Tech Digest – February 8, 2026

The Self-Improvement Loop

AI Now Writes Its Own Code and Designs Its Own Components — The Feedback Loop Is Closed

Anthropic’s Chief Product Officer confirmed that “effectively 100%” of Anthropic’s product code is now written by Claude. In parallel, Anthropic released a 2.5x-faster version of Claude Opus 4.6 through Claude Code and its API, accelerating its own development pipeline further. Meanwhile, DeepMind used AlphaEvolve to discover a new nonlinear activation function called “Turbulent” that outperforms the standard ReLU by 3x — meaning AI is now designing the internal components that make AI work better. OpenAI’s Noam Brown predicts that by year-end, agent autonomy horizons will be so extended that measuring them will become the main challenge.

Note: OpenAI’s model release cadence has compressed from 97 days to 29 — a 3x acceleration. That’s the downstream effect of this loop: when AI builds itself, the pace isn’t linear anymore. Anyone planning a digital project around today’s tool capabilities is planning against a moving target that accelerates between budget cycles.

Sources: Anthropic CPO (video), Claude, arXiv (AlphaEvolve), Noam Brown

Research Automation

Claude Opus 4.6 Takes #1 in Physics Reasoning — xAI Co-Founder: “A Claude Code Moment for Research Is Not Far Off”

Claude Opus 4.6 scored 12.6% on the CritPt physics benchmark, surpassing GPT-5.2’s 11.6% and nearly tripling its predecessor Opus 4.5’s 4.6% score. Igor Babuschkin, xAI co-founder, tested the latest generation of major AI models on theoretical physics research and described Claude 4.6 as having “a very detailed understanding” of the field. He concluded that a “Claude Code moment for research is not that far off” — meaning AI agents running sustained, autonomous research workflows.

Note: When a competitor’s co-founder publicly praises your model’s research capability, the signal is hard to dismiss. Research automation doesn’t replace scientists — it changes the cost and speed of investigation. Public health agencies, environmental bodies, and any institution that depends on research outputs will feel this shift within the next funding cycle.

Sources: Minyang Tian (CritPt), Igor Babuschkin

Hardware & Infrastructure

Nvidia Explores Co-Packaged Optics to Eliminate Electrical Bottlenecks in AI Servers

Nvidia is working on server designs that use co-packaged optics — integrating fiber-optic connections directly into chip packages to bypass the electrical bottlenecks that limit data transfer speeds in current AI infrastructure. The shift, reported by the Wall Street Journal, reflects how quickly compute demand is outstripping traditional interconnect technology. Corning, the fiber optics manufacturer, is a key supplier in this transition.

Note: This is a supply-chain signal, not just a tech story. When the compute layer restructures its physical interconnects, procurement timelines for data center equipment shift, lead times change, and infrastructure specifications written today may be outdated by delivery. Any institution planning major IT infrastructure should build in flexibility for hardware generations that haven’t shipped yet.

Sources: Wall Street Journal

Biomedical Acceleration

Polygenic Risk Screening Could Prevent Nearly a Quarter of Premature Deaths

A study published in Nature found that polygenic risk screening — using genomic data to identify individuals at elevated risk for common diseases — could reduce premature deaths by 23.3%. Separately, South Korean researchers published a spray-on hemostatic agent in Advanced Functional Materials that stops bleeding on contact, potentially transforming emergency trauma care into a rapid field procedure.

Note: A 23% reduction in premature mortality is not an incremental improvement — it’s a structural shift in how public health systems allocate prevention budgets. Genomic screening at population scale is moving from research to policy discussion. For health ministries and insurers, the question is becoming “when,” not “if.”

Sources: Nature, Advanced Functional Materials

Capital & Workforce Signals

Top 5 AI Startups Now Outvalue Every Dot-Com IPO Combined

Silicon Valley Bank’s latest State of the Markets report shows that the five most valuable AI startups have surpassed the combined valuation of all dot-com era IPOs. The concentration of capital in a handful of AI companies signals where the market expects long-term returns — and where institutional procurement budgets will increasingly flow.

Sources: Silicon Valley Bank

January 2026: Worst Month for US Job Cuts Since the Great Recession — Europe Already on the Same Trajectory

Forbes reports that January 2026 was the worst month for job cuts announced by US employers since the Great Recession. AI is cited as a direct factor, alongside tariffs and broader economic slowdown. The pattern is not US-only. The European Trade Union Confederation reports that the EU has lost 2.3 million manufacturing jobs since 2008, with nearly a million lost since 2019 alone. Layoff announcements continue across Europe’s automotive sector (Porsche, Michelin, Valeo, ThyssenKrupp), banking (Commerzbank, Santander), aerospace (Airbus), and tech (Infineon, Nokia), with the ETUC describing Europe as “hemorrhaging quality jobs.”

Note: The US numbers are a leading indicator, not an isolated event. Europe is tracking the same curve with a slight lag — and with fewer of the AI companies generating the replacement jobs. For any institution planning workforce programs, training budgets, or service delivery staffing over a 3-5 year horizon, the baseline assumption of stable employment is already outdated.

Sources: Forbes, European Trade Union Confederation