As 2025 draws to a close, the annual ritual of tech predictions begins anew. Dr Lewis Z. Liu, co-founder and CEO of Eigen Technologies, offers a distinct perspective shaped not by abstract forecasting, but by the hard realities of building a venture from the ground up. After a year of founding a company, investing in early-stage firms, and consulting with over 300 industry executives and policymakers, Liu presents 12 foundational assumptions for AI in 2026 that he is actively betting his business on.
The Core Shifts: Privacy, Context, and Human Value
Liu argues that the coming year will see a necessary merging of two critical concepts: context and privacy. He posits that effective, 'live AI' integrated into daily workflows cannot exist without deep contextual understanding, which in turn cannot be shared without granular, robust privacy controls. This integrated approach, currently underdeveloped as the industry focuses on large language models and simple applications, will become a hot topic.
Furthermore, the backlash against generic AI-generated content will intensify. "We're all sick of GPT-laden lazy AI slop," Liu states, predicting that genuinely human-originated creativity will be valued as a foil to the sea of beige, automated text. The compromise may lie in personalised AI tools that amplify unique human voices rather than replace them.
Market Realities and Economic Corrections
The hype around certain AI tools is due for a reality check. Liu highlights the 76 per cent drop in activity last quarter for AI coding platforms like Lovable and Cursor, a phenomenon dubbed 'vibe-coding'. The issue, encapsulated in a popular meme, is that such tools allow small teams to generate the technical debt of a much larger one, creating code that humans struggle to debug. However, Liu believes these tools are powerful and will mature with improved guardrails.
Economically, 2026 will see a shift from 'vibe revenue' to substance. After years of start-up growth fuelled by repackaged services with poor margins, real recurring revenue and solid unit economics will regain importance. This maturation of AI business practices will separate sustainable ventures from fleeting trends.
A significant market correction is also anticipated, driven by two factors: the rise of cheaper, open-source models like DeepSeek, and potential over-investment in data centres. Liu notes that DeepSeek 3.2 is benchmarked at 96 per cent cheaper than offerings from OpenAI or Google, questioning the need for top-tier GPUs for most applications. This, combined with trillion-dollar data centre build-outs, could reshape the industry's economic foundations.
The Geopolitical and Industrial Landscape
Liu's assumptions paint a picture of a fragmenting global AI landscape. He predicts a rise in populist, anti-elite skepticism towards Silicon Valley, fuelled by rhetoric about human replacement and growing inequality. This resentment will spread from blue-collar to white-collar professionals, pressuring builders to create AI that works for broader humanity.
In response to China's lead in physical AI applications like automated 'dark factories', the West will dramatically increase investment. Liu cites Jeff Bezos's launch of Build AI with $6.2 billion for manufacturing as a key attempt to close this gap. Concurrently, Chinese AI models will continue gaining traction among Western builders, with venture firm Andreessen Horowitz estimating 80 per cent of Silicon Valley start-ups already use them.
Perhaps most intriguingly, Liu positions Europe as a potential third pole in AI. This will not be based on regulatory prowess or breakneck acceleration, but on an old-fashioned focus on trust, human-centricity, and a thoughtful perspective on technology's role in society. This approach will offer a counterbalance to US accelerationism and China's application-driven pragmatism.
Finally, traditional industries like legal, finance, and energy will begin to outcompete peers through thoughtful AI adoption. Law firms are now avid users for knowledge management, and energy companies are exploring AI for site construction, suggesting the greatest gains are for sectors with significant room for digital leapfrogging.
For Dr Lewis Z. Liu, the overarching theme for 2026 is a necessary re-centring. "AI should not be 'done to humans'; it must be developed for humans," he concludes. His twelve assumptions are a bet on a future where the most valuable AI applications make us more human, not less. The alternative, he warns, is not just bad business—it's inhuman.
