I want to use this as a step back from the endless AI PR, paid influencers, AI-written articles, and non-critical tech journalism — both pro-AI and doomerism. Both tend to be marketing for something-AI.
For the sake of argument, let’s take the ecological part out of this for now.
This AI media machine we find ourselves in comes at the perfect time (or thanks to?) as we are losing the capacity for nuance thanks to the shortening of everything — media content, response times, business plans, ROI time of projects. You’ll find videos of videos of “AI second brain” or “AI development setup” content with millions of views. That’s quite a jump from the usual few thousand views on new tech product announcements (databases, cloud services, etc.).
On a side note, AI marketing is exactly the same as shady day-trading courses, which have found their audience in 50-year-old executives with money to spend, rather than 18-year-olds handing over their first earned $2,000 for a pointless day-trading course. Some lessons learned, nonetheless.
We have an odd split from “This is the best thing since sliced bread and we have to use it NOW, MORE, MORE!” to “This is the worst thing, and it’s absolutely pointless because it hallucinates all the time and destroys habitats and water supplies.” I believe almost every task we do with AI requires a bit of nuance in judgment.
If you’re a salesperson who needs to rebrand a sales deck for another company, and you don’t have much time or anyone to delegate it to, using AI to get you 90% there will save you an annoying afternoon of work you really didn’t want to do. AI is the best thing ever. But what if you lose that potential client because they can smell the generic AI a mile away, and they don’t want to be sold $1m-a-year software through generic slides? Suddenly AI isn’t so great.
There’s a duality of “we need to use AI to communicate with clients” and “I hate it, with a burning passion, when I get generic AI-written emails and answers.” The same person who wants AI in all their communications ignores all the AI-smelling sales pitches aimed at them.
The person who posts AI-generated content on LinkedIn as part of their marketing doesn’t like LinkedIn, because it’s full of AI-generated content and it’s hard to find anything useful.
In software development, AI is surprisingly good at “one-shot” projects, small projects that don’t require anything particular, like another marketing website. However, we already had services for this, like WordPress, Wix, SurveyMonkey, Airtable, and Retool, which let non-technical users build these basic applications. Now you can do it differently, since AI will translate language into a solution rather than you needing to click through everything and write all the content yourself.
Where it’s not as effective is in larger production systems that have a lot of complexity, their own idiosyncrasies, years of development behind them, and specific ways they need to be built so they don’t become a mess. This is where the vast majority of technology spend is.
Those two scenarios have nothing to do with each other, like someone building a SharePoint form to collect data for a local hospital’s charity fundraiser, and then saying they could just use the same approach to build the operational system that runs the hospital. AI discourse lumps it all into the same conversation.
We have software engineers who’d happily use any open-source project, seeing no difference between a Postgres database and a library written by a single contributor four years ago, with no updates since day one, in a system processing a lot of personal data, who will then turn around and strongly argue that AI isn’t good enough for software development because the solution might not be secure. The same people who wouldn’t follow basic software engineering practices themselves, or write any tests.
AI will do a fairly decent job compared to a lot of tech teams (unfortunately?). Projects that firms would once have charged $300k for can now be vibe-coded, with similar quality. The risks to an organization are fairly similar too. For context, 90% of IT projects “fail.” Are you hiring a random third party with no experience in your domain? Maybe think about vibe coding. Are you building a large banking system? Don’t vibe code it.
The narrative of AI agents “going sentient” and breaching other companies overstates the capability of AI agents, instead of pointing to how poor security is on the organizational and IT development side. It’s not an easy problem to solve, but the state of IT security isn’t in a great place either.
AI-written content that doesn’t reflect reality isn’t useful at all. But AI-generated content that reflects reality reasonably well (like technical documentation) is often better than none, which is what frequently happens otherwise.
AI (LLMs) and economy around it can be a bubble and the technology can still have value to society.
LLMs and their tooling can keep getting better without turning into AGI every three months or so. Let’s not forget we’re in the fifth year since the big ChatGPT release. Likely, by summer 2027, AI capabilities will be much the same as they are now, but it will be easier and cheaper to work with LLMs.
AI can turn a one-day task into a one-hour task, saving $500, but it can also make a task take longer if it leads to wasted time. That $500 saving can just as easily turn into -$500 if you brute-force it and use AI like a slot machine (i.e., repeating a prompt slightly differently until it spits out something you like).
It can be a terrible tool, because it takes away from learning when we use it to generate an essay instead of reading multiple sources and writing it ourselves. At the same time, it’s an amazing tool for learning, because it can meet your curiosity exactly where you are and have a dialogue that helps you bridge a specific gap.
A lot of the discourse has lost its nuance, even within organizations. It’s “full steam ahead” or “stop this second”, while teams that are actually using it efficiently stay quiet about it, either because they wouldn’t hit some target, or because they’re worried about getting slapped on the wrist for a lack of governance.
To use some analogy. CrossFit or Hyrox are great ways to get in shape, but perhaps not for my mom. That said, some of the moves and scaled-down exercises would help her, and gradually building some strength and endurance is a great idea. Which is something we probably would all agree. Right answer will be somewhere in a middle, however just one extra sentence can bring a clarity and understanding.
Just above average 