged when others needed to contribute.
Sharing the context matters as much as the output
AI-assisted work often starts with one person. They give the tool instructions, add sources, correct its output, and make decisions as they go. By the time they share a draft or prototype, they have accumulated context that may remain inside their AI tool.
To overcome this, participants described increasingly sharing that context alongside the output, including prompts, skills, instructions, and the original source materials. For them, that context matters because a finished draft does not show a colleague everything they need to continue the work. They may need to understand the assumptions behind it, reproduce the process, or use the same instructions to make changes.
Participants used tools such as Slack, Drive, and GitHub to share this information, but setting up and maintaining those systems introduced friction. Their workarounds point to a broader shift: The context used to create an artifact is becoming part of what teams need to share in addition to the output itself.
Faster creation can create more work for collaborators
Participants described several new challenges that have emerged as AI became part of their shared workflows:
- Verification tax: Participants spent extra time checking colleagues’ AI-assisted work because they couldn’t always see its sources, how AI was used, or how thoroughly it had been reviewed.
- Attention fatigue: Participants described AI producing drafts, prototypes, and possible directions faster than their teams could absorb them, making it hard to give each artifact the close attention it needs.
- Unclear contributions: Participants wanted greater visibility into what each person contributed and what the AI system generated or changed. They mentioned the need for a surface that would let them see exactly how agents and humans were collaborating and who was doing what.
Together, these challenges suggest that faster individual output does not automatically translate into faster team progress.
What teams need to collaborate effectively with AI
The interviews suggest three needs for teams working with AI:
- Context that stays with the work: Teams need prompts, skills, sources, feedback, and decisions to stay with the work they shaped. They also need a record of what people decided, why, and what happened next. That history helps teammates continue the work without having to reconstruct it from old messages and meetings.
- Contributions people can trace: Teams need to see what a person contributed, what AI generated, and what someone reviewed or approved. That record helps colleagues decide what they can trust, where they need to look more closely, and who made the final decision.
- Access across roles: Shared AI work also needs an environment where technical and nontechnical colleagues can understand the work, contribute their expertise, and continue it without having to learn a developer workflow or default to GitHub.
Ultimately, this research suggests that collaboration with AI depends on what happens after the output is created. Can a colleague understand the context, assess the contributions, and move the work forward?
That’s the opportunity we’re exploring with Superhuman Docs: helping teams carry the context and clarity behind AI-assisted work into the next era of collaboration.
