LLMs are Dissolving & Creating Work | #288

I’ve had a weird experience over the last month. I’ve basically become a passable junior software engineer. With LLMs writing most of the code in Cursor, I’ve built and deployed multiple web-based apps for myself, a few for fun, and some for friends, for kicks. I wrote about my initial explorations here a few weeks ago.

Last week, I built two things that were a little more advanced and likely would have taken me weeks if not longer before LLM-assisted coding tools.

First, a dashboard for my pathless path sales (which automatically updates based on a spreadsheet I was already using (4 hours of work)

Second, a personal health dashboard of my blood tests over time, something I’ve always wanted to have and likely would have paid for if it existed

This has been pretty shocking for a few reasons:

1. Before this I knew enough to be dangerous. I could trial-and-error / google to hack my way to basic solutions in HTML and CSS. I sat down a couple times and spent a couple days learning the basics of coding but gave up twice.

2. Now, because of the power of current LLMs, I have this skill. It feels like cheating. It would have taken me months of really grindy effort to be able to build my own programs in the past.

3. Some people might say that I’m not actually coding. I do agree to a point. A lot of what I’m doing is writing requirement documents and prompting the coding agent stuff like this:

   1. “That didn’t work, what’s wrong?”

   2. “Here’s this screenshot, pls fix?”

   3. “Still not working, can you walk me through your reasoning?”

   4. “Why am I getting this error?”

   5. update: new claude 3.7 models just keep working, and can do 25 tasks in one-shot, now this amount of back and forth is diminished quite a bit

4. The funny thing about this is that I am learning quite a bit. I’ve made it much further than I ever have in the past in terms of learning about IDEs, coding best practices, terminal commands, git commits, and how to use a wide range of developer tools like github, vscode, vercel, netlify and understanding various frameworks like Tailwind, Next.js, and more.

5. The reason I’m learning this stuff is simple: it’s fun. These tools have removed a lot of the friction from coding and because of that, I am less likely to give up.

What this means about how work might change in the coming year(s)

Experiencing this sudden ability to code has been jarring.

It has made me realize how much our perception of work is shaped by completing tasks. Many people derive satisfaction from finishing a to-do list, no matter what kind of work they are doing. People talk about having “productive weekends” like it is their whole purpose in the world.

Looking back on my consulting career, it’s kind of funny to think about how much low-skill admin work is being done by highly paid people: taking notes, cleaning and reorganizing data, reformatting and moving things between documents. A lot of this has already been replaced by LLMs or will soon. When I was at BCG, it was popular to use the word “crank” to talk about being locked in and focused. I was a weirdo who took lunch away from my computer each day and I can’t even count the number of people who declined invites to join me by saying, “I’m just going to crank through lunch.”

I suspect a lot of this kind of crank-work will disappear in the next couple of years. I use the word dissolve in that it’s not that tasks will be replaced or automated in a traditional sense, but rather that they simply won’t need to be done at all. They’ll either become instant, free, or obsolete.

In consulting, working as a researcher in one job, I had the task of putting together an in-depth research report on labor unions and how they impacted productivity. For a few weeks I did intensive research: googling, reading reports, compiling information, and more. Now, this crank-task is done in 64 seconds using Grok 3’s “deepsearch.”

This clearly changes the role of the knowledge worker from someone who is moving knowledge around to someone that needs to generate unique knowledge. For the organization, as

Ethan Mollick

writes, it means shifting away from aiming for automation and to thinking about augmentation:

”First, the focus needs to move from task automation to capability augmentation. Instead of asking "what tasks can we automate?" leaders should ask "what new capabilities can we unlock?" And they will need to build the capacity in their own organizations to help explore, and develop these changes.”

There were always people focused on these things, in lean operations groups, process redesign groups, and so on, but they were never the core groups of people getting stuff done. Those people will be much more important and will be the kinds of people we need in organizations in the future.

But many people will not like this. If my work in consulting is any indication, people hate change. They will rebel, slow things down, and refuse to try new things no matter what you try. I don’t know if people can really resist this kind of technology but there does appear to be some time…

I did an informal poll in my Substack newsletter's chat this week, and it seems many organizations are still one or two generations behind in deploying LLMs throughout their organization. Most people are using them to rewrite emails and summarize information. They aren’t yet seeing tasks disappear or weeks of work dissolve. But as these models roll out, they will become embedded into the software layer and the median employee will see some of their work disappear.

