Noor’s Newsletter — Issue #13
This isn’t a recap of the weeks—it’s an attempt to understand the forces reshaping how we live, govern, and evolve.
AI is making waves
Recursion Pharmaceuticals reported positive Phase 1b/2 data for its AI-discovered drug REC-4881 from the ongoing TUPELO trial. REC-4881 is an allosteric MEK1/2 inhibitor originally discovered by Takeda (as TAK-733) and later repurposed by Recursion after their AI phenotypic platform identified its potential for Familial Adenomatous Polyposis (FAP), a rare genetic disorder that leads to hundreds of precancerous polyps in the colon. The important point here is not only the clinical signal, but the path that got them there: REC-4881 emerged from Recursion’s AI-driven phenotypic screening engine — the “Recursion OS” — and is now showing human proof-of-concept. The company plans to engage the FDA in the first half of 2026 to define a potential registration pathway and broaden the trial’s age range. The market reacted quickly, with Recursion’s stock jumping around 8% following the announcement.
Cancer diagnostics are also seeing their own wave of momentum. The FDA has proposed down-classifying many nucleic acid–based oncology companion diagnostics from Class III to Class II. If adopted, this would move a large category of oncology CDx tests from the demanding PMA pathway to a lighter 510(k) route with special controls. The agency argues that the technology has matured, with over a decade of use without major safety concerns, and that risks can now be adequately managed. Should these rules be finalized, the impact could be significant: faster development cycles, cheaper validation pathways, and more agility for precision oncology. It would also ripple into adjacent areas, including companion drug development and AI-driven diagnostic models that depend on standardized molecular testing infrastructure.
The implications extend beyond regulatory workflow. Oncology diagnostics has historically been underfunded — long approval timelines, uncertain reimbursement, and the lack of clear regulatory predictability have all held back innovation. A down-classification could unlock investment and restore the kind of experimentation the field desperately needs. We have already seen renewed excitement around early cancer detection from liquid and non-invasive samples, largely powered by companies such as Exact Sciences, Freenome, GRAIL/Illumina’s spinouts, Guardant, and Roche. The space is enjoying meaningful capital inflows and substantial M&A activity, most notably Abbott’s acquisition of Exact Sciences for $21bn.
While established players continue to scale and validate the market, a new wave of AI companies is quietly building the pipes for what comes after detection: disease characterization, diagnostic refinement, progression prediction, and treatment-response modeling. Companies like Noetik, Artera, and my own, SpatialX, are pushing these boundaries. The field still feels early, but not for long — the moment clinical workflows stabilize around molecular and imaging-rich diagnostics, these AI tools will inevitably become part of day-to-day practice.
Interesting Things in Research and Beyond
The OpenAI State of Enterprise AI 2025 report landed this week, and it proves that AI is no longer a side tool but an increasingly central part of how we work. Usage is exploding — weekly message volume has grown roughly eight-fold, and token consumption per organization has risen around 320× year over year. The underlying data, pulled from enterprise customers and a 9,000-person worker survey, points to workers saving 40–60 minutes per active day on average.
What’s more interesting, but not entirely surprising, is that adoption in the unusal suspects, such as in healthcare, manufacturing, and other traditionally slower sectors, is accelerating quickly. And while this report focuses on OpenAI, I doubt the pattern differs dramatically for Anthropic, Google, or others. Most of us already mix tools — Claude for coding and creative writing, Gemini increasingly for research and analysis, and ChatGPT for synthesis and exploration. The future, in my view, belongs to whoever owns the vertical stack and can integrate most deeply. Google likely has an advantage here: one of the biggest frictions today is the constant copy-and-paste between apps. The moment LLMs are natively wired into Docs, Sheets, Slides, Pycharm, and Gmail, our workflows will be reinvented, and the meaning of “work” will shift again. I’m personally very ready for that world.
Governance turbulence — UK legislators push for tighter AI controls
As AI becomes embedded in more industries, political pressure is rising. On December 8th, more than 100 UK parliamentarians called for binding regulation on “the most powerful AI systems,” arguing that unchecked development of high-capability models — especially those with self-training or emergent-intelligence potential — poses existential risks on par with nuclear-era technologies. The call is coordinated by the nonprofit Control AI and includes former defence and environment ministers, and the defence secretary who warns that superintelligent AI “would be the most perilous technological development since we gained the ability to wage nuclear war”.
The UK government has promised AI legislation in 2025 and potential binding agreements for developers under the AI Safety Institute. These are reasonable steps, but with the U.S. and China leading the race and the UK already trailing, it’s worth asking whether tighter brakes will close the gap or widen it. The global race for AI supremacy is accelerating; whether new rules will be timely or effective remains an open question.
On a related note, this recent piece from the Financial Times is worth reading; Graphcore’s founder offers a candid diagnosis of the UK’s challenge in building globally competitive tech companies. His argument is that capital is only part of the story. The real gap is experience — the U.S. has decades of founders, operators, and investors who have repeatedly built unicorns and know exactly how to guide new ones. Firms like a16z and Sequoia bring not just money but customer networks, connections, and early revenue pathways that are very hard to match in the UK. It’s a compelling take on what it actually takes to build global AI champions.