This was in my head. I just wanted to write it.
So many layoffs. Engineers and engineering adjacent roles being rm -rf’d faster than you could imagine it could happen. Square, 4000, LinkedIn 600, Cloudflare 1000 and yesterday Facebook did 8000. Sympathies to anyone affected, it sucks. These are big numbers in the sense of companies but tiny numbers in the context of the global workforce. But where do they all go?
Humans
Some no doubt will be re-hired. Some may change industries, but the majority will likely remain in roles related to software. If you have to keep the lights on and put food on the table, that’s what you do – you crack on. But where to crack on? I’ll get to that, but lets go back up a level to the companies doing the “resizing”.
Borganisms
I don’t think for one second companies are done getting rid of people. Why would they? “we must squeeze every ounce out of the workforce” proclaims every board meeting. Harsh sure, but business is business. Dismantling that one is a topic for another day, but stay with me. Sure some will re-hire some engineers back, but I think the days of global software brands are here. Google, Shopify, Amazon, Xero and some smaller ones like Remote are all versed in the art of acquisition and it always made sense. Buy up the talent / IP in a given space and make it yours. But then there’s the issue of the messy humans and their pesky salary. Oh no.
CI/CD of the future
There’s a problem in the slopiverse. Too much god damn code is being written and we cannot review it. At least that’s the complaint. Personally think you either trust the code that AI writes, or you don’t. Before you jump down my throat this is not a carté blanché to send it out into production without any quality control. But I think the days of a human with a title being the ultimate adjudicator are coming to an end.
Yes, pipelines do catch predictable failures. Human reviewers are great at catching situations that do pass the pipelines, but will fail in production. So whilst human review is certainly a thing for now, as the models get better I can certainly see a future in which model capability meets human capability. That’s when human review ceases to be a thing.
When we get there, we’ll be at a point where CI pipelines are sophisticated enough to be able to bring “compliance compatible”, “auditable confidence” to allow code to be entirely assessed by AI. It’ll probably be expensive. Right now Claude Security isn’t available to max customers but I’d be curious to see what happens when it is. Maybe it’s similar to Code Rabbit, who knows…
At least we’ll be able to work at a scale that could match the firehose of slop.
Factories
Imagine a factory.
This factory doesn’t produce shoes, food or hardware. It produces software. To a spec. There’s no large warehouse. There’s no office. Just a team of five people remotely working for their co-operatively owned software factory.
Tech Brands
Now imagine a future where large software companies are nothing more than data centres and products. They stopped writing software and don’t have engineers on staff anymore than nike has cobblers or seamstresses on staff. Like Nike, this big tech company is basically a brand. It buys in all of their code.
Specs are issued, vendors are selected and code is produced. It is shipped from the factory to the vendor in a container and ingested into the tech brands infrastructure through the most heavy duty security and feature analysis the world has seem. It makes Mythos look like a JavaScript version of Eliza.
The ingestion approves the image, the containers are loaded and moved into production where they are continuously monitored against microsecond precision SLA agreements.
Software factories
Some of it feels awesome. The opportunity to win big contracts for teams that previously had no chance of doing so feels like a massive win. What happens with that wealth is another topic. Imagine winning the contract to supply the algorithm that handled captchas because you could demonstrate your factory’s solution issued 70% less CPU instructions and therefore reduced energy usage by 80% of the baseline specified in the tender? Does anyone even engineer for this right now? Another topic for another day.
Some of it feels terrifying. Imagine working for pennies on a production line being nothing more than the meat based carbon life form that gets fired if when the code they approve code - later malfunctions. No thanks, I don’t want that future even if does come with really nice grey uniforms. Nope. Do not want.
Most of it feels like it will never happen. But that’s the thing with AI, it has affected the way we operate. Not in totality, but certainly in parts. Sometimes all it takes is someone to move first and then everyone copies. Few people remember Vine, but it was the OG platform that gave the world the attention sapping technique of short clips. That technique metastasised into the format of choice for the delightfully-engaging-yet-ultimately-brainrotting-experience that is TikTok, YouTube shorts and Snapchat. It even morphed instagram from a once large network of excellent deeply human photographs, into something else entirely.
What part of what tech will be the first to test the waters and where will it go?
Brand Moat
But it may go that way. Acquisition is expensive, due diligence takes time. Attracting and maintaining talent is expensive. All it takes is one vendor to think “We’re the brand, we have the compute power and access to <insert Mythos like model>. Why don’t we just buy the tricky bits in?” Any shoe factory could make Nikes but they don’t because of trademark, copyright, contracts and the definate scenario of being sued into oblivion.
Does the algorithm underneath Spotify’s random shuffle have to be exclusively by Spotify? No, but the audio must be served through Spotify1. What about the compression alogrithms? Who even notices that? What if they wanted to reduce their network costs and that became a key factor in the tender. I think you can see my point.
The brand for tech many becomes the moat in the same that the brand for Nike is. In some senses we’re already there. Lots of people didn’t want to be associated with Twitter when Musk took it over and those people moved over to Bluesky. Voting with your dollars in tech is just as much a thing as it is in fast moving consumer goods (FMCG) or vehicles to give a few examples.
Wrapping up
People affected by AI layoffs are going to continue working. The talent has to go work somewhere right? Hiring is a nightmare (another topic for another day), too few jobs and too many engineers.
Not everyone wants to be an indie studio or make lifestyle SaaS or mobile apps. Talent attracts and teams are powerful things. Who knows what will happen when these people get together? I don’t and probably none of this will happen.
But it was a thought that felt novel enough to publish after a lovely cup of tea with my friend.
I appreciate your attention.
Cheers,
Jamie.
-
Spotify - please fix your shuffle mode. For some reason when it has hundreds of tracks to choose from, it always seems to select the same ones. More so with certain genres. ↩