MeshThink

This article from the New Yorker is a must-read for anyone in tech. The author believes that we’re at the end of software development as a high-value career path. I see the point, but I don’t think it’s quite on the mark. Of course, AI is absolutely poised to change software development. Just ask Stack Overflow. But I don’t see it ending the role of strong developer. You see, it’s really good at writing a certain kind of code…code that’s been written before.

Now, for a typical digital product, ‘code that’s been written before’ might be 95% of the code base. Maybe even 99%! Sure, you’re fitting the blocks together to build a new and interesting final product, but the building blocks themselves? Mostly same ‘ol same ‘ol – new languages, different platforms, but the same logic, the same algorithms.

And here’s the worst part – as a developer, ‘code that’s been written before’ is the least interesting part of the job. Spending a couple of days chasing a bug in a method you wrote that’s been done thousands of times by thousands of other devs? It’s not engaging – it’s tedious. Doubly so if it’s been written before by me – now I have to write it again, debug it again, just in a new language? Give me a challenge – something I really have to puzzle out how to build!

As I’ve worked with the emerging generation of AI-powered coding tools, it’s been a fresh new experience – because of how little time is spent on the tedious ‘code that’s been written before’ parts. I can jump past that and get to the interesting stuff – the differentiating code.

How does this change the development process? It’s an accelerator – developers are able to get all of the basic blocks in place faster. Instead of spending 95% of their time on commodity code and 5% of their time on differentiating code, they can flip it in the other direction.

This has a big impact on product strategy. As the coding tools become better, you’re able to test out the differentiating idea much faster, because the commodity pieces can all be generated from AI assistants. And when it comes to product roadmaps, AI tools should be a huge accelerant – because the Big Ideas almost always need a lot of foundational blocks to prop them up. Now, your development staff can spend most of their energy on the thing that’s most important to your product strategy, and speed-run the rest. This might mean less development staff is needed in some cases, but I suspect the smart companies will use this to double down on investment in the thing that sets them apart.

In product strategy, we often call this the ‘moat’ that defends your product vs. competition. If the thing that makes your product special is a really tough technical problem to solve, it won’t be an easy target for others to copy. AI assistants won’t change that – competitors will still need top developers to be able to replicate your differentiator, because it’s *not* code that’s been written many times before. This is why I think software development jobs won’t go away – smart companies will recognize that the amount of moat-digging they get per developer has skyrocketed, and will put their company in a much more defensible position.

But this is also a critical time to take a long, hard look at your moat. Historically, some products have dominated their market with moats that are built on cumulative effort, not difficult effort. The differentiator isn’t uniqueness, it’s critical mass. Some examples:

“We have out-of-the-box adapters for 300 different partners.”

“We’re a one-stop shop – you can replace these 3 platforms and just use our consolidated solution.”

“We’re the most feature-rich platform – everyone can do some of what we do, but no-one can do all of what we do.”

I think of these kinds of moats as wide and shallow. No one step is particularly hard to take – you’ve just got a looooong way to go to get across the moat. Long enough that it hasn’t been feasible for most startups to try to cross. AI provides planks that make the shallow parts of the moats an easy stroll.

So, if you’re a market leader with a wide, shallow moat – run, do not walk, to your engineering team and figure out how to refocus your roadmap on technically difficult differentiators. If you are a startup looking for a market to disrupt, find one of these players with a wide, shallow moat and set your team directly on a hard-to-develop differentiator. Those players are much more vulnerable to being overtaken by a small team than they were a year ago, so get your AI stilts on and start wading across. 

And, whatever you do, don’t build a moat with AI code generation. Because, if you can, it’s not actually a moat at all.