A few weeks ago I stopped opening one of the main tools I run my company on. I didn't cancel it. I still pay for it every month. It still does its job every day. I just don't look at it anymore.

When Microsoft's CEO said business software would "collapse in the agent era", the internet shortened it to "SaaS is dead." But read what he actually said. Most business apps are really just a database with some rules bolted on top, and in the agent era those rules move up into the AI. That isn't death. It's a shift. The software isn't dying. The way we use it is. And I've spent the last few weeks living inside that change.

The short version: SaaS isn't dying, and neither is the screen. What's changing is the screen's job — from the thing you work in to the thing you check. The value, and the money, is quietly moving to the engine underneath.

The problem every small team has

We're a small team. Nobody here wakes up and does email outreach full time. So it runs the way it runs in most small companies. In bursts. You set up a campaign, a real lead comes in, and everyone drops everything to answer them, build the relationship, close the deal. The outreach quietly stops. Weeks go by.

And under all of that sits a fear. What if the AI gets it wrong? What if we get marked as spam, or blocked, or we embarrass ourselves in front of a stranger?

Here is how I finally got past that fear. You weren't reaching those people anyway. There are thousands of possible customers out there for whatever you sell, and only a tiny slice will ever say yes. Doing it imperfectly, but actually doing it, beats doing it perfectly in your head. Worried about getting blocked? You were already at zero.

The honest comparison isn't "AI outreach versus perfect human outreach." It's "AI outreach versus nothing at all."

First, I tried using the tool the normal way

We use a tool called Apollo. It's powerful. It's also a maze. Lists, sequences, labels, buttons everywhere. I build software for a living and it still took me real time to work out what was happening on the screen. They run actual training seminars just to teach people how to use it.

It has an AI feature that is genuinely decent. You describe your ideal customer, and instead of blasting the same template at everyone, it reads each person and writes them a message based on your company. Sounds great. Two honest problems though.

The first is cost. Each personalised message quietly uses up a credit or two. So a plan that looks like 3,000 messages for about $59 a month is really closer to 1,000. Still cheap. But thinner than it looks, especially if you haven't spent days narrowing down exactly who you're writing to.

The second is context. The tool wants one clean, simple type of customer. If you sell one product to one kind of buyer, wonderful. But we're a services company with different offers for different people, and the AI has no room for that. It can't hold your whole business in its head.

And the wider reality is brutal for everyone right now. Reply rates on cold email have been falling for years as inboxes fill up and spam filters get stricter. One large study put the average reply rate at under half a percent. For us it was hundreds of emails for a single real reply, plus a few automatic out-of-office bounces. Templates are finished. And the moment a message smells like it was written by a machine, people don't even open it.

Then I went around the screen

So I stopped using Apollo's screens and built a small AI assistant that works with Apollo quietly in the background instead.

The assistant reads our own company documents. Our offers, our past clients, how we describe ourselves. For each person it does far deeper homework than the built-in feature ever would, writes a genuinely specific message, and for the first month it handed me a short summary to check before anything went out. I read it, I approved it, it sent. It runs on a cheap always-on server, so my laptop doesn't have to stay open.

Honest results so far? A few replies. Nothing dramatic. I'm still learning and still improving it. But it runs consistently, every day, which is the one thing we could never manage by hand.

Then came the part that changed how I think. Halfway through building it, I realised I didn't really need to build most of it. The newest AI tools can drive Apollo for you directly. You connect them, describe your company and who you want to reach, and simply say "go find them, research them, and write to them." The thing that used to need five seminars and a wall of buttons is now just a conversation. (One tip if you try this. Use the strongest, most expensive AI to set it up, then switch to a cheaper one to run it day to day. That's where the costs stay sensible.)

I never open Apollo's screens anymore. The tool still does all the heavy work. I just never see it.

And it isn't only about outreach

We also build a product of our own called Atlas. It gives small online brands their daily sales and website numbers and tracks them against their goals. The first version was classic software. Log in, look at your dashboards.

Then we added a WhatsApp plugin. Now the numbers come to you. A short note at 9 in the morning on what happened yesterday. Another at 6 in the evening on how today is going, measured against your goals. Our users mostly stopped logging in. Not because the product got worse. Because the screen stopped being necessary.

Same pattern, from both sides of the table. As a user of someone else's software, I went around the screen. As a builder of my own, we quietly removed ours.

So what is actually changing

To be clear, SaaS isn't dead. Apollo still finds the people. Atlas still crunches the numbers. The engines underneath matter more than ever.

And the screen isn't dead either. It's still useful. Sometimes sitting down and looking at a dashboard is exactly what you want. What's changing is its job. The screen is turning into one way in, not the only way, and often not even the main one. More and more, the software just does its work in the background and reaches you through a chat, a voice note, a morning message.

The screen also picks up a new job worth naming. It becomes the place you supervise the engine, not the place you operate it. Where you check what the AI did, set the limits, approve the big calls, and step in on the odd case it gets wrong. And that matters, because nobody sane is handing the whole thing over yet. In one recent survey, only 6% of companies fully trusted AI agents to run core work on their own. So the interface doesn't vanish. It moves from the thing you work in to the thing you watch from.

What's really ending is the old assumption that the screen is the product. That the pretty interface is the thing you're paying for. Once you drop that assumption, the value moves underneath, to your data and to the core thing your product actually does. If your main edge was a good-looking dashboard, that's now a problem. If your edge is the engine, then these AI assistants just became your best way to reach people.

There's a money side to this too. For twenty years, software was sold by the seat. More people logging in, more you pay. But if an agent is doing the work, how many seats is that really? The industry is already feeling it:

When the software does the work instead of the person, you slowly stop paying for chairs and start paying for results.

That's a big shift. For how software gets built, and for anyone whose job is designing screens. The real design question for the next few years is less about where to put the button, and more about how easily a machine can use your product with no buttons at all.

We're a services company that builds these kinds of systems. Imperfectly, out in the open, learning as we go. But after this month, I can't unsee it. The login page is slowly becoming optional.