28
votes
How has your industry changed in the past decade?
The other day I had to get new glasses, and I braced myself for my lenses to be incredibly thick and expensive to boot again - but then I had them made, they look normal, and they barely cost me a Benjamin. Clearly, the optometrist crowd has made some major developments in the past decade or so, which leads me to ask - if you're working in an industry most people don't really think about, what's happened in your space in the past ten years?
I'm a developer so... enshittification.
Products used to do something useful for users. That has become increasingly rare.
I don't socialise on the internet any more, I pay subscriptions and consume 'recommended' content. Nobody I know posts anything to their personal social media pages any more because it's inevitably lost in the sea of ads and influencer jockeying.
A few niche services have gotten a lot better.
I had to return something on Amazon earlier this year and their chatbot was amazing.
An online game Ive been playing for about 20 years now has really found their stride, and the updates for the past ten years have all been amazing. Anyone who plays Oldschool RuneScape knows what I’m talking about.
I do love Discord. Its a significant improvement over AIM, IRC, Ventrillo, Skype, etc etc. Its not a replacement for message boards and I really wish people would stop using it for that, but it is an amazing messaging/video call app and I love that it works so well in a Firefox web browser.
I work in international security, and my last decade has felt like the famous Lin Wells memo.
In 2014, Counter Insurgency (COIN) was still being heavily pushed as a legitimate solution to Europe and America's greatest perceived threat - ideologically motivated violence by groups occupying what were then termed 'ungoverned spaces'. The entire national security apparatus was focused on the Middle East and stopping violent extremism, particularly in Europe.
Ten years later, a major land war has broken out in Europe and conventional, large-scale combined arms warfare is back, which we were told could 'never happen'. Despite this war being ongoing, global attention turned to the Middle East again, where (by the definitions of International Humanitarian Law, as they were taught to me), a genocide is being comitted. But even that doesn't hold the world's attention, as Christian nationalism and fascism has overtaken the world's only military superpower.
All of that is being exascerbated by climate change, divided societies, and above all information operations by major state actors. If you couple those information operations with the rise of unregulated algorithmically driven social media platforms becoming the new gold rush, I'd say the major shift in the past decade has been fractured civil society.
The operators of these platforms act as transnational feudal lords, able to shape the polities of any country whilst providing very little back to the concept of the nation-state in the form of taxation from their wealth.
Disinformation and misinformation are opportunities for profit or gain, and the concept of a global community has eroded. The UN is as weak as I've ever seen it, whereas in 2014 it was still an effective body for dispute resolution, peace keeping operations, and attempts to address truly global scale security concerns - food insecurity, the Sustainable Development Goals, and climate change.
To put it more simply, in 2014, even a minor terrorist attack would be a stop-everything emergency, requiring presidential addresses, and civil/military response. In 2024, ICBMs are flying in Europe, tens of thousands of people have been killed by the United States's closest ally, and if there is a presidential address, nobody listens or cares.
We're at major risk of a global military conflagration, we're sleepwalking into a climate nightmare, but there is no agreed 'we' to even begin discussing solutions.
Slack has become the prevailing method of IM for my team, more documents are saved and shared via cloud storage instead of a centralized document folder, so they can also be worked on in real time remotely, everything is digital now (thankfully), more of my team is fully remote, so office management of the team needs is very different, but still somehow the same. Everyone needs the same stuff, just at home instead of in the office.
Oh yes, cloud file sharing. I definitely do not miss the days of trying to manage a shared Samba drive across two continents over VPN with crappy rsync scripts.
At one of my first jobs we used IRC for internal communication while everyone was still on email. Good simple times. Which is what made the company that created Slack pivot in the first place.
I don't miss the days of working on events or marketing materials with my sister office and having to literally drive up to SF from the south bay because we only had desktop computers and having to have a meeting scheduled to work on some Illustrator or InDesign document together because we needed to make sure it was compliant with the brand package.
I'm in and around emergency medicine. The explosion of mental health emergencies has absolutely devastated much of emergency medicine. I started in the field as a firefighter/paramedic responding to 911 calls back in 2010. Calls for suicidal ideations/attempts were there, but not anywhere near the frequency I was responding to those calls when I left in 2019. I left to go work in a pediatric emergency department, when I started in 2019, we had 3 "lockdown rooms" situated for patients experiencing such emergencies. Situated so that the room could be "locked down" and remove access to any practical way to self harm. Since then, the hospital had to build an entirely separate behavior unit with 7 new lock down rooms. They are typically all occupied, and patients experiencing these behavioral health emergencies often need to be overflowed to 1 of the 3 old lock down rooms. Room space aside, these patients are often ranked very low in priority as well. Not that suicidal ideations aren't an emergency, they are, but their vitals signs are stable and the person down the hall has decompensating vitals and failed their Bi-PAP trial and is about to need intubation and vasopressors. The need for dedicated mental health resources, providers, facilities, is there. But the overlap of "we can't treat the mental health emergency til we know the physical health is stable" has led to a culture of "screen 'em in the ED, however long that takes, then ship 'em to a psych facility", and it feels very bureaucratic and inefficient, not to mention minimally helpful (if at all), to me.