But what happens then? In theory, this shift should free people up for deeper thinking and more creative work. But I suspect most people don’t want to work like this. A lot of people don’t like thinking too much. They just want a job and a sense of accomplishment. They want to work like most other people are working and go about their lives.

This shift also raises uncomfortable questions about work and careers. If low-level tasks disappear and we now have intelligent thought-partners that can co-create with us, why would companies structure their organizations in the traditional pyramid? Why spend your time training 10-15 people, of which 1-2 may stay with the company? Why not hire 1-2 that are already really good and pay them much more?

More broadly, if the median person can do more things, what does the labor market look like? In my example above, I basically acquired a very rudimentary junior developer skillset. This will only get easier (and it actually has with claude 3.7 release this week - see this quick video creating a website on claude with only my book as input):

At Zillow and other companies, non-coders are already shipping production ready code.

This has two implications I think:

• The labor pool is more fluid in terms of what people can do

• This also creates a lot more potential work because the cost of doing any random task in the economy falls (humans will always create work!)

I suspect these two things will shape the future economy: labor and labor rates will drop dramatically in some sectors while endless amounts of work and opportunity will be created: new kinds of companies, more opportunity for solopreneurship & gig work, and new ways of building teams and organizations.

Of course this will take time to shake out.

But it’s clearly starting and I don’t think people quite realize it yet.

My advice: play with these tools. Experiment with them at least once a week. Try to do something you didn’t think you could do. See how it changes your perception of your own work and your future path. It might be jarring but it might be exciting.

Some more half-baked guesses about all of this:

1. People with hard-won tacit knowledge no longer need junior teams doing the grunt work. This benefits entrepreneurial individuals who want to go solo.

2. I also suspect this loosens the constraints on many people, including executive types to act prosocial. With a smaller “clueless” group in an organization, the sociopaths will be less constrained by norms.

3. Startups can gain massive cost advantages against established incumbents. They’ll still have to compete on other things like relationships and brand but the cost differentials are going to be hard to defeat. When I worked in executive search, the industry thrived on labor-intensive processes that LLMs can now dramatically streamline. I suspect these firms will be really slow to shrink the workforce as the industry is a reputation based oligopoly, but it is now inevitable.

4. There's both diminishing returns to "putting in the hard work" on a traditional path and much higher returns to doing it on work you deeply care about. Find your Good Work!

5. The nature of expertise is shifting from information gathering to information synthesis and creative application. I’ve been teaching things like synthesis for years, and 60-70% of people really struggle to do it. Now they have the basic version of that skill at their disposal.

6. People with strong work identities will struggle. Authors who pride themselves on slow, grinding work and painful writing processes will not be happy about the millions that can write relatively close to their own level. This has essentially already happened to codes. Don’t identify too strongly with a task or role. You may need to reinvent yourself in the next 5 years. Remember, you’re just human.

7. The gap between personal tech capabilities and workplace restrictions will create tension, as employees increasingly recognize they could work more efficiently with tools they're already using at home.

The returns to sharing more, writing your own ideas down, developing your own principles, and riffing more unstructured thoughts are increasingly worthwhile activities (of course this last one is to justify the unstructured nature of this post).

Onward!

Here’s a quick reflection I did on writing too:

Paul MillerdFeb 25

some half-baked but informed by hacking around with LLM + writing for a month or so:

- It's a huge advantage to have a past body of work and style in terms of fine-tuning and training. It can help you understand your own style and keep evolving that over time independent of LLMs

- Vibe Writing will be a thing in 6-12 months if not the next couple of months. LLM suggest edits => accept. The biggest thing stopping this is reliable output based on an input/preferred vibe. It is getting CLOSE

- Mu…

34

10

1

hey

I’ve been having fun and hanging out online since 2015. I’ve somehow figured out how to hack a living doing things like this for more than seven years. Wild.

If you like what you read here, you’ll probably enjoy my books The Pathless Path and Good Work.

If you’d like to meet others on “pathless paths”, you can join The Pathless Path Community.

Some things I endorse: Crowdhealth, an alternative to US health insurance, Kindred, a home sharing app, and Nat Eliason’s Build Your own AI Apps course.

If you received this and find yourself in a state of outrage thinking, “How the hell is this drivel allowed in my inbox?” I strongly encourage you to unsubscribe below. It is your god given right and there is plenty of good stuff online.

A reminder: I don’t check unsubscribe alerts and never look at my subscriber list. So if you feel like unsubscribing, you can do so below.