In most deep tech companies, the technology is not the constraint. The team can build the thing. What it cannot reliably do is get a scientist, an investor, a regulator and an enterprise buyer to hold the same understanding of what the thing is, what it does, and why it matters, at the same time.
That failure has a name in cognitive science, and it is not a personality flaw. Once someone understands something deeply, their brain loses reliable access to what it is like not to know it. Experts systematically overestimate how much of their knowledge is shared, and underestimate how much explanatory work a listener needs1,2. In a company where the deepest expertise sits furthest from the customer, this is not an edge case. It is the default operating condition.
The operating system gap is not a failure to communicate. It is a failure to model what another mind does not yet know.
Why expertise creates the gap it needs to close
The curse of knowledge is well documented in economic and organisational settings: people with private information cannot fully discount it when predicting how a less-informed party will judge, price, or understand the same situation1. Later work extended this to expertise directly, showing that experts reliably overestimate a novice's ability to follow their reasoning, even when explicitly asked to imagine the novice's perspective2. Debiasing instructions barely move the effect. It is not a lapse in effort. It is a structural limit on perspective-taking under deep domain knowledge.
Bridging that gap is the job of the brain's mentalising network, the medial prefrontal and temporoparietal circuits that construct a model of what another person knows, wants and expects3. Building an accurate model of an investor's or a buyer's mental state is metabolically and attentionally expensive, and it degrades under exactly the conditions deep tech runs on: technical pressure, time scarcity, and a scientist's attention already consumed by the problem itself.
Private knowledge
The team knows the mechanism, the failure modes, the caveats. That knowledge cannot be fully discounted when predicting what a buyer or investor will understand.
Mentalising load
Modelling another person's actual state of understanding draws on limited cognitive resources, and is the first thing to degrade under deadline pressure.
Silent miscalibration
Nobody flags the gap in the room. The explainer feels clear, the listener nods, and the misunderstanding surfaces weeks later as a stalled deal.
None of this means the science needs to be simplified. It means the translation work is a distinct skill from the technical work, and most deep tech teams never staff or train for it.
Three points where the operating system fails
1. Cognitive overload disguised as rigour
Working memory has a hard capacity limit, and instruction that ignores it does not transfer, however accurate it is4. A forty-slide technical deck delivered to a buying committee is not thorough. It is unlearnable in a single sitting. The audience leaves with fragments, not a model, and fragments do not survive an internal approval process.
2. No shared mental model across functions
Team performance under complexity and time pressure depends on members holding compatible mental models of the task, the roles, and the equipment5,6. When R&D and commercial teams have never built a common model of what "ready for revenue" means, they are not disagreeing about strategy. They are running on different maps of the same territory, and every handoff between them loses information silently.
3. No audience design in the language itself
Effective communicators continuously monitor whether their listener is actually tracking, and adjust in real time7. Deep tech teams rarely build this monitoring into their process. The same explanation goes to the regulator, the investor and the enterprise buyer, calibrated for none of them, and each stakeholder is left to reconstruct relevance on their own.
Building the operating system on purpose
If the gap is cognitive, the fix is structural, not motivational. Four mechanisms do most of the work.
A modular narrative, not a single deck
Build the story once, in components, then recombine it for each audience's actual prior knowledge instead of reusing the technical version everywhere.
Named translators between R&D and commercial
Someone whose explicit job is to hold both mental models and check them against each other, before the gap reaches a customer or a term sheet.
Chunked, sequenced explanation
Segment complex material into pieces that respect working memory limits, with the mechanism last, not first, for any audience that is not technical.
A shared, written definition of commercial readiness
One document both R&D and commercial sign off on, so decisions stop being renegotiated from two incompatible models of the same milestone.
This is also why a strong team often outperforms a group of strong individuals. High-functioning teams develop a transactive memory system, a shared understanding of who knows what, so no single person has to hold and translate the entire picture alone8. Deep tech companies that build this deliberately convert their most exposed vulnerability, expert isolation, into distributed, checked understanding.
The bottom line
Deep tech companies rarely fail because the science does not work. They fail because the people who understand the science cannot reliably model what everyone else does not yet know, and no structure exists to close that gap before it costs a deal, a round, or a regulatory approval.
The technology is the foundation. The operating system, built to translate expertise into shared understanding, is what makes it move.
References
- 1.Camerer, C., Loewenstein, G., & Weber, M. (1989). The curse of knowledge in economic settings: An experimental analysis. Journal of Political Economy, 97(5), 1232–1254. Link
- 2.Hinds, P. J. (1999). The curse of expertise: The effects of expertise and debiasing methods on prediction of novice performance. Journal of Experimental Psychology: Applied, 5(2), 205–221. Link
- 3.Frith, C. D., & Frith, U. (2006). The neural basis of mentalizing. Neuron, 50(4), 531–534. Link
- 4.Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. Link
- 5.Cannon-Bowers, J. A., & Salas, E. (2001). Reflections on shared cognition. Journal of Organizational Behavior, 22(2), 195–202. Link
- 6.Mathieu, J. E., Heffner, T. S., Goodwin, G. F., Salas, E., & Cannon-Bowers, J. A. (2000). The influence of shared mental models on team process and performance. Journal of Applied Psychology, 85(2), 273–283. Link
- 7.Clark, H. H., & Krych, M. A. (2004). Speaking while monitoring addressees for understanding. Journal of Memory and Language, 50(1), 62–81. Link
- 8.Wegner, D. M. (1987). Transactive memory: A contemporary analysis of the group mind. In B. Mullen & G. R. Goethals (Eds.), Theories of Group Behavior (pp. 185–208). Springer. Link