Been working in IT for around two decades now, and some things are the same. Like I still use vim and all the basic Unix tool skills I picked still applies. SQL is still SQL and so forth. Of course the whole containerization and Kubernetes is radically different than just ssh into a server and manually do stuff. Which is for the better.
On the less technical side, I feel there is definitely a time before and after GDPR. Before data was just data, and there wasn't much thought into whether anyone kept more data than they needed. Not just internally, but also customers expected we sort of always had their data and that it was easily accessible. Now we have to deal with a whole lot of compliance work, to justify what we keep in our logs, what we have in our database, for how long, who has access, logs of who accesses and for what purpose and so forth. This is a great thing to be forced to handle and good for consumers in the end, but it also comes with a whole lot of frustrations when people are interpreting laws very differently and lawyers without technical knowledge who has absurd demands almost forbidding to keep any sort of backups or system logs, while still demanding we process their data.
Ooo GDPR is a good one, which I definitely echo. I'll add onto that that with GDPR came a heavier focus on data security in general - probably because GDPR forced companies to hire actual data security consultants. Suddenly there's courses on preventing social engineering we have to pass, and the guillotine for those who leave their desk with their computer unlocked! Not complaining mind you - I used to be the only data security 'nag', but now suddenly every other day an email went out from up top scolding everyone about the shit I'd been pointing out for years, LOL
I’m a developer so…. AI.
I don’t Google things any more I ask ChatGPT.
Can you share more about how you effectively use genAI? I've tried time and again to use it and without fail it's incorrect to the point of being a waste of my time. In fairness, I can't use ChatGPT specifically due to corporate policy, but can use other state-of-the-art models.
I think part of my problem is with the code I work on: the context is just too large, with too many internal libraries. But even asking questions about well-known libraries in widely-used languages (the interaction of GSON and Lombok being my latest attempt) has yielded results that are useless at best, and often have made the problem worse.
Not OP, We use a combination of things here but MS CoPilot seems to have improved a lot over the last 6 months. It's still not the same as ChatGPT, but we are getting a lot of great use cases set up with it.
Don't try to get it to do anything smart with your entire code base is my advice. I don't trust current LLMs to write code directly for me. Except maybe for one-off throw away scripts I need to use once.
I basically use LLMs as rubber duckies who actually talk back. I explain issues, get a response and sometimes the response is really useful and other times it is crap, but through the process I have gained new insights. I also like to play different models out against each other if I need ideas on how to approach a specific code issue. I ask them both for input and then ask them to review the response of each other. Works pretty well!
Basically, I mostly use them as tools to get through various tedious tasks. Helping me go through obscure errors, stack traces, decipher spaghetti code someone else wrote and a lot of the tedious Google searches.
The benefit of a chat style LLM is that you can follow up on it and ask it questions. I don't expect it to be perfect, but it gives me a jumping off point for a lot of things that previously would have cost me more time.
To give a few slightly more concrete examples.
Again, it isn't perfect. In fact, I can only use it like this because I have knowledge and experience myself that allows me to ask the right questions and validate answers.
I'm a developer who recently became a manager. I have been using ChatGPT to help me write job postings. It's all stuff that I know enough about to check for accuracy. I just hate writing that style of document, and ChatGPT does a really good job of producing the first draft that I can then polish.
I use it for things I know I can Google.
GitHub worklfow documentation
It didnt have the up to date documentation (a feature was added) so I gave it the link and then it was able to answer. The most complex part about it was creating a timestamp in one step to reuse in multiple other steps, I had to get it into a specific format. That bit would have taken me a whole day by itself if I had to Google it and read through linux documentation.
I needed to choose a text similarity algorithm, so I asked it what are the best python libraries for my use case and I talked to it for like 30 mins to narrow it down one.
I use it to generate text samples for test data all the time. Stuff like “write me a customer review thats 180 characters” …. “Shorten that to 100 characters”
Not OP, but I use it to look up simple things. How can I refactor this basic algorithm more cleanly? Why isn’t this laying out as I expect? How can I return an array dropping the first index. Things like that size. I will rewrite the code as a simpler generic example before giving it to ChatGPT.
I have also started using Xcode’s ai code completion. It’s kind of ok, but kind of bumpy too. Want to try code pilot.
