Politics

Practical foresight: Liu’s 12 AI assumptions for 2026 focus on governance, economics and societal trust

Thursday, 18 December 2025 6:07AM UTC

Dr Lewis Z. Liu’s pragmatic 2026 AI assumptions highlight the importance of privacy, responsible development, and economic sustainability amid rapid technological and geopolitical shifts, offering a strategic blueprint for industry and policymakers.

‘Tis the season for 2026 predictions, a ritual many dread, but Dr Lewis Z. Liu’s piece laying out “12 AI assumptions for 2026” deserves attention because it is rooted less in idle prophecy and more in a year of building a company, advising policymakers and conversations with more than 300 executives. His framing shifts the debate from sensationalist forecasts to practical assumptions that founders and managers can organise around. ^[1]

Central to Liu’s thesis is the convergence of context and privacy: the argument that truly useful AI requires rich contextual signals while those signals can only be shared if granular privacy controls are in place. That proposition helps explain why organisational proprietary data , long siloed and underused , could become a genuine competitive moat if paired with rigorous governance. Industry observers note the same dynamic: unlocking unstructured enterprise data will be pivotal for automation and insight, provided privacy and control scale with access. ^[1]

The rise, wobble and likely maturation of AI coding agents is another throughline. Liu warns of a trough of disillusionment after early exuberance, citing precipitous drops in activity among some platforms and the familiar pitfall that AI-generated code can accumulate unmanageable technical debt , “Vibe coding allows two engineers to develop the technical debt of 50 engineers.” Yet the market is consolidating around agentic coding tools from major players. Microsoft has moved to embed multiple coding agents into GitHub, signalling a more neutral, plural approach, while startups and incumbents alike are improving guardrails and agentic capabilities to reduce hallucinations and boost reliability. This evolution supports Liu’s view that coding agents will not disappear but will require better integration and human oversight to reach their promise. ^[1]^[3]^[6]

That agentic evolution is not confined to software development. Industry reporting shows a rapid push by financial institutions into agentic AIs that can plan, learn and act autonomously. British banks are already piloting customer-facing agents for budgeting and investing in collaboration with regulators, a development that exposes new systemic and governance risks even as it promises scaled personalisation. Regulators are flagging threats that flow from speed, autonomy and interaction effects between agents, and expect accountability to be enforced under existing regimes. These tensions underscore Liu’s point that organisational adoption will reward traditional industries that pair thoughtful governance with practical AI use-cases. ^[2]^[6]

Anthropic’s recent model upgrades and rival entrants sharpen that competitive and technical backdrop. The company’s Opus 4.5 release extended Claude’s reasoning, memory and agentic capacities, enabling more complex code generation and autonomous tasking , exactly the capabilities Liu anticipates will power both promise and peril. At the same time, new, leaner coding models from other entrants emphasise speed and cost-effectiveness, reinforcing Liu’s claim that cheaper, capable models could blunt the competitive edge of elite chips for many production workloads. The practical upshot is a market where capability, cost and governance interact to reshape product economics and deployment choices. ^[4]^[5]

That shifting economics is visible in start‑up performance too. Liu argues 2026 will privilege recurring revenue with healthy unit economics over the “vibe revenue” of earlier years. The point is timely: as agentic tools move from R&D to customer-facing services, firms will be forced to demonstrate durable margins and measurable outcomes, or face a market correction. Investors and operators alike are already separating durable subscription businesses from repackaged services with weak margins, raising the bar for sustainable AI commercialisation. ^[1]

Geopolitics and capital flows also shape Liu’s assumptions. He predicts a surge of Western investment into “physical AI” , robotics and automated manufacturing , as part of broader industrial policy. That mirrors public moves by high-profile investors and firms to bring AI into the real world at scale, and explains why some Western players are redoubling efforts to match China’s factory automation and robotics deployments. At the same time, Liu expects Chinese models to remain influential among Western builders; venture capital sentiment and pragmatic engineering choices mean Chinese models will continue to figure in many stacks, complicating both procurement and policy debates. ^[1]

