As AI amplifies individual productivity, where should teams channel their efforts for collaboration?

AI has greatly boosted individual productivity, enabling a single person to build product prototypes rapidly on their own. However, critical thinking and judgment work such as defining core problems and evaluating solution value rely more on team collaboration. In the future, the focus of teams will no longer be on division of labor and implementation, but on integrating diverse perspectives to calibrate directions. Managers also need to adjust organizational and innovation processes to concentrate on high-value decision-making links.

AI tools have become powerful enough to continuously improve the feasibility of solo entrepreneurship, allowing innovators without technical backgrounds to produce usable products. With the popularization of AI, this "solo development" model not only appears in startups, but is also gradually implemented within large enterprises. Throughout the entire innovation process, the execution links have been universally accessible via AI: product managers can build runnable prototypes without writing a single line of code, while developers can use AI to accelerate software development, testing, documentation writing and data processing.

This raises a core question: does team collaboration still have lasting value? If the basic unit of work is shifting from "teams" to "individuals equipped with AI", the organizational structure, talent recruitment standards, and core value of employees will all undergo tremendous changes. This topic has long gone beyond the simple scope of discussing productivity.

We explored this issue through an innovation development challenge co-hosted with students from the Stern School of Business at New York University. We asked teams with diverse skill backgrounds to use unified AI tools and solve real, complex practical problems within one day. We aimed to study how teams use AI in the processes of creative divergence and final decision-making, as well as which links in the innovation process still cannot do without collaboration.

Eight Teams, Real Challenges, One Day of Intensive Work

33 students were required to design solutions for a series of thorny social problems in New York within 6 hours. The topics covered the affordability of groceries, public transport equity, bike lane safety, access to childcare resources and more. These problems have complex variables, and even defining the "success criteria" itself is very difficult.

Participants were randomly divided into groups of 4 to 5 people, all using the same set of tools. Everyone used AI research and creative tools to understand the problems and locate target users; then designed specific solutions through prototype tools; finally connected to intelligent code platforms such as Lovable and Vercel to develop demonstrable functional prototypes. All collaboration was completed on the Miro shared digital whiteboard, where everyone could see the progress of others in real time.

It is worth noting that most of the students had no engineering background at all (the grouping did not deliberately balance the skill ratio), the team members were not familiar with each other before, and had never used this set of tools. At the same time, the competition had an incentive mechanism, where the winners could get awards beneficial to their career development. The judging criteria clearly gave priority to problem definition and solution design, rather than technical implementation. The competition was not about who delivered the fastest, but about who could create a product that truly meets the needs.

If AI could really turn innovation into a solo task, this competition would be an ideal verification scenario: a group of outstanding people, powerful AI tools, urgent time pressure, everyone had sufficient reasons to work separately in parallel, and let AI take on the heavy implementation work.

But from the practice and feedback of each team, we saw a completely different conclusion.

Problem Definition and Solution Design

After observing the team workflow and interviewing participants, almost all groups showed two clear patterns:

At the beginning of the challenge, each team would discuss the selected topic, sort out and define the core problem, and clarify the target users. At this stage, team members would use AI (some collaboratively, some independently in parallel) to collect materials, integrate information, summarize and analyze the results before holding collective discussions to align their understanding of the problem. The scenario is very typical, just like the brainstorming in an MBA class: a group of people sitting together and communicating enthusiastically.

After completing the problem definition, most teams began to split the implementation tasks: some focused on prototype design and development, while others prepared the solution presentation and narrative. At this stage, members used AI to complete their respective assigned tasks, but they would still hold centralized discussions to verify whether the solution design is feasible.

For the work that AI can undertake or even complete independently - information integration, generating multiple design schemes, building interfaces, etc., the efficiency is indeed higher when everyone splits the tasks and advances independently in parallel.

However, problem definition is a link that all teams spend a lot of time on collaborative discussion: everyone checks each other's research conclusions, questions each other's assumptions, and jointly finalizes the core problem to be solved and the solution ideas. One participant mentioned: "Defining the problem is the most important step." Another also said: "Even with AI assistance, it is very difficult to lock in a specific problem."

