If the last decade was about democratizing innovation—putting lean sprints, low code‑ tools, and cloud into more hands—the next decade is about automating it. This was the overarching vision that drove us to create Extuitive.
At Extuitive, we believe product creation and reinvention are on the verge of running like a living, evolving system: generating many viable options, applying selective pressure, letting the fittest survive, and carrying the learning forward. That’s not science fiction; it’s the practical shape of what’s already working when modern AI, human judgment, and nature’s own playbook meet.
Polyintelligence: Many Minds, One Loop
We call this approach polyintelligence—a deliberate union of three kinds of intelligence:
- Machine intelligence for relentless search, simulation, and generation
- Human intelligence for initial user intent data, intuition, and actual behaviors.
- Nature’s intelligence, the logic of emergence to create optionality and apply selective pressure.
Flagship Pioneering has recently articulated the overall vision of polyintelligence1 - the push to unify the organic, the human, and the technological into a connected intelligence—especially in biotechnology.
However, polyintelligence is not just limited to biotech. In consumer innovation, this same synthesis now gives small teams unimaginable capabilities previously only accessible to large companies with large budgets. And it begins with emergence.
Emergence: Why the Best Ideas “Surface” When You Widen the Search
Emergence isn’t mysticism; it’s mechanics. When you deliberately create variation and systematically select against clear goals, better ideas surface. Evolutionary design and newer quality and diversity methods embody this: instead of chasing one “optimal” concept, they illuminate the whole landscape, thereby populating a grid of high-quality alternatives to reveal alternative outcomes with a different claim, form factor, or use case. Designers use these methods to overcome fixation and discover options they wouldn’t have drawn themselves2.
This is how Extuitive’s custom trained AI agents work. They expand the option set—formulations, bundles, packaging, copy, visuals—and preserve the promising variants under constraints. Variants could be product appeal, purchase intent, uniqueness, or other factors. Then they hand those variants to the next layer of intelligence: consumers.
Consumer Truth at Software Speed
Imagine if you could reliably test a product or advertising concept with an audience in real time, without having to recruit participants - at a fraction of the cost and time compared to traditional approaches.
Recent research shows generative AI agents can sustain believable, memory-based behaviors and even exhibit emergent group dynamics—useful properties when you need to “rehearse” how people will respond to a product or story. In parallel, social scientists have shown large language models can be conditioned as proxies for specific sub‑populations, reproducing response distributions (“algorithmic fidelity”) when properly grounded and calibrated3. Together, those findings open a path for synthetic consumer panels that complement small, fast human checks.
Extuitive’s AI consumer agents stress test concepts and creative content across large, diverse synthetic panels. You see where purchase intent spikes, which claims trigger skepticism, and which segments are price sensitive—before you spend time and money on tooling or media. At Extuitive.com, we showcase how Shopify merchants generate and validate ads in minutes and preview lift across 100,000+ user personas all based on real user data collected by us.
It’s worth noting that with recent advancements in LLMs, groups of LLM agents can now spontaneously form shared conventions—and even develop collective biases—from simple local interactions. That’s a powerful reminder: multi‑agent systems can produce emergent social behavior. We should harness this capability carefully and audit continuously4. We have built that caution into our loop with regular human‑sample calibration and bias checks.
Once a product and its story are chosen, our agents generate the creative content—images, posts, reels, headlines—then run adaptive experiments that push spend toward winners as the data arrives. In settings with limited budget, fast fatigue, and many variants, multi-armed bandit methods (e.g., Thompson sampling5) consistently beat static A/B by learning while earning. That means less “learning tax” and more revenue sooner.
Inevitability Meets Opportunity: Advantage, Entrepreneurs
- The cost of knowing is collapsing. Enterprise surveys show the fastest gen AI adoption in exactly the functions we automate—marketing/sales and product/service development—with organizations now using gen‑AI in multiple functions rather than isolated pilots. As adoption compounds, automation of the full loop goes from plausible to standard6.
- Polyintelligence compounds advantage. Human AI collaboration doesn’t always outperform AI alone on every metric, but met analyses and position papers converge on the same design lesson: teams need to orchestrate complementary strengths and evaluate as centaurs (human+AI), not in isolation. That’s exactly what our loop operationalizes7.
- Nature’s algorithm scales down beautifully. Optionality + selective pressure is how scrappy teams win: you don’t need more meetings; you need more good shots on goal and a rigorous way to prune. Evolutionary and quality diversity methods give businesses a principled way to do both.
An Invitation to Be Part of Our Journey
Automating consumer innovation isn’t about sidelining people; it’s about amplifying them—using machines to open the search and nature’s rules to focus it, while humans set the direction and guardrails.
That polyintelligent loop is how small and medium businesses can punch above their weight: more viable ideas, fewer costly bets, faster launches, and stories that actually convert. This isn’t just plausible—it’s the new baseline taking shape in the data and in the market.
[1] https://polyintelligence.com/
[2] https://link.springer.com/article/10.1007/s10710-023-09477-9
[3] https://arxiv.org/abs/2304.03442
[4] https://www.science.org/doi/pdf/10.1126/sciadv.adu9368
[5] https://web.stanford.edu/~bvr/pubs/TS_Tutorial.pdf
[6] https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2024
[7] https://mitsloan.mit.edu/press/humans-and-ai-do-they-work-better-together-or-alone
