I’m pretty convinced that the more the software engineering industry adopts LLMs for coding, the more specialists, not generalists, will benefit from this adoption.
This seems counterintuitive. About 12-18 months ago, I started seeing opinions like “with enough LLM assistance, any generalist can now contribute to any part of the system”. So, it seemed like highly specialized roles would disappear.
Nowadays, the dialogue has shifted from the generalist/specialst debate to whether software engineers need to review LLM-assisted code at all (or whether they’re needed in the first place).
However, I’d like to return to the specialist question, because it’s extremely relevant in the context of this newsletter. I do believe that becoming a specialist is the easiest path to solopreneurship, so will AI make specialists obsolete?
I don’t see this happening in practice. In fact, I see more demand for my services as a specialist, and I see more success stories of software engineers becoming solopreneurs.
Here’s a recent anecdote. I was on a call with a potential client a few months ago. They wanted me to take a small project to audit a part of their infrastructure after a migration. The client said, I quote:
Yes, I checked it with AI. But, you know…
NOTE: this is not a post about the success of AI adoption. This post is not pro or against AI. I simply try to claim that given some AI adoption, specialists will still be in demand.
The Fundamental Flaw
The fundamental flaw of LLMs (and their greatest strength) is their ability to generate plausible content.
LLMs can generate code that compiles, passes tests, and looks reasonable. LLMs can then convince you that it’s the best possible implementation. And yet, the implementation can be overly complex1 or suffer from performance and security issues. Or it can be completely unnecessary - LLMs happily design and implement any feature you ask for without questioning the need in the first place. I’ve seen it over and over again. You can add any number of plugins, skills and MCP servers, but LLMs will rarely push back on implementing something in the first place: they want to use the tokens! They were designed for that.
The most dangerous thing about this situation is not just the occasional merging of questionable code into the codebase. It’s the unfortunate compounding effect.
Of course, many engineers and many companies can be ok with that. Some call it “adding levels of abstraction”. Some (in fact, most startups) happily trade short-term velocity for long-term success.
The Specialists
I see two reasons why specialists are still very much needed in the LLM-accelerated world:
Rapidly generated code is like a ship that accelerates all the time. But will it arrive at the destination? You need a set of specialists who use their expertise to steer the ship. You still need judgement and critical thinking; you need to know best practices and standards, and you need to understand the business context.
You also need to ensure that whatever is built remains adequate. Engineers or LLMs follow reasonable approaches. Software design is not forgotten. Performance and security are still respected. Modularity and reusability sound boring, but they’re still very much needed.
Here’s a great post from Jonathan Stark:
“Do you actually know what your clients are buying from you?”
Lots of people don’t realize that what they think they are selling isn’t what their clients are actually buying.
If you think you sell code or features or white papers or logos or ad creative or some other artifact of your craft, then I can understand why you’d be worried about AI taking your job.
But your clients aren’t buying artifacts.
They’re buying expertise.
Things like trust, judgment, taste, risk reduction, peace of mind, connection, understanding, etc.
They can’t get any of that from AI.
In fact, I believe that the value of specialists will grow with time if the current wave of LLM models keeps improving: 1) there will be more generalists, because LLMs discourage deep specialization 2) existing specialists will retire… How do you get the new specialists? The law of supply and demand tells me that the value of specialists will go up.
It’s Still an Amazing Tool
AI in the hands of a specialist is much more valuable than AI in the hands of a generalist. You can move faster AND feel comfortable with the progress.
Here’s my recent experience. Take a look at the comments!