Your Favorite Tool Is Getting Worse. That's Not Bad Product Management, That's Panic.

You know the tool. The one you picked years ago because it solved exactly one problem that was making your week miserable. You learned its shortcuts. You stopped thinking about it. That's the highest compliment software can get: you forgot it was there.

Then you opened it last month and the button had moved. There was a little sparkle icon sitting where the thing you actually use used to be. And the bug you reported, the boring one, the one that's been there for two years, was still there.

Sound familiar? Here's the part nobody in that company is going to say out loud: they know they're going to be replaced. Most of what made them special is a prompt away now. They're panicking, and software feature bloat is what panic looks like from the outside.

It's Not Just You. Half the Planet Thinks the Update Is a Trap.

54% of Americans believe software updates exist to push them toward premium features. In the UK it's worse: 61% believe software is deliberately degraded to force paid upgrades.

Those numbers come from a UserTesting survey fielded in April 2026 across 4,000 people in the US, UK and Australia. A few more from it, because they kept getting worse as I read:

• 78% of Americans, 81% of Brits and 85% of Australians avoid changing anything about their devices unless they have to
• 40% need days to adjust after an update. 25% need weeks or months
• 15% of Americans name "unwanted AI features" as a reason they don't update at all
• 56% feel anxious or annoyed before they even click

But the number that actually matters is this one: 68% of Americans said they would install an update if it were security-focused with minimal design changes. In Australia it's 77%.

Read that again. People are not change-averse. They are redesign-averse. They will happily take your patch. They will not happily relearn where you moved the export button. Every vendor I've watched in the last two years has gotten this exactly backwards, and the survey suggests they've now trained a majority of their users to assume the worst about every release note.

The Math That Turned Your Roadmap Into a Panic Room

In February 2026, roughly $285 billion evaporated from SaaS valuations in about 48 hours. Forbes ran it under the headline "$300 Billion Evaporated: The SaaSpocalypse Has Begun" on February 4th. Reporting across that week put the total software market cap loss above a trillion. Forrester, who are not exactly known for hyperbole, published a piece titled "SaaS As We Know It Is Dead".

Here's the mechanism, and it's simpler than the panic makes it look.

Per-seat pricing assumes a human being sits in front of the interface. That's the whole model. You have 40 employees, you buy 40 seats, the vendor's revenue grows when your headcount grows. It worked beautifully for fifteen years.

Now put an agent in the loop. The agent doesn't need a seat. It doesn't need the interface at all. It needs an API and a well-shaped output. If the work happens without a human in the UI, nobody buys the seat, and the entire revenue model of a company with a thousand employees quietly stops making sense.

The market noticed. Per-seat pricing dropped from 21% to 15% of SaaS companies in twelve months, while hybrid pricing climbed from 27% to 41%. Forced AI bundling was named the single largest driver of SaaS cost increases in the first half of 2026.

And every product manager at every one of those companies read the same headlines you did.

This is the part where I have to be fair to them: their panic is rational. If your unique selling point was "we structure your data nicely and give you a decent interface for it," that genuinely is a prompt away now. That's not a slur. That's just true. It's the same dynamic I wrote about in Every WWDC Kills a Few Apps, except the thing that used to eat your product was one company's keynote once a year, and now it's a model that gets better every eight weeks.

Six Tools, Six Dates, One Pattern

I'm not going to editorialize between these. Just the dates.

Figma. UI3 announced at Config on June 26, 2024. Rolled out to all users October 10, 2024. Became the mandatory default in spring 2025, legacy UI removed, no opt-out. The forum threads are titled things like "Forcing UI3 on us is a huge mistake, let us choose" and "Figma's forced updates: a masterclass in user-hostile design." Moved features, changed shortcuts, muscle memory gone.

Slack. Killed the standalone $10/user/month AI add-on after August 17, 2025 and folded AI into Business+ at $15/user/month annual. If you were paying for the add-on, your renewal option was the pricier tier. There wasn't a second option.

Notion. Bundled AI into Business and Enterprise in May 2025 and removed the AI add-on for new Free and Plus customers. For a five-person team on Plus, getting back the thing you already had works out to roughly an 80% higher per-seat bill.

Firefox. The new CEO announced turning Firefox into "a modern AI browser." After the backlash, Mozilla promised an AI kill switch in Q1 2026. Notice the tell. A kill switch means the features are opt-out. Nobody builds a kill switch for something you had to turn on.

WeTransfer. July 2025, the terms of service quietly gained language allowing your files to "improve performance of machine learning models." After the outcry they stripped the machine learning, derivative works and commercializing references within days and said they weren't training on anything. Fine. But somebody wrote that clause, and somebody approved it.

Duolingo. The 2025 "AI-first" memo. Users deliberately broke streaks they'd kept for hundreds of days, on purpose, as protest. The CEO walked the memo back and scrapped the AI-usage performance rule in May 2026.

Seven, if you count the one I already complained about at length: Dear WordPress, nobody asked for AI connectors, we asked for caching.

