SOPs for AI: How Pharma's Most Underrated Skill Became the Key to Agentic Workflows

When you’re working in Pharma, you’re intimately familiar with SOPs. Standard Operating Procedures are the backbone of the pharmaceutical industry - the one constant across every department, every company, every country.

Over the years, we’ve all learned to read and write these meticulously detailed procedures. And if you’re being honest, you’ve probably wondered more than once: why do we need this level of detail? If you’re well-trained, you could figure out the steps on your own.

Here’s the twist: in the age of AI agents, that skill - the ability to explain a complex workflow in precise, unambiguous, step-by-step language - is pure gold.

The Connection Between SOPs and AI

As you probably already know, the better you can explain your intention to an AI, the better the output. The more precisely you can describe the exact workflow, the desired format, and the quality criteria, the more reliable your results become.

This isn’t a coincidence. It’s the same principle behind SOPs.

Think about what makes a good SOP: it assumes nothing about the reader’s prior knowledge. It specifies every decision point. It defines success criteria. It handles edge cases. It’s written so that anyone - regardless of experience - can follow it and arrive at the same outcome.

That’s exactly what an AI agent needs.

The difference is that your human colleague might tolerate an ambiguous instruction and figure it out from context. An AI agent will either guess wrong or ask you to clarify - and in automated workflows, it often guesses wrong silently.

The Problem: Nobody Wants to Write an SOP Every Time

Here’s the catch. Even if you know that structured, detailed instructions produce better AI outputs, nobody wants to write a two-page procedure every time they ask an AI to summarize a paper or draft a medical information response.

You’d spend more time writing the instructions than doing the work yourself.

This is the gap that Anthropic (the company behind Claude) addressed with a concept that should feel immediately familiar to anyone in pharma: Skill files.

What Skill.md Files Are

Anthropic introduced a context management system that lets you write reusable instruction sets - essentially, SOPs for your AI. These files are stored as Markdown documents (.md files) and are automatically loaded when you trigger a specific keyword or phrase.

The concept is simple:

1. You write a Skill once - a detailed, step-by-step procedure for a task you do regularly
2. You define trigger phrases - keywords that tell the AI when to load this Skill
3. The AI retrieves the Skill automatically whenever you mention a trigger phrase in conversation

For example, you might create a Skill for “medical inquiry response” that includes:

• The exact structure your response should follow
• Which data sources to check
• What safety language to include
• The tone and level of scientific detail expected
• Quality criteria for the final output

The next time you say “draft a medical inquiry response about [topic],” the AI doesn’t just wing it. It loads your procedure and follows it step by step - just like a well-trained colleague following an SOP.

Why This Matters for Medical Affairs

Medical Affairs runs on repeatable workflows. Literature monitoring. Medical information requests. KOL briefing preparation. Congress content summaries. Advisory board follow-ups. Training material creation.

Each of these tasks has a specific structure, specific quality standards, and specific compliance requirements. And each one is a perfect candidate for a Skill file.

Consider the workflow for responding to an unsolicited medical inquiry:

## Skill: Medical Inquiry Response

### Trigger Phrases
- "medical inquiry response"
- "unsolicited medical information"
- "MIR draft"

### Procedure
1. Identify the specific clinical question
2. Classify as on-label or off-label inquiry
3. Retrieve relevant SmPC sections (4.1, 4.2, 4.8, 5.1)
4. Search approved standard response library
5. If no standard response exists:
   - Summarize relevant clinical data
   - Include safety information
   - Add appropriate disclaimers
6. Format response using approved template
7. Flag for medical review if off-label

### Output Requirements
- Scientific, balanced tone
- No promotional language
- Reference approved data sources only
- Include adverse event reporting reminder
- Maximum 500 words

Once this Skill is saved, every future medical inquiry response starts from this foundation. The AI doesn’t forget a step. It doesn’t skip the safety language. It doesn’t accidentally use promotional tone. The procedure is baked into the workflow.

The Markdown Connection

If you’ve been following the discussion about structured communication with AI using Markdown, you’ll notice that Skill files are the logical next step.

Markdown gives you the syntax to structure individual interactions - headings, lists, tables, bold markers. Skill files take that syntax and turn it into persistent, reusable procedures that your AI applies automatically.

The progression looks like this:

1. Unstructured prompts - inconsistent, unreliable outputs
2. Markdown-structured prompts - better outputs, but you have to structure every time
3. Skill.md files - structured, consistent outputs on autopilot

Each step builds on the previous one. You can’t write effective Skills without understanding Markdown. And once you understand Markdown, Skills are the natural evolution.

You Don’t Even Need to Write Them Yourself

Here’s the best part - and the one that should resonate with every Medical Affairs professional who’s already drowning in documentation:

Your AI can write Skills for you.

If your company has existing SOPs - and let’s be honest, you have hundreds - you can feed them directly to your AI and ask it to generate a Skill file. The AI extracts the key steps, identifies decision points, defines output criteria, and formats everything as a reusable Skill.

The workflow is straightforward:

1. Upload your existing SOP to your AI assistant
2. Provide the Skill file format specification as context
3. Ask the AI to convert the SOP into a Skill
4. Review, adjust, and save

You’re not starting from scratch. You’re translating institutional knowledge that already exists into a format that AI agents can execute reliably.

Think about what this means at scale. A Medical Affairs team with 50 documented SOPs could convert them into 50 AI Skills - creating an AI assistant that handles routine workflows with the same consistency and compliance awareness that took your team years to develop.

From Individual Skills to Agentic Workflows

Individual Skills are powerful. But the real transformation happens when you chain them together into agentic workflows - sequences of Skills that the AI executes autonomously.

Imagine this pipeline for congress coverage:

1. Skill: Abstract Screening - Scans published abstracts, filters by therapeutic area and relevance criteria
2. Skill: Key Data Extraction - Extracts primary endpoints, safety signals, and competitive insights into structured tables
3. Skill: Internal Summary - Generates a concise summary for field medical teams
4. Skill: KOL Mapping - Cross-references presenting authors with your KOL database
5. Skill: Follow-Up Draft - Drafts personalized follow-up communications for relevant KOLs

Each Skill is a self-contained SOP. Together, they form an automated workflow that would have taken a team of three people two weeks to complete manually.

Getting Started

1. Identify your top 3 repetitive tasks - the ones you do weekly or monthly with a predictable structure
2. Check if SOPs already exist - most Medical Affairs teams have documented procedures for their core workflows
3. Convert one SOP into a Skill - start simple, test with your AI, iterate based on output quality
4. Build your Skill library gradually - add one new Skill per week, refining as you go
5. Share with your team - Skills are just text files, making them easy to distribute and version-control

The Pharma Advantage

Here’s the irony: the pharmaceutical industry’s obsession with documentation - the thing we’ve all complained about at some point - turns out to be a massive competitive advantage in the AI era.

Other industries are scrambling to figure out how to write clear, repeatable instructions for AI systems. We’ve been doing exactly that for decades. We call them SOPs.

The teams that recognize this connection - and start converting their procedural knowledge into AI-executable formats - will move faster, more consistently, and more compliantly than teams that treat every AI interaction as a blank-slate conversation.

Your SOPs aren’t just compliance artifacts. They’re the training manual for your AI workforce. Time to put them to work.