Do you remember when Google pissed off a lot of people by killing its Reader app even though it still had a loyal following?
The announcement came a few months before the actual deed, with the understanding that people needed time to adjust.
That still happens with certain tech-related retirements (6 months for the “send as” feature in Gmail), but the grace-period cycles are definitely speeding up. AI models come and go with little notice, website portals and account settings change with the seasons, and you wake up one morning and your apps opted you in to a bunch of “personalization” features you never asked for.
And that’s only outside of work.
Different workplaces are adopting AI at different speeds, and the tech companies themselves are early and robust adopters. This is good and bad. Good, in the sense that testing what works in your own environment can save you heartache when scaling externally. Bad, in the sense that it feels like everyone’s building the plane while flying it.
Not only do you have all the external changes to keep up with, you’re also subject to a slew of internal changes you might not have known about until they happen. This is because old-school communications plans are now too slow. Then, when you go to try the new tool, you realize it’s only an MVP. Meaning, not fully baked and probably nowhere near the same functionality as either the old experience or the ideal one. The plan is for everyone to be a semi-permanent beta tester so teams can iterate on the feedback as they go rather than developing themselves into a corner too early. Kinda like Agile on meth.
I’ll admit, though, it’s entertaining to watch when there’s a particularly egregious miss on what constitutes an MVP and the torches and pitchforks start swinging. Schadenfreude. And of course, that would never happen to my own team. 🙂
Iteration and whiplash
I get it. AI is changing the nature of software development so fast, it’s difficult to plan ahead. Not to mention, anything you do build and release is also subject to the whims of rapid change. That’s one of the selling points for using open-weight LLMs in your AI stack. You control the stability rather than relying on the kindness of OpenAI and Anthropic.
I do wonder how long it’ll be before people get so sick of having new stuff pitched randomly over the wall at them that they revolt in droves. Sure, the early adopters (usually) don’t mind, but they’re also the folks who have a higher risk tolerance and more patience/desire/skill to help troubleshoot the glitches. In change management parlance, we’d call them the “champions” who help drive adoption by demonstrating their own positive experience.
But the champion system doesn’t always work out the way you intend. If an early fan turns sour on the product later on, that sentiment will hold a lot of sway among the undecided, and it’ll be hard to turn it around. Championship can also go to someone’s head, such that they get arrogant about their superior knowledge and are noticeably off-putting with people who aren’t as adept with the technology.
With AI, the arrogance is manifesting in some particularly ugly ways, driven by fear that if you’re not perceived as a AI-native expert, you’re on the short list for a pink slip.
That’s also why when someone rolls out a tool, thirty clones or “enhancements” suddenly come out of the woodwork and pile on the change. Don’t get me started on the Quip-replacement cage match that’s occurring in my own workplace right now.
Side Quest: Quip is — or should I say was — Salesforce’s low-barrier notes/collaboration tool, and it’s on a path to deprecation in favor of Slack Canvas, which does and doesn’t serve the same function. The official IT recommendation to switch to SharePoint or OneDrive got a big yeah, naw from most folks. Sorry, Microsoft.
There’s also a noticeable us vs. them dynamic among the builders who are hardcore AI natives and the regular folk who like to ask their chatbot questions sometimes. The product cycles are skewing toward the builders — the daring volunteers — even though the numbers don’t bear that out.
Codex (ChatGPT’s main coding-agent tool) had 8 million weekly active users in mid-July, moving steadily toward 10 million. That’s roughly 1% of vanilla ChatGPT’s user base of 1 billion weekly active users, yet features like its desktop app have skewed toward the builders. My theory: It’s because they’re more tolerant of the frequent iteration and because they’re using a disproportionately large number of tokens that makes it a better business bet.
How can we make this suck less?
The recent social media crackdown on slop is a big sign that the general public’s tolerance level for AI-related churn is eroding outside the workplace. So, now that we’ve had a nice kvetching session, let’s chat about ways to cope inside the workplace when your tools start shuffling around like a deck of cards.
Don’t add to the problem
Can you build your own solution and/or replicate the old experience? Yes, that’s easier than ever now, thanks to our code-writing agent friends. Should you? Probably not, unless you have a really unique angle or know you’ll have a big enough user base to justify the time it takes to build and maintain. If you want to do it just for yourself or your team, go for it. Chucking it into the broader mix as an alternative, though, creates learning-and-evaluation churn for your colleagues and undermines the folks who built the official solution.
Give constructive feedback
Still itching to do something, anything about that solution that’s just not quite right? Ask the other party if they accept feature contributions, and if so, put those code-writing agents to work. If not, ask if they’re collecting feedback anywhere. Be respectful about what is and isn’t working. This is a collaborative problem-solving situation, not a shade-throwing opportunity. And be patient while a fix is underway. If the change broke systems that can’t be patched or worked around without a lot of investment, then have a chat with your management chain.
Listen to the champions
If you’re in the skeptical-this-will-work category, try to keep an open mind until it does or doesn’t prove out. Pay attention to what the power users and champions are talking about — they’re usually pretty vocal — and see if you get the same results. Or ask them directly about how they’re dealing with whatever it is you hate about the new tool. The champions often have a direct line back to the product team, which can expedite fixes.
Don’t dump-and-run
If you’re the exuberant creator/announcer of a replacement tool, please, for the love of all that’s holy, let people know what’s coming as early as you can. Nothing will wreck someone’s day faster than seeing a two-paragraph Slack announcement that a tool they rely on suddenly changed and broke their workflows. If you even have the slightest suspicion you’ll be leaving people high and dry, make sure your support system is more robust than “see the wiki FAQ.” (Ask me how I know.)
Most of all, in the rush to demonstrate how AI-cool you are, don’t forget there are humans on the other side, either building the systems or adopting them or both.
Be kind to each other. We’re all dealing with a lot right now.
All opinions here are my own. All text is my own, too, including the em dashes. I welcome constructive comments and discussion on LinkedIn and Bluesky.
Leave a Reply