Atlassian reported earnings this week. Revenue up 32%. The stock jumped 23% in a single day. Salesforce is up about 20% over the past month. Neither of those numbers makes much sense if you believed the story everyone was telling back in February — that agentic AI was about to gut companies exactly like these.
Back then the numbers ran the other way. Software stocks had one of the worst stretches anyone could remember. Salesforce fell 26%. Atlassian dropped 36%, and at its low point for the year was down more than half. Over a trillion dollars came off enterprise software valuations in about a week, and the financial press started calling it the “SaaSpocalypse.” Satya Nadella had basically called it more than a year before it happened, telling an interviewer that business apps are essentially CRUD databases with some business logic bolted on, and that logic is what’s about to move to agents. At the time that sounded like a founder talking his own book. By February it read like a prediction that came true early.
So what changed? Not the thesis, really. Just who it applies to. Datadog is up 76% off its March low. CrowdStrike, 55%. Meanwhile, HubSpot was still down 46% by late spring and Monday.com 45%, even with both companies posting decent revenue growth. And now the two loudest February casualties are having their best quarters in years because of enterprise AI agents, not despite them. Atlassian’s CEO credited the beat to customers running their teams and workflows through what the company now just calls its platform. Salesforce’s growth is coming almost entirely from Agentforce, which crossed a billion dollars in annual revenue and delivered close to 4 billion units of agent work last quarter alone, more than double the quarter before.
Here’s the pattern once you sit with it for a minute: nobody’s getting punished for having software. They’re getting punished for making a human operate it.
Stop making people log in to find out what’s true
Think about what a login screen actually is. It’s a wall between a person and whatever they need to know or do, and the only way through it is learning your particular arrangement of tabs, fields, and menus. For forty years that was fine, because there wasn’t another option. Now there is. A clinician doesn’t want to learn where the discharge summary button lives in your EHR. A sales rep doesn’t want to remember which of four tools has the field they need. They want to say what they need and have it happen, on whatever surface they’re already sitting in: Slack, a chat window, or a text field embedded in the app they already have open.
That’s the real shift, and it’s a smaller, more specific claim than “software is obsolete.” The winners aren’t the ones without a product anymore. They’re the ones who stopped requiring you to come to them. Salesforce didn’t get replaced by an agent, it built one and pointed it at the same data and workflows the app was always sitting on top of. The app didn’t go anywhere. The requirement to open it did.
Most companies can’t build that themselves or don’t have the data plumbing to try. That’s the actual bottleneck, and it predates all of this by decades. People bought software to do the work, and eventually operating the software became the work. A knowledge worker logs into a dozen systems a day without thinking about it. A clinician retypes an order that already exists in another screen. A rep pastes a deal update into three tools because none of them talk to each other. Nobody’s job description says “be the glue between systems,” but that’s what a lot of people spend a chunk of their day doing.
Enterprise AI agents – what that looks like in practice
Onix ran into this directly this year, building a clinical workflow for a large EHR provider on Google’s Gemini Enterprise AI Agent platform. The provider wasn’t going anywhere. The goal was to make its existing system work the way the companies above are learning to work. Labs, orders, notes, and scheduling data were spread across systems that were technically connected but left clinicians bouncing between screens for routine things. Onix’s platform, Wingspan, started by mapping that data landscape: what existed, where it lived, who was allowed to see it, before a single interface got built. Only after that got consolidated onto Google Cloud’s managed FHIR store did the team put a chat and search interface in front of it, embedded right inside the EHR clinicians already had open. Nothing new to learn. A clinician can ask for a pre-op panel and get it ordered with the allergy history already checked, then ask in the same breath for a discharge summary drafted and sent to the right physician, without switching windows. Every step is logged, because an agent working inside a hospital’s systems needs the same paper trail a person doing that job would leave.
The point isn’t the healthcare specifics. It’s that the hard part was never the chat window. It was building something with enough access, structure, and accountability underneath it that the chat window could actually be trusted to do things.
AI for enterprise – what are the numbers?
Anthropic’s 2026 State of AI Agents report puts real data behind this: 80% of enterprises running agents say they’re already seeing measurable returns, and 57% have moved past single tasks into multi-stage workflows. PwC’s survey is more conservative, with two-thirds of adopters reporting real productivity gains, but notes fewer than half of companies have actually changed how they operate around any of it yet. Both of those can be true. Adoption is real. Most organizations just haven’t caught up to their own agents yet.
How to implement enterprise workflow automation
Don’t go looking for the vendor that’s about to disappear. That’s the wrong question, and February proved it. Go looking instead for the workflow where your own people are doing the most manual stitching between systems right now. That’s where an agent pays for itself fastest, and where you’ll learn the most about what breaks when you try.
Build the boring part first. Identity, permissions, and an actual map of where the data lives. Skip that, and the agent isn’t smart; it’s just guessing politely.
And treat the agent like you’d treat a new hire, not a feature you flipped on. Give it a defined scope, keep a record of what it did, and let it earn more trust over time instead of handing it all upfront.
Nobody’s app is dying. But the app that only works if someone learns its screens is going to keep losing ground to the one that shows up wherever its customer already is.
Conclusion
In the age of AI-enabled enterprise workflow automation, enterprise applications are not dying but are transforming to changing business needs. At Onix, we’re at the forefront of this transformation with our suite of AI workflow automation tools and platforms.
Are you ready to embrace the future of enterprise applications? Connect with us.
