top of page

Co-Intelligence: Why Judgement Is Becoming the Real Currency of the AI Era

Generative AI has made intelligence, in the raw sense of processing information and producing plausible output, abundant. Analysis that once took teams of analysts weeks can now be produced in minutes, at near-zero marginal cost, and access to capable models is no longer restricted by budget or specialized expertise. The open question for organizations in 2026, as well as in 2027, is what still differentiates them, and a growing body of research and practitioner experience points to the same answer: not the model, but the judgement applied around it.


That was the starting point for Elaine Barsoom, former Head of AI and Tech Innovation Partnerships at Nike and founder of Wave Co AI, at this year's Bucharest Tech Week 2026. She opened her session with a question: "What is the last decision at work that you made that only a human could have ever made?". 


The Case for Co-Intelligence 

Barsoom told the audience that strategic analysis which once took months can now be produced in twenty minutes, and that models which once cost hundreds of thousands of dollars to build are now freely accessible. Her point is not that AI is unimportant, but that once a capability is universally available, it stops being what separates one organization from another. 


That framing lines up closely with what Wharton professor Ethan Mollick termed "co-intelligence": generative AI functioning as a collaborative partner rather than an oracle, with a human still responsible for evaluating and directing its output (Mollick, 2024). Elaine Barsoom builds an organizational version of the same idea, distinguishing between two postures a company can take toward AI, one reactive, focused mainly on cost savings, and one deliberately designed with humans in mind.


The first buys short-term efficiency and long-term fragility, in her framing; the second compounds. She structures the "designed" posture around three pillars, creativity, accountability, and trust, arguing that neglecting any one of them undermines the others, regardless of how capable the underlying model is. 


Worth pausing on before reading further: which of those two postures does your own organization actually resemble right now, the one measuring AI purely by cost saved, or the one that's redesigned how decisions get made around it? 

Tacit Knowledge and Embedded Partnership 

Barsoom's clearest articulation of what gets lost in a purely reactive approach came in an exclusive interview we had with her before her session. Asked how she advises executive teams, she said the answer rarely comes from a roadmap: "Companies carry decades of accumulated context: tacit knowledge embedded in workflows, relationships, workarounds, and institutional memory that rarely lives cleanly in a data system" (written interview, Q4).


She described her own approach as "embedded partnership, not advisory at a distance," arguing that frontline employees "can see where decisions slow down, where customers get stuck, where exceptions happen, and where value disappears in handoffs" (written interview, Q4). 


This shapes how she believes AI initiatives should be framed. Rather than starting with what a given tool can do, she suggested organizations ask "where is value trapped, where is human potential underused, and what work needs to be redesigned to unlock both" (written interview, Q5). It's a distinction she also applies to partnerships more broadly: successful ones, she said, "start with a sharp business problem, not with a tool, a trend, or a press release," while the ones that fail are treated "as side projects or vendor relationships," without a clear internal owner accountable for making them work (written interview, Q3). 

When Speed Outpaces Judgement 

Barsoom's go-to illustration of what happens when this discipline is skipped is Klarna. In 2024, the fintech's AI customer-service assistant reportedly did the work of roughly 700 human agents and cut resolution times sharply; by 2025, its CEO was publicly acknowledging that the shift had over-indexed on cost at the expense of quality, and the company began rehiring humans. Barsoom's reading of the episode is that the system "can optimize for speed... but it's not [able to] optimize for the... decisions [where] human judgment was actually the competitive advantage".



A field experiment with Boston Consulting Group and Harvard and Wharton researchers, published in Organization Science, found that generative AI significantly improves performance on tasks within the model's "capability frontier," but measurably reduces performance on tasks outside it, largely because users extend their trust in AI output beyond the point it is actually warranted (Dell'Acqua et al., 2026).


The researchers call this a "jagged frontier": strong in some places, confidently wrong in others, with little visible difference between the two from the user's side. Barsoom raises a closely related point about accountability from the stage: outcomes can be right or wrong for the right or the wrong reasons, and responsibility for those outcomes was never something a model could hold on its own. 


The broader data backs up how common this gap is. McKinsey's 2025 State of AI research found that while generative AI use is now nearly universal among large organizations, only a small fraction report having scaled it into measurable enterprise-wide impact, a gap the report attributes to unredesigned workflows rather than the technology itself (McKinsey & Company, 2025). 

The Skills That Compound 

Asked what would separate top performers over the next three to five years, Barsoom pointed to two things: deep expertise, which she called a genuine competitive moat, and the adaptability and curiosity to keep working with the technology rather than deferring to it by default. Without that combination, she argued, there is no real competitive moat left, organizations start to look interchangeable. 


Her closing advice in the written interview struck a more personal note, and one she returned to repeatedly: that innovation and strategy careers are, underneath the technology, about people. "The people who take risks. The people who build conviction. The people who earn trust, change how work gets done, and bring others along before the outcome is obvious" (written interview, Q6).


Her advice to anyone building a career in this space was to "invest in relationships with the same seriousness that you invest in expertise" and to "stay close to the human reason for the work" (written interview, Q6). 

The Boundary Worth Designing 

Here's the thing nobody puts on a slide: none of this means slowing down on AI. If anything, it means getting a little more intentional about it. The tools aren't going anywhere, and honestly, they shouldn't because they're useful. What's worth protecting is the part that made your team good in the first place: the judgement calls that come from experience, from having been wrong before and learning from it, from actually knowing your customers and your craft.


So, we are curious: what's one decision in your own work you'd never want to fully hand over to an AI? 

References 

Dell'Acqua, F., McFowland, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Organization Science, 37(2), 403–423. 


McKinsey & Company. (2025). The state of AI in 2025: Agents, innovation, and transformation. McKinsey & Company, QuantumBlack. https://www.mckinsey.com 


Mollick, E. (2024). Co-intelligence: Living and working with AI. Portfolio/Penguin.

Future Summit: AI 2026 Conference - presentation by Elaine Barsoom, Founder of Waveco

 
 
 

Comments

bottom of page