I’ve been spending a lot of time with LLMs lately — primarily Claude Code, and Gemini — to try to bring organization to a couple of projects that I’ve been working on. The first is an inventory report internally to compare three different sets of data; the second is a file permissions report for Google Drive to review access controls for some of our shares that we’d like to be less permissive; the third is a dashboard for some data that we’ve been working on pattern-spotting in.

All of these efforts are what I’d call scratchboard work:

• They’re important to help look for patterns
• They’re not mission critical resources
• They don’t prescribe courses of action on their own

I would class them as part of the thinking process, but not the final result of thinking process. LLMs have become very powerful tools in the pursuit of thinking, but by themselves, they are not thinking.

Not one of these tools is one that I would rely on without an engineer to help make sure they’re accurate, performant, and built for the longterm. They’re hacks that I’ve worked on to help solve a problem once, not a piece of engineering that I would rely on for more than one project.

The best thing that we’ve found so far is that we’ve found data that could benefit from a real statistician’s treatment. The LLM couldn’t find what we were looking for, but it’s shown enough signal in the data for us to find and hire someone who can help us build the real predictive analytics that we need to make better choices around a specific set of business decisions.

We are midway through a contract engagement to do that important work, and mine that data for actual signal. The tool was good enough to show us a pattern, we interviewed to confirm our suspicions, and now we’re applying a different set of tools to build a better path forward for the business.

---

When I was first in IT, I worked for NCEE, helmed by Marc Tucker. Marc had built his career talking about career development and educational development, starting with a report called “Thinking for a Living: Work, Skills, and the Future of the American Economy”. The report made some strong claims that the future of our (meaning, the United States) economy depended on jobs that provided employers with the ability to make big decisions, steer the course of the economy in new directions, and do so based on the strength of the reasoning, skills, communications, and efforts of its employees.

Marc was a spectacular human to work for – empathetic, encouraging, smart, and generous – and he was one of the better CEOs I’ve ever met or worked with. He was constantly encouraging his teams to do more, to stretch their skills, and to go back and learn more. It was his encouragement, and that of another close friend, that encouraged me to go back for my Master’s at Virginia Tech in STS.

Marc was very clear that we all had to go on learning, that we had to build skills throughout our careers, to be truly successful and to be part of a successful economy.

---

As part of that program at Virginia Tech, I spent a lot of time learning and thinking about what it means for technology to be in use within an organization. My final projects spent a lot of time on social media adaptation, focusing on patterns of utilization. One of the really important parts of the program was discussion about how technology gets adopted and made. Pinch & Bijker published a lengthy tome called the Social Construction of Technology. It’s a sociology-of-technology book that’s somewhat famous, and has become very central to how I see technological adoption rates in IT.

In the work they contend that:

• Technology does not determine human action.
• How technology is used is highly dependent on its human adoption and adaptation.
• One technology’s superiority over another is not purely technical in origin.

So, too, must it be with modern tools.

In a world with LLMs and other similar tools, what does it mean to be someone who writes? What part of the technological advances allow us to democratize opinion and expression — something promised for a long time by the Internet overall, going back to the early days of blogs — and what stays as authentic craftsmanship of writing in a world where LLMs exist?

I don’t have all the answers. Hell, I’m not sure I have an answer.

I can tell you how it makes me feel when I know I’m reading something that was LLM-enhanced or -derived, and that’s overall negative. Perhaps it comes down to my own processes. Let me explain.

When I write a conference talk — which is most of the writing I do now — I build an outline, starting from index cards. The process is about charting ideas. An index card might have a few words, or a few words plus a sentence or two on the back. Cards get strewn about the table where I’m working, moved and moved again, until I have most of an outline. From there, I gather things into a formal outline, and if necessary, a full script.

Once I’ve got the script, then I start the slides, get them to a rough-and-ready point before working to align everything and testing the process.

A conference talk, from start to finish is anywhere between 40 and 100 hours worth of work to put together, done all by myself. What I get from the experience is the process of thinking through all the parts of the talk, crafting a narrative from start to finish, and opportunities for engagement from the broader audience.

Could I do it in less time?

I’m sure I could. I’m not sure I want to, though.

The process I have works well for me in terms of getting myself into the talk, understanding all the parts of it, exploring the concepts well, and creating the final process. I think my talks have been well received the last few years: my speaker ratings have been high, and my invitations to speak have continued to arrive. Neither are a perfect way of evaluating a talk’s success, though: ratings are not compulsory, and it’s hard to know if there are other reasons I might be invited to speak.

In the end, for me it comes down to craft: what does it mean to think for a living and communicate well? Does it mean building a list of points and then using an LLM to sort out the flowery bits of writing and connective tissue? Or does it mean spending the time to choose the right words, the right level of clarity?

For the sort of blogging I do now, I think I far prefer the latter, but it’s become impossible to spot what’s good writing done by hand, and what’s been done by an AI with a good prompt.

So, I have to turn back to Pinch & Bijker, and my education, which taught me that perhaps how something is built could provide key understanding as to how something is used. A friend recently compared LLMs for writing to the typewriter, the word processor, and spell check. Each of these felt not-quite-right to me as analogs.

Typewriters changed how legible writing was, but it didn’t make word choice for the author. Word processors made it easy to store and recover writing, but it didn’t do the writing for the author. Spell check could help you avoid embarrassing mistakes (well, some of the time!) but wasn’t doing the writing.

At the end of the day: for me, writing is thinking. Writing is the physical expression of a thought process, of the communication that makes us all humans, and in wont of expressing our ideas and evidences.. Perhaps that’s why I am not in line with some of my friends and peers in the industry with the use of LLMs for writing.

The professional writing that’s done publicly (some might say performatively if they’re not being charitable), is the sort of reputational writing that has become a part of professional culture. Those articles get re-shared on LinkedIn, or passed from Slack to Slack, because they represent a memetic kernel of knowledge that Cory Doctorow once called “whuffie” in Down & Out in the Magic Kingdom as a way of building street cred.

But then I realize that for some out there, coding is thinking, also. The tools that I benefit from in the opening are the same ones that I struggle with when others use them for communicating.

None of this is straight forward. None of this has clear answers.

What this calls for is exploration, for patience, for “no hard decisions,” while we see how the rest of things shake out. There are challenges with LLMs that are also outside the technical realm that may make their point-of-use more controversial for the more human adaptations: cost of use, cost of environmental use, corporate structures and ethics, etc. There’s no shortage of options here.

It’s time for us to ask more of the questions regarding authenticity, disclosure, and utilization of these new tools, before we just embrace their wholesale use. What are we gaining? What are we losing? What does this mean about how technology works? What does this do to productivity? Is what we’re getting good? How can we be sure?

There’s plenty to question.

But we can’t stop thinking for a living if we want to be successful.