Writing on pharma AI validation and GxP AI
I write about choosing, running and validating AI projects in pharma. These pieces cover what to do and what it costs, including where GxP rules actually apply. I also look at how regulated teams decide how much evidence a system needs.
For help choosing AI projects or running them well, see Work with me. I also help teams use AI effectively and validate where GxP applies.
Master batch record vs batch record: where generative AI fits
GenAI can draft a master batch record or recipe change. FDA and draft Annex 22 say a qualified person must review and own it. What that review and the records look like.
Deviation investigation in pharma with AI: what AI drafts and what QA decides
What an AI agent may read, draft and suggest in deviation, CAPA and complaint work, what QA must decide and sign, and how to pilot it in shadow mode first.
Context of use statement for an AI tool: how to write one
What a context of use statement for an AI tool must say, using FDA's definition and draft Annex 22's intended use section, with a short sample and the gaps that make one fail.
ALCOA+ principles for AI-generated records
How each ALCOA+ principle applies when an LLM drafts an SOP, spec or report: who is the author, what counts as the original, and what the quality unit must keep and sign.
When a GxP AI tool gets it wrong
A wrong GMP AI output is a quality event. Stop the use, keep the evidence, assess what changed, and release any fix through change control.
Which drug companies sign the most AI drug discovery deals
A count of public AI drug discovery deals by drug company from January 2025 to October 2026, with the partners each one chose.
Takeda AIMI: how Takeda's medical information AI agent keeps answers accurate
How pharma medical information AI agents keep answers accurate, using Takeda's public description of AIMI and SEIMEI: approved sources, citations and a human who signs off.
Annex 22 readiness checklist for GMP sites (EU GMP draft)
What a GMP site can do now for EU Annex 22 while it is still a draft, what to wait on, and where generative AI stands after EMA's 2026 workshop.
More AI drug deals, the same thin slice of cash
From January 1 to October 26, deal counts rose from 7 in 2024 to 11 in 2025 and 18 in 2026. Headline value was $28.8 billion in 2026 versus $31.14 billion in 2025. Disclosed upfront packages were 4.1%, 3.1% and 3.2% of headline value.
Pharma is hiring people to steer AI agents
Confirmed pharma job postings put people in charge of building and directing AI agents. Measured gains remain unproven.
Full agentic workflows: when they are worth it, and what to make deterministic
When a full agentic workflow is worth it, and how to tell which steps can be deterministic instead.
How to choose which AI projects to fund in pharma, and which pilots to stop
In my experience, a lot of pharma AI ideas don't need AI at all. I rank the rest by value against full cost, and I say when a pilot should stop.
Can you use GitHub Copilot, Codex or Claude Code for GxP code?
In my reading, usually yes. Code for non-GxP apps that is reviewed before release is ordinary work. Code in a validated system is validated for that use.
Twelve AI concepts for pharma teams
Twelve AI terms for pharma and Quality teams, from agents and RAG to evals and model drift. Each one defined, how it shows up in regulated work, and an example.
Pharma AI jobs, Q4 2026: about 1 in 10 non-AI job postings ask for AI skills
What nearly 1,000 pharma job postings with no AI in the title ask for, what the AI jobs ask for in return, and what that means for your career.
Solutioning AI in GxP: how to choose the route and prove it in phases
Once an AI idea has passed the assessment, the next call is how to build or buy it and how to prove it. A seven-step pattern for pharma: carry the value case forward, check what you already own, compare routes by how the validation work is divided, design for adoption, and gate every phase on quality first, then value, usage, reviewer burden and cost.
Can you use ChatGPT, Claude or Copilot for GxP work?
In my reading, usually yes. Context of use decides: once output becomes a GxP record, an SOP, or a specification, normal review applies.
Lock the process and what it costs before you look for AI
How I map a process as it actually runs, put a baseline on it and lock it with the people who do the work before anyone proposes a tool. One made-up lab-supply example, from the first walk-through to the short candidate list handed to Assessment.
Before you ask if your AI is high risk, list it
You cannot tier what you have not listed: a plain case for an AI use-case inventory first, with twelve fields, three internal control tiers and a separate EU AI Act legal-review flag.
One AI warning letter is not a trend. The rules behind it are.
FDA has published one warning letter about a firm using AI to create GMP documents. What the record shows, what it does not show, and nine practical checks for AI-assisted GxP work.
Where flagged GenAI disclosures appear first
As of Oct 2026, flagged GenAI entries in the Pharma AI Tracker appear most often in enterprise tools, and enterprise is often the earliest area.
The most valuable AI call you'll make is often 'not this one'
When a pharma AI idea is first proposed: write the value case, sort it into four buckets, and use generative AI only when a rule will not do. Three sourced cases and one example workflow, with some GxP points for each.
AI validation in GxP doesn't have a settled playbook yet. Here's where regulators stand as of fall 2026.
A dated map of the AI guidance a CSV, QA, or AI-governance lead can actually lean on: what is final, what is still draft, where CSA fits and where it does not, and a practical checklist anchored to the parts that won't move.
The GenAI routing layer is the least-documented part of the pharma stack
When a company assistant sends a question to a model, the public record rarely says who picks that model. In the Pharma AI Tracker, a search of the September 2026 snapshot, 349 entries across 49 companies, for gateway and routing terms returns nothing. That is a gap in what companies say in public, not a finding about what they run.
Where pharma AI actually is: what I saw at the 2026 ISPE AI Summit
Two days inside the room at the 2026 ISPE AI summit: where regulated-pharma AI is really deployed, where it's going, and the one control almost nobody owns yet.
Security is now a GMP requirement
The 2025 Annex 11 rewrite puts pen-testing, MFA, and audit trails into GMP text. For AI systems, that makes the security controls and the validation controls the same controls.
How much validation does your AI actually need?
GMP AI risk turns on influence, consequence, uncertainty, complexity, and whether a failure is easy to see. Workshop input, not final guidance.
What a CSV validator should look for in an LLM gateway
A pilot went well and the business wants it in production. The CSV team opens the architecture diagram and sees 'LLM gateway' in the middle. Here's what that means for validation and which three failures to reproduce before signoff.