The AI Race Everyone’s ‘Winning’ — But Where Are They Actually Going?

My browser currently has 47 tabs open. I discovered this when it crashed earlier today, and upon restarting, the tab count greeted me like a silent reprimand. The vast majority pertain to artificial intelligence — think tutorials, academic preprints, model benchmarks, and release notes. Each one was opened with the sincere goal of thorough reading. Instead, I followed my usual pattern: saving them to a ‘To Read’ folder, closing the tabs, and carrying on with my routine. That folder now contains hundreds of links; I’ve only managed to get through a small fraction.

By profession, I am a hospital pharmacist. My everyday responsibilities involve reviewing prescriptions to verify that the medications prescribers order are safe, suitable, and accurately dosed for each individual. I also conduct research in personalized medicine, aiming to uncover why a treatment that proves highly effective in one person might fail entirely in another. This work demands patience, precision, and attention to detail — it lacks flashiness, but it carries significant weight.

Yet over the last twelve months, I’ve experienced mounting pressure to reinvent myself into a different type of professional. During the early part of 2026, major U.S. tech players — OpenAI, Anthropic, and Google — all rolled out substantial upgrades to their AI systems. Chinese firms arguably outpaced them, with Alibaba, Moonshot, StepFun, and Zhipu unveiling new models in rapid sequence. Numerous tech giants introduced subscription tiers for their AI-powered coding assistants, and the U.S. Food and Drug Administration issued updated guidance regarding AI in pharmaceutical development. The underlying message was constant and unmistakable: the world is hurtling toward AI while you remain motionless.

AI and science: what 1,600 researchers think

I first encountered this velocity back in 2023, when I was a visiting scholar at Tsinghua University in Beijing. I contributed to curating and validating biomedical training datasets for large language models (LLMs). I wasn’t part of the core model design or training pipeline, but I was close enough to the cutting edge to witness firsthand just how quickly everything evolves.

Then OpenClaw showed up. OpenClaw is an open-source ‘agentic’ tool that hooks LLMs directly into your machine, granting the models the ability to surf the web, read and modify files, take commands from messaging platforms like WhatsApp and Telegram, execute code, and carry out complex multi-step workflows with barely any human oversight. Launched last November as a developer utility, it was quickly repurposed for a far broader set of jobs and ballooned into a worldwide sensation — especially in China, where leading tech companies scrambled to produce their own equivalents. A hospital-administrator friend of mine, someone with zero technical expertise, reached out asking me how she should set it up. Once your non-technical coworkers are asking about agentic AI, the FOMO (fear of missing out) becomes nearly impossible to shrug off.

So I made an attempt. I installed OpenClaw on my laptop, two cloud servers, and my phone — then never touched it. I enrolled in an online machine learning course and bookmarked tutorials I’ll likely never complete. All the while, my genuine research — a pharmacogenomics project I’ve poured two years into — had barely moved forward. I was burning so much energy getting ready for a future that may or may not materialize that I completely neglected the work sitting right in front of me.

On top of it all, my unease only intensified. Rather than calming my FOMO, my reading was stoking it — it felt as though everyone on the planet had figured out AI except me.

Shifting sands

So I did what I should have done much earlier: I revisited my To Read folder. What I encountered was humbling. The overwhelming majority of those breathless product reveals and tool introductions — the ones cluttering my inbox from newsletters, WeChat groups, and lab Slack channels — had basically gone nowhere. Tools that had seemed groundbreaking in January were forgotten by March. Meanwhile, foundation models (AI systems trained on enormous datasets and architected for a broad variety of tasks) are advancing at such a pace that specialized tools layered on top of them frequently become outdated within a matter of months.

Six tips for better coding with ChatGPT

Chasing every single update in an environment where the underlying technology is this fluid is often an exercise in wasted motion. The most capable AI users I’ve observed — bioinformatics colleagues who use it daily and collaborators at the AI labs constructing these systems — aren’t always the earliest adopters. Far more frequently, they’re the ones who wait just long enough for meaningful signals to emerge from the noise: for practical use cases to surface, for failure patterns to become apparent, and for the wider community to rally around proven approaches.

That realization led me to the questions I should have been posing from the start: what exactly do I do? What are my genuine strengths? And where, precisely, do I actually need support?

As a pharmacist, I identify mistakes that fall in the space between a clinical guideline and a real patient. When a treatment that all the evidence says ought to work doesn’t, I don’t just chalk it up to bad luck in a statistical lottery; I probe whether another factor is at play. My whole research program is built around this very question: when a population-level recommendation falls apart at the patient’s bedside, is that simply random statistical noise, or is it a meaningful signal hinting at a biological mechanism we haven’t yet uncovered?

Once I was able to express that with clarity, the daily avalanche of AI news stopped feeling like a string of personal indictments and started to look more like a menu I could pick from deliberately.

No mo’ FOMO

Over recent months, I’ve picked up three habits that have helped me stay connected to AI developments without drowning in them.