Six companies, wildly different sizes, wildly different products, same eighteen months. That isn't six product teams independently losing their judgment. That's software feature bloat arriving on a schedule, and the schedule belongs to the market, not to any of their roadmaps.

The Uncomfortable Half: Their USPs Really Are a Prompt Away

I can't sit here and act superior about this, because I'm one of the people doing it to them.

I've cancelled paid tools this year and replaced them with a script and a model call. Not because the script was better. It wasn't. It was worse in about four ways I could list. But it was mine, it cost me an afternoon, and the recurring bill went to zero. I wrote a whole post about which parts of my stack I moved off SaaS entirely: Your SaaS Stack Is Bleeding You Dry.

Multiply me by however many technical customers a vendor has, and you understand the meeting.

There's a stat that makes it concrete: generative AI functions are now among the most duplicated functions in enterprise software portfolios, showing up around seven times across an average company's stack. Seven vendors, all charging, all shipping the same capability, all watching the others do it. If you're the fourth of those seven, you are not adding an AI feature because your users asked. You're adding it because the sales deck needs a checkbox and because the alternative is being the one product in the category that doesn't have it.

That's not stupidity. That's fear with a budget.

Now the Part Where I Argue Against Myself

Software feature bloat did not start with AI, and I'd be lying by omission if I let you think it did.

In 2002, the Standish Group presented data showing that 45% of features in a typical software product are never used. Another 19% are used rarely. Only 7% get used always. That's twenty-four years ago, long before anyone had a model to bolt onto anything.

Niklaus Wirth wrote "A Plea for Lean Software" in 1995, and his observation, now called Wirth's Law, was that software gets slower faster than hardware gets faster. Also decades old.

And then there's Zawinski's Law of Software Envelopment, which is the one that actually keeps me honest: "Every program attempts to expand until it can read mail. Those programs which cannot so expand are replaced by ones which can."

Look at the second sentence. That's the part people forget when they quote it as a joke about bloat. Zawinski isn't only saying products expand. He's saying the ones that refuse to expand get replaced by the ones that do. Sometimes bloat is what survival looks like from the inside, and the lean tool you're nostalgic about died because it stayed lean.

Sometimes the AI feature also just lands. GitHub Copilot hit 4.7 million paid subscribers as of January 2026, up 75% year over year, deployed at around 90% of the Fortune 100. Those are Microsoft's own numbers, and vendors publish flattering numbers, but that's not a rounding error either. Somebody's paying, repeatedly.

I'll go further. There's an AI feature in a tool I use daily that I complained about publicly when it shipped, and I now use it every working day. I was wrong. I don't have a clean rule for telling a panicked feature from a real bet when I'm standing on the outside looking at a changelog, and neither do you. Anyone who tells you they can is selling something.

How to Stop Caring What Your Vendor Does Next

The good news is that you don't have to diagnose their panic. You just have to stop being exposed to it.

Test the export now, not at renewal. Actually run it. Open the file. Is your formatting there? Your attachments? Your history? If the export is broken or lossy, you don't own a tool, you own a hostage situation, and you'll find that out on the worst possible day.

Score your stack on switching cost, not on features. The question isn't "is this the best option." It's "how many hours does it cost me to leave." A slightly worse tool you can walk away from in an afternoon beats a slightly better one that holds four years of your work in a proprietary format.

Buy from vendors too small to panic. A team of eight with no investors doesn't have a board asking about the AI story. It has users who pay. That's the whole reason Obsidian beats note apps with a hundred engineers.

Read the renewal email like a contract, not a newsletter. The AI bundle is a price increase wearing a feature's coat. Work out the actual per-seat delta before you click accept. On a five-person team, the Notion-style move I described above is real money for something you were already getting.

Keep a one-page "if this dies" note per critical tool. What it does, what the alternative is, how the data gets out. Fifteen minutes each, once. It is the difference between switching on your schedule and switching in the middle of a client deadline.

And stop reporting the boring bug. I know. It's a real bug, and reporting it is the right thing to do, and I still do it. But it is not on the roadmap and it isn't going to be. Route around it and stop spending emotional energy on a ticket nobody is reading.

What I'd Want Someone to Say to Me

I ship my own software. A couple of small macOS apps with paying customers, nothing that would show up on anyone's radar. And when everything in my feed for six straight months was about agents eating categories, I absolutely felt the pull to bolt something AI-shaped onto a product that did not need it. I didn't, that time. I'm not confident I always won't.

So I don't think the vendor is your enemy. I think they're scared, and scared people make fast decisions with other people's workflows. That's not an excuse for them. It's a forecast for you: this keeps happening through the rest of 2026, and probably well past it, and no amount of forum threads is going to slow it down.

Build the exit before you need it. Not because your tool is definitely going to get worse, but because the cost of being wrong about that is now much higher than the cost of a fifteen-minute note.

If your business is running on a stack that keeps shifting under you and it's costing real hours every month, that's a thing I untangle for clients. Let's talk if you'd rather not audit it alone.