Not to be dismissive of your specialization, but LLMs are a lot more useful when there’s a wealth of relevant training data — and (anecdotally) I’ve never heard of GSON or Lombok, so I presume those are just too obscure to have meaningful coverage. I work with JavaScript and React, which are embarrassingly common these days but the upside of that popularity is that AI is really good at answering questions about those. I hit more issues when I ask for help with TypeScript, and it’s totally unreliable when I need help with specific JS libraries, even ubiquitous ones. By this point I can usually intuit what subjects are safe to ask about, and which are likely to end in nonsense hallucinations.
One surprising use for AI I’ve been leaning into more is Copilot’s terminal integration (in VS Code, maybe available in other ways too?). If there is a specific command or args I’m looking for but don’t know off the top of my head, I just type a couple words explaining what I want and it’ll put the right thing in. Previously I would’ve had to pull up a man page or google around; this has been an unexpected timesaver and for my needs it’s been quite reliable.
In case you ever move to Java later, Lombok takes care of the getter/setter boilerplate. It looks like gson is a json serializer, not the one I used to use when I was in Java land.
You not knowing GSON an Lombok is simply down to the ecosystem as anyone working with Java will very likely know both. I also haven't had real issues with LLMs and java as there is a plethora of data out there they are trained on.
I've found the question / answer pattern of chat bots to be really unhelpful, it doesn't seem to line up well with the way I think or work.
What I've found far more useful are the AI autocompletion tools like Codeium in vsCode. I admit that at first I found them more annoying than helpful, but I started just letting them do their thing during an RSI flare-up and it drastically cut down on the amount of typing I needed to do. I still need to go in and fix things, but that involves a lot less typing. I think it has sped me up, in a similar way to how snippets speed me up.
I'm in the Credit Union industry, and technology vendors have become quite the "fire hose" over the last few years. I'm sure it's the same in many industries but AI is in literally everything we are pitched. Even when it isn't necessary. The problem we are seeing is there is way too much overlap in what vendors can do for us, and the costs have skyrocketed because of the "AI".
I lead the institution in selecting vendors mainly because of my mid ground knowledge of AI, it has been busy for sure. We also have to wade through the good partners in the sea of technology to find the one that truly fits ours needs.
There are some groups trying to help our industry find the correct partners but I feel like there is a bit of lobbying and money passed around for those recommendations.
Looking at your comment and @jackson's comment above, genAI is creating both time and financial costs for workers and companies. What's fascinating to me is that some users insist that, when wielded properly, genAI can save tremendous amounts of time and mental energy. Others find that it requires so much prompting and checking over copious output that it doesn't actually save them any effort.
I wonder where the discrepancy in perception of genAI's usefulness lies. Is it truly that some people just aren't putting the time and effort into effectively using genAI, or are they doing different kinds of work, or are they inaccurately estimating the time/energy expenditures with/without it, or something else?
For myself, I prefer to work on a problem myself than supervise a computer that pretends to think. Prompting output is so much worse than producing output. I don't want to deny that some people are having positive experiences with such tools. At the same time, I see high costs and low benefits, resulting in negative value overall.
I’m convinced everyone who finds it cumbersome is asking too much of it.
Before, when Id google things, its 1. Type the question 2. Look through the search results 3. Pick a search result that seems promising and click on it. 4. Wait for that page and all its ads to load depending on what machine I’m using and then sometimes 6,7,8 rinse and repeat cause all these search results suck or I’m asking the wrong question.
With chat gpt, I ask a question, I get an answer. If the answer makes no sense, I’m asking the wrong question and I need to rethink how I’m going about the problem. The whole turn around time is down from 5-10 mins to like 1-2 mins. Its a huge life improvement.
I fully expect chat gpt will one day introduce ads and all the nonsense that enshittified Google results, but today its still amazing.
That might be. I usually find the answers to be some combination of wrong, fictional, and overly generic. Also, while there are some heuristics for the reliability of a webpage (domain authority, amount of ads/popups, formatting), all of that is stripped away in genAI. Even tools like Perplexity that cite their sources are not trustworthy on their own, so (in my case) I'm back to reading through source material to determine accuracy and reliability. For finding answers to questions, I find it god-awful.
Maybe my questions are just more simple?
I tend to ask it things like syntax, best way to do something in blah language, generally things that don’t need a lot of context.
I wouldn’t google how to use some internal tool that my company has, or why a previous employee chose one data structure over another or what a hard coded error message could possibly mean.
I've been bit hard even doing so. Gemeni told me to use a non-existent Android property and it was acting so confidently, and I ended up wasting a good hour trying to find it kn docs till I realized it doesn't exist.
This means I can't trust an LLMs output at all: 90% accuracy is as good as 0% accuracy, and if I'll end up reading through sources anyway, I might as well cut the middleman and google those sources directly.
The problem is sometimes you don’t even know what to google. LLMs give you a starting point.