The second link that highly relies on collaboration is solution design. Nowadays, the cost of rapidly generating prototypes has almost dropped to zero, and prototypes have been given new meaning. As one student said: "Prototypes are the starting point of discussion, not the final product. This is especially important in the AI era." After the team built different versions of prototypes, key questions would arise: Who are the real users of this product? Through what channels will users access the product? What if relevant data is missing? The prototype itself cannot answer these questions, but it makes these questions impossible to ignore.

The experience of the winning teams confirms one point: Problem definition completed through collaboration based on insights can spawn innovative solutions, and AI is their execution engine for building usable prototypes within a day.

· The championship team created the New York Food Price Index project: collecting price data from supermarkets across the city to evaluate cost-effectiveness, while developing a consumer-facing App and in-store display solutions. The usable App completed in 6 hours is already very impressive, but the real highlight is their redefinition of the problem: transforming the problem of price burden into an information transparency problem, which can be improved through convenient price comparison.

· The team that won the Highest Social Impact Award relied on the 13,000 existing grocery stores in New York to transform them into community wholesale distribution points. Instead of building new infrastructure, they revitalized existing stores to allow surrounding residents to purchase in bulk at low prices. The team realized that the solution already existed, and what was missing was only a better resource coordination mechanism.

· The prototype award team transformed the public complaint database into a co-built community map, which intuitively marked the hidden danger points of bike lanes. Residents can use this data to promote municipal repairs, and these data originally belong to the city, but were not open to the public.

Teams Need a Unified Collaboration Space

Most of the students participating in the challenge had no engineering background and were not familiar with the tools used, but all teams produced usable, logically consistent prototypes within 6 hours. As one participant said: "You don't need to know how to write code to make an excellent prototype!"

However, if teams directly split up and work in separate AI dialogue windows in parallel, they tend to get into trouble. Some students reported: "This model is very tricky, everyone is using independent dialogues, information overload, and a lot of time is wasted."

In contrast, some teams adopted a unified group AI dialogue and broke through this bottleneck: "AI works best when a single person enters prompts, but after everyone uses AI collaboratively in the same space, I realize that the upper limit of the team's joint use of AI is higher."

Many participants mentioned that using AI in a shared collaboration space is a completely different experience: "Usually using AI is often working independently. I really like the mode of collaboration in a shared space, the whole process can be collaborative, and you can see the progress of others at any time to provide support and ideas."

If people use AI in isolation, their cognition will be limited to their own perspective. Existing research has proved that the quality of results of teams equipped with AI is better than that of individuals using AI alone, and also better than teams without AI assistance. The observations of this challenge further show that when the team uses AI in a shared space, everyone can view the generated content in real time and give immediate feedback, AI becomes a carrier for rapid verification and creative thinking, to implement the team's collective judgment.

Collaboration and Competitive Advantage Shift Upstream

One participant pointed out the core trend: "AI amplifies the ability of a single person to produce products. The focus of team collaboration is shifting to process design, rather than the implementation of specific tasks."

This is a profound transformation. In product development over the past few decades, collaboration has mainly been used to coordinate implementation work: who develops which function, how modules are connected, and ensure that everyone's results can be integrated. This kind of coordination work is still necessary, but it is no longer a bottleneck.

The bottleneck has shifted upstream: Is the product we are building the right one? Do we understand the users well enough? Can the problem we define lead to valuable solutions?

Such problems cannot be split and processed in parallel. It requires the speculative friction generated by the collision of different viewpoints, and the team needs to slow down for full discussion before quickly moving to development.

Many participants mentioned that AI is "good at execution, but it is difficult to generate original ideas", "very useful, but still dependent on human judgment", and some concluded that "human participation in decision-making is always required". When AI takes on more of the implementation work of "how to do", humans will be more responsible for "what to do" and "why to do", and such problems naturally require collaboration.

This article does not mean that after AI simplifies implementation, collaboration is only valuable at the problem definition stage. Product development still requires professional competence, aesthetics and judgment, and teams still need to coordinate implementation. AI will not make collaboration obsolete. On the contrary, it makes the tedious, divergent but crucial work of jointly sorting out problems more critical than ever.