Finally, Liu outlines broader societal currents: growing scepticism towards Silicon Valley elites, Europe’s potential emergence as a “third pole” focused on trust and human-centric AI, and the moral case for building AI that amplifies human originality rather than replacing it. His admonition that “AI should not be ‘done to humans’; it must be developed for humans” echoes regulatory and industry conversations about accountability, explainability and the distributional effects of automation. Those are not merely rhetorical points; they frame the decisions boards, CTOs and policymakers must make as agentic and autonomous systems enter everyday services. ^[1]^[2]

Taken together, Liu’s dozen assumptions form a pragmatic checklist for 2026: solve privacy while enabling context, temper invective about imminent replacement with investments in human originality, force the economics of AI businesses into healthier territory, and manage the dual technical and regulatory risks of agentic systems as they move into production. As model capabilities accelerate, the quality of governance and product economics will determine which firms convert potential into durable advantage. ^[1]

📌 Reference Map:

##Reference Map:

^[1] (CityAM) - Paragraph 1, Paragraph 2, Paragraph 3, Paragraph 6, Paragraph 7, Paragraph 8, Paragraph 9
^[3] (Reuters) - Paragraph 3
^[6] (CIO) - Paragraph 3, Paragraph 4
^[2] (Reuters) - Paragraph 4, Paragraph 8
^[4] (Reuters) - Paragraph 5
^[5] (Reuters) - Paragraph 5