I find them particularly useful for things like finding the idiomatic way to do things in a language, because that’s very hard to google.
I actually do agree with you here. LLMs are a great calculator for languages, good for searching for a word I don't remember and fuzzy search in general.
You describe the exact problem I have with it - Trust.
If I'm asking it about a topic I don't know the answer to, I would be foolish to trust any output because it might be confidently incorrect. There is real value being able to trust answers first time and I don't see how LLMs can achieve that.
A reasonable middle ground might be to flag an answer with the text "This output may be inaccurate due to insufficient training data" or something, but god forbid these companies admit the limitations of current tech.
Oh, and your point about putting ads in AI answers? It's already happening :)
https://www.androidauthority.com/google-ai-overviews-ads-mobile-3500919/
I'll mention one thing among many:
The average web developer seems much more tolerant of friction now. Cloud deployment, devops configuration, forests of dependency packages, supply chain injection, app bundling, SPAs, microfrontends (hello again, iframes), test suites too big to run locally, flaky tests, frameworks upon libraries upon other libraries, linting just to avoid talking about linting, containers, containers needing other containers, containers for local dev not fitting in local dev, PR actions, merge checks, pre-commit checks, commit text linting, secret management, force pushes...
I miss the days when I could just Save in my editor, then see my change in my browser.
This is much of what drove me from Java/J2EE to Ruby/Rails then to management then to unemployment and now try to start my own (physical) product business but not expecting much but hoping hard. That product isn't about getting rich but trying to do some actual small good in the world and just maybe paying my bills if I'm really really lucky.
I'm so disillusioned with the tech industry. We haven't just enshittified the customer experience but also our own daily lives. At some places the pay is terrific—you just have to trade your soul to get it. The industry is inhuman; we have all been cogs in a machine that pretended to have a heart. That heart has proven out to be an automaton for converting people into capital for billionaires.
It all seems a little crazy to me too. All that fuss when some large percentage of things deployed could function just fine as a Rails app behind nginx+unicorn (and maybe varnish if load is high enough that cache helps) on a cheap VPS box. No it’s not sexy or trendy but I it also gives me a reasonable chance of being able to understand and hack on everything I’m deploying.
Of course uploading PHP files over FTP is even simpler but I personally find the structure offered by something like Rails to be appealing, even if I don’t want to complicate it too much further.
We're not allergic to complexity like we should be. Simplicity can mean less functionality, sure, but it also means more resilient, easier to learn and fix. Unfortunately, designing for simplicity is hard. It's less work to just keep adding stuff and the downsides only come home to roost years later.
I work as a ‘creator’ I guess we’re called (I consider myself more as an artist for a mainly online audience), and in the eight years I’ve been doing this, there has been a tremendous power struggle between creators - social media companies - tech/payment companies which has been very interesting to be a part of. AI has only intensified that struggle, because some social media companies are betting on generating their content without any need for humans/creators.
For us, we’ve always focused on owning our own platform and mainly use external platforms as advertising outlets for our work, which has served us well. Creators that depend on external platforms for revenue and connection with their audience our struggling + it has become a lot harder to convince people to sign up for an external mailing list. I think that’s only going to get worse.
I’m wondering if those more private social media apps, like WhatsApp with group chats and communities, will ever generate the amount of ‘free publicity’ you need as a creator to get a working business of the ground.
Interesting years ahead!
I’m a mobile app dev, so:
Back then for company infrastructure you'd rent a colo and rack a couple of servers for your business to get a file share, email or install a business app. Your side of the business was essentially scoped to a particular country. Now I can explore public API documentation, get an API key, write some code and get something provisioned in minutes that teams or entire departments in other countries can start using tomorrow or in a few hours.
Want a marketing system, let's see what Hubspot can do. Need monitoring for our SaaS app? Sentry and Elastic looks good over there. Hiring a bit of a mess? Let's try out Greenhouse.
We run the company completely online connecting a bunch of disparate services together with a level of automation that can be done between them thanks to the ubiquity of APIs and excellent open source software. We're in a position where we are not totally beholden to Windows and Office (except maybe the salespeople), so we run a lot of Mac and Linux workstations, Slack/Zoom/Google Workspace for collaboration and a lot of open source software to support the rest of the infrastructure on Google Cloud for cheap. Our production Kubernetes clusters auto-upgrade, which would give fits in a more traditional setup which I'd argue forces us and the dev teams to build resiliency and scaling properly from the start.
Our offices across multiple countries are essentially closets with a few desks and an access point on the rare occasion someone wants to come in.
I’m a brewer and all my special/seasonals have become IPAs. Only thing that sells consistently and that the most vocal customers want. Sure i could make something else but then it sits eating fridge space. Make an IPA and it’s gone.