Manifesto

What this is and why it exists.

Pharma and biotech R&D used to run on a mentor–mentee system. Senior scientists trained junior scientists. Junior scientists did the execution. Seniors reviewed, corrected, taught. The next generation of seniors came up that way too. That system is being replaced — quietly in some companies, openly in others — with a mentor–AI system. The execution work that juniors used to do is now AI's. The senior's role is shifting from training the next human to architecting the human–AI workflow the team works inside. This site is the field guide for that transition, written for the scientists actually doing it.

Part 1: The decomposition thesis

The first move, before any architecture, is to decompose your own work honestly. Decomposition is the test: can this part of my work be packaged into a reusable protocol an AI can run? If yes — literature scans, first-draft writing, regulatory checklist generation, code review, summary generation, exploratory data analysis — install a Skill that does it, and stop spending your hours there. If no, that's the part that's uniquely yours. The part your eight years of dissertation depth actually paid for.

A PhD biostatistician contains ten years of concentrated knowledge about how clinical data fails in ways the model doesn't predict. A medicinal chemist contains years of judgment about which structural modifications will fail in vivo even when they look beautiful on paper. A regulatory affairs scientist contains tacit knowledge about which sentences in a Type B meeting briefing book the FDA will actually read. None of these decompose cleanly. All of them sit on top of work that does. The decomposition problem is the visibility problem: how do you free yourself from the work that AI can handle, so the work that only you can handle becomes the thing a hiring manager (or a paying client) actually sees?

The answer is not to dumb your expertise down. The answer is to delegate everything around it — accurately, deliberately — until what's left is the part nobody else can do.

Part 2: The architect thesis

Decomposition is necessary but not sufficient. The senior scientist who only decomposes their own job is buying themselves time, not relevance. The senior scientist who builds a human–AI lab — designs the workflow, writes the protocols, encodes the review patterns, sets the handoff rules — is doing a job that capital actually needs done.

Capital does not preserve career ladders. It preserves whatever produces output cheapest at the required quality bar. In bio/pharma R&D specifically, the cheapest path to acceptable output is increasingly "one senior scientist + AI workflows + minimal junior headcount." That's not a forecast. That's a hiring trend already visible in industry data: pharma entry-level R&D positions dropped meaningfully through 2025, while senior-IC roles continue to be filled. The companies aren't hiring the same shape of team.

The opportunity inside that shift, for late-start technical professionals: if you can demonstrate that you can both do hands-on AI-leveraged work and architect the system around it, you become the person worth keeping. The 22-year-old AI-native MS holder can ship a workflow fast. They can't yet tell when an analysis plan is statistically optimistic, when a regulatory submission has a methodologically subtle flaw, when a trial design is underpowered for the actual question. You can. The question is whether you can make that judgment visible and reusable — encoded into the workflows your team runs, instead of locked in your head.

That is what an architect does. This site is for scientists who want to be that.

Who this is for

Late-start technical professionals in biotech, pharma, and clinical research. Specifically:

  • PhDs and MS holders who finished their degrees between 27 and 35
  • Biotech and pharma R&D scientists at the bench-to-data interface
  • Biostatisticians, clinical data scientists, computational biologists, ML engineers in life sciences
  • Postdocs actively considering or in the middle of an academic-to-industry transition
  • Clinical research professionals in trial design, regulatory affairs, medical affairs, biomarker discovery
  • Scientists working on FDA-regulated AI/ML, real-world evidence, or pharmacovigilance
  • H-1B holders managing the additional constraint of visa status while doing all of the above

The common thread is not the degree. It is the combination of deep domain expertise in the life sciences and a late start in a job market that prices early career velocity. The conventional career advice for this group is useless because it was written for people who optimized for velocity from the beginning. This site is for people who optimized for depth, and now have to figure out how to leverage that depth at the speed AI made the new floor.

What we actually do here

Three things, all in service of helping you build your own human–AI lab:

Tools. Open-source software evaluated for the specific bottlenecks bio/pharma scientists actually hit. Not every tool that exists. Tools that solve real problems for this audience: literature management, lab notebook systems, statistical workflows, AI-augmented coding for analysis pipelines, regulatory document drafting, clinical data wrangling.

Guides. Long-form articles on the decisions, transitions, and architectural choices specific to this audience. Not "how to get a job" generically. How to navigate an H-1B layoff with 60 days. How to translate dissertation depth into industry-readable evidence. How to know when a postdoc has run its course. How to wire AI into a biostatistics analysis plan without losing methodological rigor. Topics the general career internet does not cover with the specificity this audience needs.

Claude Code Skills. Reusable AI workflow components that take a chore off your plate and make the workflow inspectable to your team. This is where the architect thesis becomes literal: you write a well-specified protocol once for a thing you used to do by hand, the team runs it consistently, and the review pattern stays visible to everyone who needs to evaluate the output. We test these and document what works.

What we are not

This is not a newsletter that sends ten links every Sunday. It is not a LinkedIn presence competing for engagement. It is not a community platform. It is not a job board. It is not generalist career advice with bio/pharma flavor sprinkled in.

It is a reference and a field guide, scoped tightly. You come here when you have a specific problem in your bio/pharma R&D work that AI tooling could plausibly solve. You leave with a tool, a skill, a workflow recipe, or a guide that addresses it — written by someone who has actually done the work in your domain.

About Vera

This site is written and maintained by Vera — a working biostatistician with PhD-level training and active research in human–AI collaboration frameworks for clinical and pharmaceutical R&D. Her papers under review focus on the methodological discipline required when AI workflows are inserted into FDA-regulated analysis pipelines: where it improves reproducibility, where it introduces silent failure modes, and how to design review patterns that catch the latter before they ship.

The "Tested" badge on tools and skills means Vera personally ran them in real R&D contexts, broke them, and wrote up what she found. Currently no tools or skills are marked tested — that bar moves up only with first-hand use, not with vendor copy or anyone else's review.

We are not sponsored by any of the tools we cover. We do run AdSense to keep the site running. That is all.

If you are in this audience, the newsletter is the right next step. One email when something genuinely useful ships. Nothing else.