William Padula highlights a critical gap in healthcare: while cures exist, the real challenge lies in establishing systems to finance and effectively deliver these expensive treatments.

VentureBeat's Q1 2026 Pulse Research reveals a crucial gap dubbed the “Governance Mirage” — enterprises often lack the control and security they believe they have with AI governance. While 43% of respondents say AI governance is owned by a central team, 23% couldn't identify the owner, and 31% point to vendor opacity as the biggest barrier. The key failure in enterprise AI isn’t the intelligence of the model but the runtime infrastructure that struggles to manage state and ensure durability in production. Current stateless setups, like Python scripts and LangChain chains, falter as container restarts lose context, token costs rise, and hallucinations multiply, pushing teams to spend more time fixing plumbing than building smart agents.

Engineering teams face a heavy “DIY tax,” devoting much of their capacity to managing retries, checkpoints, and state persistence rather than advancing core AI logic. State amnesia—loss of progression memory—is a leading technical hurdle, alongside cost overruns and silent failures. Microsoft’s platform requires notably more observability effort, contributing to a high “telescoping” cost for visibility. Meanwhile, a significant marketing-reality gap exists, with platforms like Microsoft and OpenAI facing criticism over overstated agent autonomy and reliability.

Security strategies for AI agents are varied and still evolving, with no clear dominant approach; enterprises are building protections from the ground up, especially as agents gain deeper system access. A critical shift is underway from stateless to durable execution frameworks, though opinions vary widely on the best architectural path. Enterprises are increasingly favoring human-trust metrics, such as User Acceptance Rate, to judge agent readiness, reflecting ongoing concerns about hallucinations and failures.

Ultimately, the survey emphasizes that success in enterprise AI requires prioritizing runtime reliability, economic viability, and architectural resilience over chasing model improvements alone. Leading-edge organizations are betting on multi-architecture, deterministic systems to build durable control layers—while others risk repeating the pitfalls of earlier automation efforts. It’s clear: the reckoning in AI adoption is about runtime and economics, not just cognitive prowess.

Biotech startup NewLimit has successfully closed a $435 million funding round following its announcement of a pioneering prototype drug that reverses liver cell aging. This breakthrough promises to advance the field of longevity medicine significantly.

Spending on AI technology is soaring alongside rapid consumer adoption, yet concerns are emerging. High-profile players like Elon Musk's SpaceX, which not only builds rockets but also develops AI models, recently announced plans to seek a $1.77tn valuation on the US stock market. At the same time, Anthropic, creator of the Claude chatbot, has filed for an IPO, with OpenAI likely to follow suit. This wave of enthusiasm coincides with massive investment in infrastructure like data centers, as businesses look to make AI investments pay off. Here are six essential charts that illustrate where the AI boom stands and how it has evolved.

President Donald Trump revealed that his administration is negotiating potential deals involving an equity stake in OpenAI. The goal of these discussions is to ensure that the American public can share in the gains from advancements in artificial intelligence technology.

Andrew Bailey, Governor of the Bank of England, cautioned that artificial intelligence might have to be rationed because current energy supplies may not keep pace with AI's growing capabilities. He emphasized that governments and businesses are facing significant societal decisions, needing to balance energy demands across different sectors. The core issue is not AI's potential, but whether the energy infrastructure can support its expansion.

Maroš Šefčovič, EU trade commissioner, has suggested introducing a "diversification instrument" to lessen Europe's reliance on single suppliers for critical chips and rare earths. This proposal, presented at the European Policy Center’s Brussels Economic Security Forum, would require companies in key sectors to secure supplies from at least three different sources to enhance economic security.

Our system excelled at one task: converting natural-language questions into API calls. Analysts, account managers, and operations leads used it to bypass the manual effort of gathering data from multiple dashboards and BI tools. By typing simple English requests like “Compile a report on sales volume for January through March 2026 for the Northeast region, broken down by city,” the system generated structured JSON API calls effortlessly.

