AI Dependency: The First Fix Was Free — and Now the Bill Is Coming Due
Those of you who've been reading Kludder for a while know that accountability matters to me. Every post is a colour palette of links and source references. That was a choice I made early on.
In an age where everyone has access to a megaphone, where conflict drives attention and bold claims seem to be the only claims, I haven't quite given up hope.
Hope that we'll find our way back to facts one day. That I can be "caught out" the day I go off on an undocumented rant in pursuit of higher readership. Kludder is meant to be that little break room that gives you a clear view of the murky. But it's also supposed to be the place that can nudge you in the right direction, should you want to make up your own mind.
This week's Kludder is going to be different. It might even feel like a blog post! Because today I'm writing based on my own thoughts for the future. I'm writing in an attempt to understand what awaits us around the next corner, when it comes to artificial intelligence.
The day AI wrote a text message for me
Claude, ChatGPT, Copilot, and Gemini have become everyday words in offices around the world. I personally no longer go a day without using a language model.
And yet there's a voice in the back of my head. The one that's terrified I'll become so lazy I stop thinking for myself. The voice that fears for my love of writing. That maybe there'll come a day when I put down the pen for good, never to pick it up again.
We're dependent on language models now. Personally, my most recent example came yesterday, when I asked Copilot to turn an email into a text message. The person I was trying to reach hadn't replied. But instead of simply summarising the email in two lines myself, I asked a language model to do it for me.
Worming its way into our lives
Slowly but surely, Claude and ChatGPT have become central to how we express ourselves. To the way we search for information. The way we code.
Of course there are advantages too! I save loads of time every day. I do things I'd normally never bother with (like setting up a budget in Excel – ugh!). But AI has also made me spend more time on other things.
The threshold for sending low-quality content has dropped dramatically – better known as workslop.
Curious about workslop? I've written about it here:
Sebastian Parry-Jones Øyrehagen
Maybe that's why Kludder has become one of the highlights of my week. The time when I sit down and just write. Not because I have to. Not because the job demands it, but because it feels good to sort through my thoughts.
The first fix is free
We've gotten used to our little AI helper. That little chat window that answers everything you want to know. The window that flatters you when you ask it to go through your LinkedIn profile. Because who doesn't recognise that classic LLM line: "You're not just XX. You're XX who also does YY, and that makes you truly stand out."
The chat window has been sitting on our phones for years now. At the office, you're falling behind if you're not using Copilot constantly. And anyone who isn't AI-generating renovation ideas for their bathroom is stuck in 2023.
The time has come for the AI companies to cash in. All year there's been speculation about IPOs for both OpenAI and Anthropic. They need to show earnings that match their user base. And that's what users are starting to notice.
Among Anthropic users in particular, there's grumbling about limits being hit faster than before. Hardly surprising, when you read about users paying $200 a month for a Pro-plan, yet they are using tokens worth $5,000 on the same subscription.
The companies behind the language models have subsidised usage from the start. Going forward, I think we'll see less goodwill from Sam Altman and Dario Amodei. That could mean a new reality for all of us.
It wasn't long ago that I wrote about bossware – software that lets your boss monitor you. In Europe we have strong privacy protections compared to the rest of the world. But I worry about the future of the workplace when the cost of AI use goes up. Companies will have an incentive to make sure you're not using Copilot to plan a stag do, touch up your CV, or create a shopping list.
Sebastian Parry-Jones Øyrehagen
In the US, software company Zapier has rolled out a service that lets you monitor employee token usage. It's being used to spotlight employees that've really leaned into the AI wave. But it can also be used to audit usage – and verify that the language model is being used for actual work.
Is it too late?
I don't want to be AI-negative. Perhaps I'm naive, but I see so many possibilities for truly exciting use. Two days ago I presented the 11Labs tool to my team at work. It can AI-generate voices, and markets itself as excellent for customer service centres. But the use cases go far beyond that. The voice can read out an article while you're out for a run, or cooking. It can read back what you've written yourself, so you can check whether the words flow the way you want.
We can certainly shake our fists about voice actors' jobs being under threat. I don't think they are in this case. And I genuinely hope we will be using voice actors in the future. But no company hires voice actors for internal articles. Used properly, the kind of tools 11Labs is developing could expand the way we absorb information.
When the fun is over, and tokens become a currency on a par with dollars and pounds, sensible use will matter more than ever. I believe AI will become a tool you're expected to use where it genuinely makes sense.
And maybe that's not turning an email into a text message. That one you'll have to manage yourself.
It's going to be hard for a lot of people. It happened surprisingly fast – forgetting how to brainstorm. Relearning it will take time.
Meanwhile, the chat window sits in the background – like an old summer fling you can't quite forget. But we all know what it's like to go back to an ex.
It's never quite the same.
Calling for honesty
The CEO of Verizon, Dan Schulman, says it's high time senior executives acknowledged that artificial intelligence will cost people their jobs. In an interview with the Wall Street Journal, he's predicting twenty to thirty per cent unemployment within two to five years.
Schulman took over as CEO of Verizon last October. In that sense, it's perhaps not surprising that he wants to come across as bullish about how AI can strengthen Verizon. As we all know: AI hype = rising stock prices.
Or maybe he's right. We'll see.