Managers Need to Make Three Changes

The conclusions of this challenge are applicable to team building for product development or other types of workflows, helping organizations succeed in the AI era:

1. Focus on Investing in Problem Definition

When anyone can make a runnable prototype within a few hours, the competitive advantage no longer comes from development speed, but from products that others cannot think of - rooted in your more accurate definition of the problem.

This in itself is easy to make people uncomfortable: compared with implementation, sorting out problems is slow and chaotic, especially after AI further improves implementation efficiency. It has no clear progress bar and is often repeatedly deliberated. It is easy for people to rush to start development, because at least they can intuitively feel the output. But as one participant concluded: "If you go straight to the solution without fully understanding the problem, you will only end up starting all over again."

Managers should actively reserve time for this work, and specific practices include:

· Start the project with problem definition, not directly brainstorming solutions

· Reward teams for in-depth dismantling of problems, not just rapid delivery of results

· Embrace seemingly iterative discussions, as high-quality problem definition often looks like this

· Recognize the value of such members: those who dare to pause the progress and ask "what is the problem we really need to solve" are more valuable than those who are eager to start development

2. Polish Problems with Prototypes

The development threshold has been greatly reduced, which has completely changed the positioning of prototypes. In the past, prototypes were expensive and usually placed at the later stage of the process, produced after the solution was fully finalized. Nowadays, prototypes are low-cost and can be produced quickly, so they can be made in advance - not to show the final answer and implement it directly, but to make the direction of debate tangible and perceptible.

The new value of prototypes: Intuitively present different opinions when the correction cost of divergence is still very low.

Managers should:

· Encourage rapid production of simple prototypes in the early stage, rather than waiting for the later stage to polish the perfect version

· When evaluating prototypes, focus on what key questions it raises, rather than the perfection of the finished product

· Accept that the first version of the prototype is likely to have deviations - this is exactly its meaning

· Leave space for the team to discuss the information exposed by the prototype

· Avoid the team being obsessed with the first version of the prototype, and treat the prototype as the basis for discussion

Old process: Sort out the solution → Develop and implement. New process: Initially define the problem → Quickly make a simple prototype → Further polish the problem based on the prototype.

3. Form Teams by Perspective, Not by Skill Division

However, prototypes can only become effective learning tools when people with diverse perspectives gather to interpret their information. This event confirms that AI allows people without technical backgrounds to complete development, so how meaningful is it to continue to build teams according to "who can code, who can design, who does research"? Skill division still has its role, but its weight has declined.

The best-performing teams do not necessarily have the strongest technical strength, but can understand the problem from diverse perspectives: some are familiar with the policy environment, some have personally experienced the pain point, and some can sort out the business model.

This points to a new team-building idea: no longer prioritize filling technical positions (the common configuration for product development: one engineer, one designer, one product manager), but fill in cognitive perspectives. Who can fully understand this problem? Who can spot blind spots in thinking?

One student predicted: "AI tools allow a single person to complete the work that originally required an entire company, and the team boundaries will become more and more blurred." This judgment is correct, but on the other hand, after the tools are popularized, to stand out, we must solve user pain points more deeply and directly.

If everyone can use similar AI tools, the core competitiveness of the team will no longer be technical ability, but deep understanding of the problem; and this deep cognition comes from bringing together people with different thinking perspectives.

The practice of these students designing solutions for New York's social problems in 6 hours confirms the long-standing conjecture of many innovation and collaboration researchers. The core question is not whether AI will replace teams, but how teams will transform to undertake the work that AI cannot complete: judging which problems are worth solving, who to serve - this kind of core work relying on human value judgment.

Teams that take the lead in completing the transformation will not only develop faster, but also create truly valuable products.

Keywords: #AI

J.P. Eggers, Sarah Ryan, Asha Dinesh | Text

J.P. Eggers is Professor of Entrepreneurship at the Stern School of Business, New York University. Sarah Ryan is the Director of the Entrepreneurship Acceleration Program at the Stern School of Business, New York University. Asha Dinesh is Head of Market Insights at Miro.

Zhou Qiang | Editor

This article is from the WeChat official account "Harvard Business Review" (ID: hbrchinese), written by HBR-China, and authorized for release by 36Kr.