|
Key takeaways
|
Most of us learned the platform from somebody else’s hard-won answer on the community. So when a few of us on the ServiceNow Developer Advocate team got together to argue about AI and the knowledge developers share, it got personal quickly. Read what we think, then tell us whether finding out an answer was AI-written has ever changed how much you trusted it.
Sharing used to be a gift
Two essays make the case from opposite ends. One, on the creative commons, argues that AI has broken the deal that made open sharing work: models take whatever people publish, whatever the license says, and putting your work in the open now mostly invites a flood of low-effort AI contributions. In its words, AI “has turned sharing knowledge from a gift to the world, into a liability for the author.” The other argues that AI has no wisdom, and that people who rely on it to write and even read their code won’t build any either, because they stop making the choices and owning the mistakes that expertise comes from.
Why this one is personal on the platform
The ServiceNow community runs on exactly the kind of sharing these essays worry about: developers going down a rabbit hole so nobody else has to. A lot of that knowledge lives nowhere else, like the workaround somebody posted after an upgrade broke their integration, and the reply three years later saying it still works.
That puts our readers on both sides of the loop. You publish the answers, articles and Share apps (anyone remembers Share?), and you will learn from whatever those look like a few years from now.
Every pass gets a little thinner
Our worry goes a step past such essays, and it’s an opinion rather than a finding. The knowledge we all learned from was built by people writing down hard decisions and what happened afterward. As more of what gets published is generated, more of what the next person reads was generated from whatever came before. It’s a photocopy of a photocopy: every copy is still readable, and every copy has lost a little more of the original.
Honestly, how the next model gets trained is the labs’ problem. What the next developer learns from is ours, and more of what they find reads like it was generated: fluent, roughly right, and missing the part where somebody made a hard call and lived with the result.
The wisdom essay is where this gets uncomfortable. Seniority on the platform has been built the same way for a long time. You make the architectural call yourself, you do the deep dive or run the experiment because nobody could tell you the answer, and you find out whether you were right. If that call gets delegated, and the human job shrinks to choosing between the options a model offers, we don’t know where the next generation of senior architects comes from, or what “senior” will even mean.
Will anyone be able to tell?
We didn’t reach a consensus on this one.
One line of argument says slop is a phase. A couple of years ago, the requirements documents we drafted with AI were painful to read. Today they’re decent. A few years from now, a model that has read everything you’ve written could write in your voice well enough that nobody can tell. At that point the only question that matters is whether the knowledge got across: if a developer follows the steps and can configure the thing, the content did its job, however it was packaged.
The counter is that capability was never the constraint. Many people already back away from content the moment they sense AI wrote it. If we wanted an AI’s answer, we’d have asked one ourselves. We came to a community post for a person’s experience. As we see it, a model is tuned to keep whoever is typing the prompt happy, and that’s the author. The developer who reads the answer six months later, and actually needs to learn from it, was never part of the deal. The wisdom essay goes further than any of us would and predicts companies will one day boast about a “NO-AI” policy as a competitive advantage - we’re not holding our breath though.
Today, the counter is plainly winning. You can still usually tell, and people mind. The first argument is about the day you can’t, and none of us could say with a straight face that the day never comes. The printing press and the internet both swung between miracle and menace before people worked out where they belonged, and from where we sit the pushback on one-prompt content is getting louder. The less flattering possibility is that we’re just a generation that happens to care.
A related question stayed open too. People have long paid more for handmade furniture or clothes than for the machine-made kind, so is the premium about AI at all? The closest we got is that AI can take someone’s ideas along with their labour, without asking, which is exactly the creative commons complaint. Mass production is also what made furniture and clothes affordable for most people, though. AI could do the same for plenty of things that used to depend on who you knew or what you could pay: a tutor that adapts to one kid’s strengths and gaps, a plain-language walkthrough of a contract or a diagnosis, or captions and read-aloud for anyone who needs them.
Where we did agree: the problem is the disguise
AI-generated content can be perfectly good. Passing it off as a person’s experience is where it goes wrong.
Picture two movie podcasts with the same script. One opens with “this show is AI-generated,” and nobody much minds. The other opens with “Hi, I’m Jim,” and there is no Jim. Find out about the second one and you stop trusting anything that show ever told you. The only difference is whether someone lied to you about where it came from.
Online communities, ours included, run into the same thing: answers and articles posted under a person’s name that turn out to be mostly generated. Some of it is good. Some of it is confident nonsense. The damage comes from the reputation that piles up around the name along the way: people start trusting an answer because of who posted it, and once it’s clear that reputation was gamed, it no longer tells you which answers to trust.
So there’s a real case for something like “AI authentic”: content AI helped make, that adds genuine value, and that says so. We think vendors should make that easier to do honestly, ourselves included. Provenance that travels with the content, so a reader can see what a person did and what a model did, would take most of the sting out of this.
Where it’s fine, and where it isn’t
Using AI to research, outline or tighten your writing is the easy yes. Some of us write in a second language, and having AI clean up the grammar is often what gets us to hit Post at all.
An AI-drafted community answer works too, as long as you ran the fix on your own PDI first and mention that AI helped write it. The test is whether you could defend every line when the person asking comes back with a follow-up.
Pasting the output of one prompt into an article or a LinkedIn post and publishing it under your name is where it starts to slip. So is answering a question with a fix you haven’t tried, however confident it sounds. Readers can usually smell both today, and both teach the next developer to expect less.
Where we draw the line is putting your name on work you didn’t do, to build a reputation you didn’t earn. That’s the disguise, and everyone else in the community pays for it.
What we would actually tell you to do
When you write an answer or an article, put your decisions in it: what you tried first, what you ruled out and why, what broke on the sub-prod instance before it finally worked, and what you decided wasn’t good enough to post. A model wasn’t there for any of that, so it can’t write it for you. It’s also the part the next developer actually needs.
On your next real judgment call, like extending a table versus building a new one, or deciding whether a flow or a business rule should own a piece of logic, write your own answer down before you open the chat window. Then compare. You still get the AI’s help, and you’ve had to think it through first, which is the bit that turns into experience.
And when AI did the heavy lifting, say so. One line is enough: “Drafted with AI, tested on my PDI.” It costs nothing, and nobody feels fooled later. We’ll go first:
An AI helped draft this post. We rewrote it until every opinion in it was one we’d defend.
Over to you
Has finding out an answer or article was AI-written ever changed how much you trusted it, even when it was right? If it never bothered you, tell us that too. That’s the answer that would make us rethink the most.
A couple of the sources that shaped the discussion:
