• Home
  • Blog
  • Projects
  • Bookshelf
  • About
← Back to all posts

The Heat Death of Software

Published ....
Last modified ....
From Stamford, CT

Share this post on BlueskySee discussion on Bluesky


I think we’re heading for a dramatic change to how software gets consumed in the next 5-10 years. Before diving into my prediction, lets recap a bit about how most software was distributed 15 years ago, and how it is today.

15+ years ago, most software was still largely distributed artifacts that you’d install into your computer and used on your computer. These applications were largely built by fairly large software engineering teams, working in collaboration to define the product surface area, the set of features, and how the application functioned and was expected to be used.

Today, most of that remains largely the same. The primary change over the past 15 years has been that most software is now primarily distributed on the web.

  • Google Workspace has largely displaced the Microsoft Office suite of tools
  • Figma has largely displaced Sketch/Adobe Illustrator

This is similar to the shift in the gaming industry as effectively every mainstream title and publisher shifted to the live service model. Software now is treated as a live service as well, instead of scheduled quarterly annual releases teams can ship changes in minutes.

Now we can get to my prediction and what I think will effectively become the heat death of software.

Thanks to LLMs, AI, and Agents - we’re seeing a dramatic change in software engineering, specifically around:

  • Open Source Software
  • Software Teams
  • Vibe Coding

Open Source Software

This is one of the areas of software engineering that I think is currently and will continue to see the largest changes thanks to agents. For the past number of years, most leveraged common popular open source libraries for a variety of problem spaces. Largely driven by the “DRY Dogma” (Don’t Repeat Yourself), most teams pulled in those popular libraries instead of re-writing it in-house.

products
services

Now with agents, the cost of maintaining that external dependency can be replaced with a few bucks and about 30 minutes of token generation with your favorite model and harness.

Why add a dependency on a left-pad package when your agent can write the three lines of code implementing it within your codebase directly?!

Now a fairly large chunk of software engineers that were sold a fulfilling career of maintaining open source code are being flooded with contributions or bug reports from agents, while their install rates are decreasing by the minute. These engineers are being pinched - some projects are starting to prevent folks from reporting issues or even opening pull requests to somehow keep their maintenance burdens relatively low, and the group of folks and teams that would either support development (by contributing code or money) is further dwindling because they no longer need to depend on the open source version of the code.

I think within 5 years the concept of an open source library/package will mostly be looked at like an antique, something that used to be prevalent but will now be seen as quaint and only appreciated by a select few.

Software Teams

By in large, most software since the industrialization of computers has been built and maintained by teams of people. Theres been notable exceptions to this, especially when looking at open source software, however almost every bit of software you use daily is maintained by a team now.

Teams were effectively the only tool that companies had to either speed up development or increase features and capabilities in their products.

Now with agents, we’ve started to see this change pretty dramatically similar to open source software. What used to be a few senior engineers and several junior engineers can now be roughly replicated by one or two senior engineers and a fleet of ephemeral agents.

The beauty of this for upper management is that you can provide roughly the same level of service as before (or better) with a far smaller team.

The dramatic trade off that I think the industry isn’t fully acknowledging and accounting for is that we’re completely destroying the only known way to train senior+ engineers (giving junior engineers time, capacity, and opportunities to fail).

I think within 5-10 years we’ll find that most software engineering teams will be extremely minimal, and we’ll additionally have a full blown crisis and decline of talented and experienced senior engineers.

Vibe Coding

If the above two problems weren’t large and daunting enough for you, I think this last one will at least plant a seedling of doubt about the software industry at large.

Most software today isn’t really free, the industry has grown so much because it has proven to be fairly profitable (no matter what the stock market says on any given day). Good products and services can be sold for quite high margins, and because of the shift from installed/local products to live services offered over the internet - the industry has moved from one-time purchases to monthly/yearly subscriptions.

However, the agents have gotten quite good at building software. They’ve gotten so good that now anyone can effectively clone another product or service with relative ease. These clones are still what I’d call vibe coded projects, meaning they are able to look and feel like the real deal, but often may take some or many shortcuts to get there. They’re essentially really good working demos, good enough demos that they can often be useful for the general individual.

If the models and agents in general continue to improve at the rate that they have in the past year+ - then I think we won’t be too far off from folks being able to clone (or create from scratch) whatever product or service they want.

When we hit that point, why would someone choose to pay a monthly fee for their favorite product (that one could argue has maybe stopped receiving valuable updates, or has gained hundreds of features that they don’t care about) when they can pay a one time cost of a few bucks worth of tokens and get mostly the same experience?

Now you start to scale that out to anyone and you can start to extrapolate what might happen to the software industry as a whole.

Counter Arguments

There’s a few key counter arguments that I can come up with to some of my predictions above;

Maintenance is often far more costly (both in terms of time and actual costs) than building out new products or features. Currently all those vibe coded products discussed above probably don’t have a long term maintenance plan or aren’t set up for one either. Especially when thinking about some of the products/services that offer fast and efficient vibe coding platforms like Loveable and v0, etc, and the customers that those services market towards often don’t know how to prepare for long term maintenance of those vibed products either.

I think there’s also something to be said about durable services/product offerings, maybe there’s a better way to describe this suite of products and services but at the moment I think of companies like Cloudflare or AWS - where they offer a suite of core services and offerings that currently humans and in the future agents build on top of.

Meme in the format of "you wouldn't download a car" saying "you wouldn't vibe code S3"


Tags:

AIpredictionsSoftware DevelopmentAgentic Engineering

Bluesky Post and Comments:

Loading comments...

Loading...

Related Posts

AI

Your Job Is To Understand Systems

....

Parallel Computers

....

The Harness Isn't Your Problem, The Router Is

....

OpenBlank

....

ASD-STE100 Simplified Technical English

....

Performative Engineering

....

The act of making noise and delivering output instead of outcomes

The Agentic Engineering Scale

....

A brief outline of how I think of the scale of adoption of AI in software development

Onboarding Your New AI Teammate

....

Are we reinventing the wheel when it comes to onboarding new AI agents to a codebase, when we already have primitives available for onboarding humans?

Cursor and Obsidian

....

A quick look at combining Cursor with Obsidian to ask questions of your notes!

The AI Development Conundrum

....

Is it a good or a bad thing to offload writing code to AI agents and Large Language Models?

Offload Complexity, Don't Offload Learning

....

A rough mental model for how you should be leveraging AI as a tool for your own growth

My Current AI Stack

....

I've been using a variety of AI tools as of late, I figured I'd document the ones I'm primarily using!

Software Development

Your Job Is To Understand Systems

....

Parallel Computers

....

The Harness Isn't Your Problem, The Router Is

....

Agentic Engineering

Your Job Is To Understand Systems

....

Parallel Computers

....

The Agentic Engineering Scale

....

A brief outline of how I think of the scale of adoption of AI in software development

Performative Engineering

....

The act of making noise and delivering output instead of outcomes

The Agentic Engineering Scale

....

A brief outline of how I think of the scale of adoption of AI in software development

Onboarding Your New AI Teammate

....

Are we reinventing the wheel when it comes to onboarding new AI agents to a codebase, when we already have primitives available for onboarding humans?

Polish is Important

....

You're Building Software Wrong

....

Slicing software: why vertical is better than horizontal.

Matt Hamlin

Matt Hamlin

@matthamlin.me

New blog post: matthamlin.me/2026/septemb...
The Heat Death of Software

The Heat Death of Software

A few predictions about where software engineering is going, including some counter arguments as well

https://matthamlin.me/2026/september/the-heat-death-of-software
1 hour agoView on bsky.app
0
0
0
0