A provocative advertising campaign recently appeared across London’s subway system and San Francisco’s bus stops, telling commuters: “Stop Hiring Humans – The Era of AI Employees is Here.” These purple billboards are from a company called Artisan. They promise AI agents that can handle entire sales workflows—find leads, create personalized outreach, manage follow-ups, and even close deals. It’s deliberately controversial, designed to grab your attention.

But beneath the hyperbole and headlines, a real-world and important development is taking shape: AI systems can now execute complex, multi-step processes that once required human coordination.

How do these “agents” actually work? What can they realistically accomplish? And what does this mean for publishing workflows that have remained largely unchanged for decades?

What Is an AI Agent? It’s an LLM with Tools

Today’s AI agents are basically large language models (LLMs) that have the ability to use external tools. Think of them as ChatGPT or Claude, but with access to functions they can call—web searches, code execution, API interactions, or file manipulation.

Agents behave in a fairly straightforward way: the LLM receives a task, figures out which tool to use, uses it, checks the results, and then decides what to do next. This loop continues until the task is completed or the agent can’t go any further.

AI Agents in Action: Claude Code

One of the clearest examples of what AI agents can do today comes from software development. Anthropic’s Claude Code was launched in April 2025. Rather than just suggesting code snippets for a developer to use, Claude Code can autonomously run commands, check if the code works, fix mistakes, and keep improving it until everything is working correctly.

A Simple Example of Claude Code in Action:

A developer asks Claude Code: “Can you help me create a Python script that processes all CSV files in a folder and combines them into a single Excel file with separate sheets?”

Claude Code then:

1. Writes the initial Python script
2. Runs it in the terminal to check for errors
3. Discovers a missing library (pandas, openpyxl)
4. Installs the missing libraries
5. Finds an encoding error with one CSV file
6. Modifies the code to handle different encodings
7. Successfully runs the final script
8. Provides the working code along with usage instructions

After giving the agent the initial task, the developer oversees the process, as opposed to copying and pasting code snippets or debugging code line by line. The AI agent might have questions as it works through the problem, much like a team member, and it will work until the task is complete.

AI Agents in Publishing: A Near-Future Example

While Claude Code is up and running today, publishing-specific agents are still emerging. Here’s a realistic near-future scenario:

An editor asks a publishing-specific AI agent: “Check if all citations in this manuscript follow Chicago Manual of Style, 17th edition.”

The agent would:

1. Use a document reader tool to analyze the manuscript and identify all citations
2. Apply a citation extractor tool to break each citation into components (author, year, title, publisher)
3. Run each through a Chicago Style validator tool checking against CMOS 17 rules
4. Use a comparison tool to identify discrepancies
5. Generate a report: “Found 23 citations: 18 correct, 3 missing page numbers, 2 with incorrect date formatting”
6. Ask: “Would you like me to fix the formatting issues automatically?”
7. If approved, call the document editor tool to update citations
8. Use a change tracking tool to log all modifications made

Like the developer using Claude Code, the editor initiates the task and makes key decisions, and the AI agent handles the manual checking.

This is just one example. Using the same set of tools, an AI agent could:

• Convert British to American spelling throughout a manuscript
• Ensure consistency of character names and descriptions
• Format manuscripts to specific publishing house styles
• Verify that all chapter headings follow the correct hierarchy
• Check that all figures and tables are properly referenced in the text

The pattern remains consistent: specialized tools + LLM coordination = automated publishing workflows.

The Reality Check

This isn’t the “fire your human employees” revolution that the purple billboards promise. It’s about connecting digital tools with LLMs like ChatGPT and Claude, so they can work together automatically.

Claude Code offers a glimpse of what happens when enough tools become accessible to artificial intelligence—when an AI can read files, write code, execute it, debug errors, and iterate on solutions. It’s a tipping point. Suddenly, complex tasks become routine. Rather than having AI to help write individual pieces of code, a developer can hand over entire features to an AI agent.

Publishing may be approaching a similar tipping point. When AI agents can access document editing, style checking, keyword research, and formatting tools in combination, the nature of delegating tasks changes. Instead of manually checking each citation in a manuscript, an editor could hand over the entire citation review process. Instead of reformatting each chapter heading by hand, they could delegate standardization across a full manuscript.

For marketing teams, it might mean asking an agent to “research the top-performing keywords for this thriller and update our Amazon description accordingly”—activating the agent to search competitor titles, analyze what’s working, rewrite the book description, and show the changes for approval. Or delegating the creation of social media posts for a book launch, with the agent generating platform-specific content, re-sizing images correctly, and scheduling posts across channels.

These aren’t revolutionary changes. They’re the accumulation of dozens of small automations. But when an editor can hand off five or six routine tasks that previously consumed half their day, or when a marketer can delegate the mechanical parts of campaign creation, the impact becomes substantial. 

And the jump from substantial to transformational happens when enough tedious, rule-based tasks can be delegated that human work genuinely shifts to higher-value activities.

FAQ: AI Agents in Publishing

What is an AI agent?

An AI agent is a large language model (LLM) like ChatGPT or Claude that can use external tools to complete complex tasks. AI agents work by receiving a task, determining which tool to use, executing the action, checking results, and repeating this process until the task is complete. Unlike simple chatbots, AI agents can perform multi-step workflows autonomously.

What can AI agents do for book publishers?

AI agents can automate complex, multi-step publishing workflows including:

• Citation checking and formatting according to style guides (Chicago, MLA, APA)
• Manuscript formatting to specific publishing house standards
• Character name consistency checking throughout manuscripts
• Localizing and adapting content
• Keyword research and Amazon book description optimization
• Social media content creation and deployment for book marketing campaigns
• Figure and table reference verification in academic texts
• And much more

What AI publishing tools does Veristage offer?

Veristage’s Insight is a secure, AI-powered platform designed specifically for publishers. Insight provides:

• AI-powered document analysis including metadata generation, editorial insights, marketing content creation, and alt text for images
• AI Chat to interact directly with your books, images, and documents
• AI document tools like translation, adaptation, and index creation

We’re actively developing publishing-specific AI agents that are integrated directly into Insight, enabling more automated workflows for publishing-specific tasks like proofreading, citation checking, manuscript formatting, consistency reviews, and others.

Are AI agents secure?

Security depends on the specific AI platform and tools used. Here are some key features to look for when choosing AI tools:

• Private cloud environment for your content
• Data encryption during processing
• No-training policies to ensure your data isn’t used to train AI models
• Information security compliance

Veristage maintains robust security standards across our products and activities, including:

• ISO 27001 certification for information security management
• Commercial agreements with AI providers ensuring your content won’t be used to train AI models
• Data encryption at rest and in transit
• Role-based access control for team security
• Dedicated server and scalable cloud infrastructure

Learn More

Are you interested in learning more about AI agents for publishing? Contact Veristage to request a demo of our Insight platform and find out how we’re helping publishers take advantage of the very latest AI capabilities.

---