The Writer’s Role

What's left for people to do?

I’ve been spending a lot of time recently learning to use Claude Code and developing new workshops. I’ll be announcing the workshops soon, but working with Claude Code has reminded me of the similarities between writing and programming.

Most people don’t think of writing and programming as related tasks. But programmers use code to tell stories much like a writer uses words. They’re called programming languages for a reason.

Both fields turn the abstract into reality. A writer envisions a world and characters and uses their skills to bring that vision to life. Developers envision a problem, imagine a solution, and create that solution by blending their technical and creative skills.

What does any of this have to do with Craft-First Writing?

Developers and writers are at the same crossroads. AI is threatening to replace much of the work typically done by developers, just as it’s threatening to replace many writing tasks. I watched this in action earlier this week:

One security threat of AI is that it doesn’t always understand context and is easily tricked. A common attack is prompt injection, in which an attacker writes a prompt that causes the LLM to perform a malicious action. For example, if someone is using AI to manage their email and the AI reads it, an attacker could send a spam email with instructions for the LLM to download a malicious file or to send files back to the sender.

Developing often means using third-party libraries and plugins of unknown origin. I wanted a way to protect myself from malicious code, so I worked with Claude Code and developed a plan for creating a scanner that analyzed plugins before downloading them. After approving the plan, I told Claude Code to implement the plan, and took the dog for a walk.

Before getting to what happens next, let me provide some perspective on how quickly technology changes. I once worked with a salesperson whose former job was selling a file-transfer product called FTP for millions of dollars. FTP has been free for decades. I also worked for a company that designed the architecture that makes up most websites. (We built pets.com in our conference room.) That company no longer exists. I worked for a company that developed one of the first products to scan source code for security risks. At the time, it was so difficult, the CTO said he’d never build one again. I just asked my ‘robot’ to build it while I walked the dog.

When I returned from walking the dog, phase one was completed. I gave Claude Code instructions for phase two and some test cases, then went out with friends. When I returned, I had a completed program that was validated with the same test harnesses most professional products use. If I had tried to build this myself without AI, I would estimate six to eight weeks of work.

What do we need people for?

If AI can write a month-long project in a couple of hours, what will programmers do? If AI can write an entire novel or a fiction book in less than a day, why do we need writers?

Answering that question is critical.

Where people — writers and programmers alike — shine is in the creative process. I’ll write more about this in a future post, but remember that LLMs are just math programs that make predictions, and they predict to the average. They don’t come up with new ideas. There is quite a bit of research showing that, while AI can create many concepts, they lack the diversity and newness of human-created ideas. If you ask three different LLMs to come up with ideas for a fantasy novel, all three will come up with the same basic ideas. It takes people to move things in new directions.

It also takes people to manage the job. My robot was able to write a program while I walked the dog, because I gave it very specific instructions. Admittedly, I developed the instructions with the help of an AI agent I built, but the system needed me to define the problem and express a desired solution. Then we had to develop a step-by-step implementation plan.

LLMs have limited memory — called context windows — that fill up during long tasks. When the window fills up, the LLM tries to capture the pertinent information before emptying the context window and restarting. This process often results in the loss of critical information, which causes the LLM to go off on tangents. I once had an LLM go rogue and start developing a brand-new story because it forgot what it was doing.

To get around this limitation, I have a ‘persistent-memory’ agent that periodically captures what the LLM is doing, what’s working, and what’s not, so it not only remembers but also learns from mistakes. This allows the LLM to work on long tasks without stopping. Every time the context window fills up, it just reads the memory file and catches up.

This memory limitation is also why an LLM can’t really write a novel from a single prompt. It’s too big a task. The context window will fill up, the LLM will forget some of what it’s already done, and it will start changing the story. Even outlining with the LLM is fraught with peril. If left to run on its own, the LLM will reintroduce dead characters, change names, or rewrite sections of the outline based on prior discussions.

Where do you fit in?

Humans are still the only ones with full context. You know what people are struggling with and can see trends or social movements long before AI. AI can identify patterns, which may help you select a genre or identify a new solution, but it doesn’t have a human touch.

AI doesn’t have the connection to human emotion. AI doesn’t know what problems the company is struggling with, what society is boiling over about, what stories let a woman in a beach chair unwind and get away from it all. You have that connection, and you are the one with the solutions, insights, and stories we need.

AI can implement, but it needs a manager. And remember that AI predicts the average. Stories written with AI will be average stories. Programs written with AI will be average programs. Writers who can create fresh images and draw unique connections will stand out. Writers who can build beautiful sentences and carry a metaphor across an entire story will be the ones readers search for. AI can plow the field and plant the seed, but it needs your nurturing touch to make ideas grow and flourish. If we let AI do everything, we’ll end up with dull, cookie-cutter landscapes.

What can you do today?

Find a book that you love, and really read it. What made it stand out? The words? The images? The connections you made with the characters? What made it different than the average book? Because that average book. That’s AI. If you want to stand out, make a concerted effort to determine what that means. Then develop those skills. Ask the AI to create a training plan for you!

And learn to accept that technology will change the way things are done. Programmers are more willing to jump on that bandwagon, because it’s always been that way for them. But change comes more slowly in the writing world. The jump from typewriter to word processor to computer wasn’t all that earth-shattering, and it took a long time. Learning to accept such a monumental shift is going to take the writing world some time.

Whether we like it or not, it’s a new world. But with preparation, you can flourish in it.