Built initially on Claude Sonnet 3.5, the system smoothly upgraded through versions 3.7 and 4.0. However, when updating to Sonnet 4.5, unexpected behavior arose: the model embedded API call parameters in the description field or started returning clarifying questions. This broke downstream systems expecting a consistent JSON structure with no need for human clarification, exposing a critical gap in our engineering assumptions.

The challenge lies in the unpredictable “blast radius” of AI model changes—unlike traditional software components, you cannot fully anticipate how an LLM upgrade will affect outputs. Our post-mortem showed our prompt lacked explicit constraints, leading to the model interpreting instructions too liberally. The solution involves adopting an “evals-first” architecture, where an extensive suite of automated tests define acceptable model behavior and gate changes rigorously.

This approach demands significant effort to create and maintain but is essential for safely deploying AI systems dependent on evolving language models. The next frontier in AI engineering is developing robust evaluation frameworks that serve as formal system specifications, bridging the gap between passing smoke tests and guaranteeing predictable, safe production behavior.

Authors: Vijay Sagar Gullapalli, Founding AI Engineer at Adopt AI, and Sarat Mahavratayajula, Senior Software Engineer at Sherwin-Williams.

For the past three years, Microsoft’s AI narrative has been tightly linked to its partnership with OpenAI, powered by over $13 billion in investments that brought cutting-edge AI capabilities to Microsoft’s enterprise products. However, Mustafa Suleyman, CEO of Microsoft AI, has announced a strategic shift. A renegotiated contract now allows Microsoft’s AI division to independently develop advanced AI, including superintelligence, using its own researchers, data, and custom hardware. This move marks the launch of Microsoft’s own in-house AI model family, MAI, which encompasses seven new models focused on reasoning, code generation, image creation, transcription, and voice synthesis. These innovations emphasize training on high-quality, commercially licensed data rather than on third-party models. Suleyman highlighted the company's long-term vision to not only collaborate with OpenAI but also build frontier AI models internally, ensuring Microsoft can maintain leadership in AI by 2030. New capabilities like Frontier Tuning enable businesses to tailor AI models with their proprietary data securely, enhancing efficiency and performance in real-world tasks. Microsoft is also investing heavily in custom AI chips, such as the Maia 200, which improve cost-efficiency significantly. This comprehensive strategy aims to create a sustainable competitive advantage through a vertically integrated AI stack that blends superior data curation, computing power, and enterprise integration. Suleyman underlined the importance of quality data and continuous model improvement, rejecting the notion that AI models are becoming commodities. Ultimately, Microsoft envisions a future with "Humanist Superintelligence" — AI systems designed to augment, not replace, human endeavors, driving a decade-long commitment to innovation in AI technology.

By Patrick Ip
Legal technology investment continues to surge, with a significant chunk focused on plaintiff-side AI applications. Last year, Goldman Sachs noted that 44% of legal tasks may be automated in the future, fueling enthusiasm for AI in legal software. Companies like EvenUp, Eve, Supio, and Darrow have collectively raised around $682 million, representing 71% of disclosed funding in legal AI. Plaintiff firms benefit from standardized workflows in client intake and case evaluation, making AI adoption more straightforward.

However, the defense side remains relatively unexplored despite its vast potential. Corporate legal teams managing extensive litigation portfolios still depend on fragmented tools such as spreadsheets and emails, lacking comprehensive software solutions that provide portfolio-wide insights into case risks, settlements, and costs. The complexity of varied workflows and longer sales cycles due to multiple decision-makers have slowed investment.

Yet, growing pressure and advancements in AI are beginning to address these challenges by enabling better risk assessment, benchmarking, and outcome analysis. For investors, this creates an appealing opportunity to back technology that can transform defense-side legal operations and build proprietary data assets. While no dominant venture-backed platform has emerged yet for defense litigation intelligence, this sector could be the next major frontier for legal AI innovation and investment.

Patrick Ip is the CEO and co-founder of Theo Ai, focusing on AI-driven litigation intelligence for defense teams and law firms.