Source: Noah Wire Services

More on this

https://www.cityam.com/12-predictions-for-ai-in-2026/ - Please view link - unable to able to access data
https://www.reuters.com/markets/funds/agentic-ai-race-by-british-banks-raises-new-risks-regulator-2025-12-17/ - British banks are rapidly adopting agentic AI—autonomous systems capable of planning, decision-making, learning, and executing tasks—to enhance customer services like budgeting, saving, and investing. Institutions such as NatWest, Lloyds, and Starling are collaborating with the Financial Conduct Authority (FCA) to run customer-facing trials expected to launch in early 2026. The FCA highlights new risks due to agentic AI’s speed and autonomy, including potential threats to financial stability and governance, and plans to enforce accountability through existing rules like the consumer duty and senior managers regime. While Britain leads Europe in piloting these AIs thanks to a supportive regulatory approach, concerns persist over systemic risks from multiple AI agents interacting, unreliability in complex tasks, and the lack of deep understanding among banking executives. Though research estimates 40% of financial firms will adopt this technology by 2026, over 40% of such projects could fail by 2027 due to costs and unclear benefits.
https://www.reuters.com/business/microsoft-add-anthropics-ai-coding-agent-its-github-service-2025-05-20/ - Microsoft has announced plans to integrate an AI coding agent from Anthropic, a Google-backed startup, into its GitHub platform. This move is part of a broader strategy to enhance GitHub’s capabilities by leveraging AI tools that can perform tasks such as fixing code bugs autonomously. The announcement was made during Microsoft's annual software developer conference, where Executive Vice President Jay Parikh also confirmed that Microsoft recently introduced its own AI coding agent to GitHub. Additionally, Microsoft revealed earlier in the week that it would incorporate a coding agent from OpenAI as well, signaling a shift toward a more inclusive and neutral approach in the competitive AI development landscape.
https://www.reuters.com/business/retail-consumer/anthropic-bolsters-ai-model-claudes-coding-agentic-abilities-with-opus-45-2025-11-24/ - On November 24, 2025, AI startup Anthropic announced the release of its upgraded Opus 4.5 model, enhancing its Claude AI's ability to write complex code, build advanced autonomous agents, and manage enterprise tasks such as spreadsheet manipulation and financial analysis. This new release significantly improves Claude’s reasoning, memory, and performance across a broad range of computer applications including financial modeling and forecasting. A key feature of Opus 4.5 is its agentic functionality, where agents can autonomously improve and retain knowledge from previous tasks for future use. The release positions Anthropic—backed by Amazon and Alphabet—in competition with OpenAI and others in the race to develop advanced large language models, potentially surpassing human intelligence.
https://www.reuters.com/business/musks-xai-forays-into-agentic-coding-with-new-model-2025-08-28/ - Elon Musk's AI startup, xAI, has launched a new agentic coding model named "grok-code-fast-1," signaling its entrance into AI-driven programming assistants. Described as "speedy and economical," the model is designed to autonomously perform coding tasks efficiently in a compact form factor. It will be available for free for a limited time, with initial partners including GitHub Copilot and Windsurf. The release places xAI in competition with other major industry players such as Microsoft and OpenAI, who have developed similar coding agents—Microsoft’s GitHub Copilot and OpenAI’s Codex. Notably, xAI also filed a lawsuit against Apple and OpenAI, alleging anti-competitive practices intended to suppress rivals in the AI space.
https://www.cio.com/article/4064998/taming-ai-agents-the-autonomous-workforce-of-2026.html - The article discusses the rapid advancement of agentic AI, autonomous systems capable of planning, decision-making, learning, and executing tasks, and their integration into various industries. It highlights the shift from traditional AI assistants to more autonomous agents capable of handling complex workflows. The piece also addresses the challenges and considerations in developing and deploying these agents, emphasizing the need for careful design to prevent issues like hallucinations and ensuring reliability and safety in production environments. The article underscores the transformative potential of agentic AI in reshaping the workforce and operational processes across sectors.
https://www.reuters.com/markets/funds/agentic-ai-race-by-british-banks-raises-new-risks-regulator-2025-12-17/ - British banks are rapidly adopting agentic AI—autonomous systems capable of planning, decision-making, learning, and executing tasks—to enhance customer services like budgeting, saving, and investing. Institutions such as NatWest, Lloyds, and Starling are collaborating with the Financial Conduct Authority (FCA) to run customer-facing trials expected to launch in early 2026. The FCA highlights new risks due to agentic AI’s speed and autonomy, including potential threats to financial stability and governance, and plans to enforce accountability through existing rules like the consumer duty and senior managers regime. While Britain leads Europe in piloting these AIs thanks to a supportive regulatory approach, concerns persist over systemic risks from multiple AI agents interacting, unreliability in complex tasks, and the lack of deep understanding among banking executives. Though research estimates 40% of financial firms will adopt this technology by 2026, over 40% of such projects could fail by 2027 due to costs and unclear benefits.

Noah Fact Check Pro

The draft above was created using the information available at the time the story first emerged. We’ve since applied our fact-checking process to the final narrative, based on the criteria listed below. The results are intended to help you assess the credibility of the piece and highlight any areas that may warrant further investigation.

Freshness check

Score: 10

Notes: The narrative presents original predictions for AI in 2026, with no evidence of prior publication or recycled content. The earliest known publication date is December 18, 2025, aligning with the current date. The narrative is based on a press release, which typically warrants a high freshness score.

Quotes check

Score: 10

Notes: No direct quotes are present in the narrative, indicating original content. The absence of identical quotes in earlier material suggests potential originality.

Source reliability

Score: 8

Notes: The narrative originates from City AM, a reputable UK-based business news outlet. While City AM is generally considered reliable, it is not as widely recognised as some other major UK news organisations.

Plausability check

Score: 9

Notes: The predictions align with current AI trends and developments. The narrative includes references to recent developments, such as Microsoft's integration of AI coding agents into GitHub and Anthropic's release of Opus 4.5, supporting the plausibility of the claims. The tone and language are consistent with professional business reporting.

Overall assessment

Verdict (FAIL, OPEN, PASS): PASS

Confidence (LOW, MEDIUM, HIGH): HIGH

Summary: The narrative presents original, timely, and plausible predictions for AI in 2026, supported by references to recent developments and originating from a reputable source. The absence of recycled content and direct quotes further supports its credibility.

AI predictions
AI governance
AI industry