The pilot project phase is finally over
For a long time, AI in enterprise meant a proof of concept that lived in a sandbox, impressed a few VPs, and then quietly died because nobody figured out how to connect it to actual systems or workflows. I watched this happen at three different companies in the same year. Same story each time.
That era is ending. The teams that kept pushing past the pilot phase — that fought through the integration headaches and the IT security reviews and the change management nightmares — are now pulling ahead in ways that are hard to ignore. And everyone else is watching and getting nervous.
The pressure to move from "we're exploring AI" to "AI is in production" is real in 2026. Boards are asking about it. Customers are asking about it. The pilot-project excuse doesn't land the way it used to.
Copilots are everywhere, but most of them aren't that useful yet
Every major enterprise software vendor has shipped some version of an AI assistant. Your ERP has one. Your CRM has one. Your ITSM platform has one. Half of them feel like someone bolted a chatbot onto a form and called it a day.
The ones that actually work — and there are some — share a common trait: they're deeply integrated with real data and real workflows, not just sitting on top of a generic model. A copilot that can actually pull your Q3 pipeline data, cross-reference it with historical close rates, and flag the deals that look shaky is useful. A copilot that rephrases things you already wrote is not.
The gap between the good implementations and the bad ones is enormous right now. Choosing the right vendor — or building the right internal tool — matters more than it did two years ago when everything was roughly equally mediocre.
The data problem didn't go away, it got more urgent
I've had this conversation more times than I can count: a company wants to deploy AI across their operations, and then someone asks where the training data or the retrieval data is coming from, and the room goes quiet.
Enterprise data is messy. It lives in seventeen different systems that don't talk to each other. It's inconsistently labeled. There are fields that mean different things in different business units. There are spreadsheets that one person maintains and nobody else fully understands.
AI doesn't fix that. It makes it more visible and more painful. The organizations that invested in data infrastructure before AI became urgent are genuinely at an advantage right now. The ones that didn't are finding out the hard way that you can't build reliable AI on top of unreliable data.
If I had to give one piece of advice to an enterprise team right now, it wouldn't be about which model to use. It would be: get your data house in order first.
Security and compliance are slowing things down — and that's not entirely wrong
AI adoption in enterprise is slower than the headlines suggest, and a big reason is that legal, compliance, and security teams are asking questions that don't have clean answers yet.
Where does the data go when it hits the model? What happens if the model hallucinates something in a financial report? Who's liable when an AI-assisted decision turns out to be wrong? These aren't paranoid questions. They're real, and the frameworks for answering them are still being written.
I've seen deals stall for six months because of data residency concerns. I've seen projects shelved because nobody could get sign-off from the legal team. It's frustrating when you're on the implementation side. But I've also seen what happens when these questions get skipped — and that's worse.
The enterprises moving thoughtfully through this, even if they're moving slower, are building something more durable.
The job displacement conversation is happening whether we like it or not
I'm not going to pretend this isn't real. I've been in rooms where a team demonstrated an AI workflow that clearly replaced what three people used to do. Everyone in the room knew it. Nobody said it out loud.
The nuanced version — which I do believe — is that what changes is the nature of the work, not just the volume. A finance analyst who used to spend 60% of their time pulling and cleaning data now spends that time actually analyzing it. That can be genuinely better for the person. But it requires the person to want to grow into that, and the organization to support them doing it. Neither of those is guaranteed.
The enterprises handling this well are being honest about what's changing and investing in reskilling before they have to. The ones handling it poorly are pretending nothing is changing until it's too late to pretend.
What the next few years actually look like
My honest take: the enterprise software landscape looks pretty different by 2028. Not because AI replaces everything, but because the companies that integrated it thoughtfully will have such a compounding advantage in speed and cost that the gap becomes hard to close.
The winners won't be the ones who adopted AI fastest. They'll be the ones who figured out where it actually created value, fixed the underlying data problems it exposed, and brought their people along instead of just deploying tools and hoping for the best.
That sounds straightforward. It isn't. But I've seen enough teams do it right to know it's possible.
The technology is ready enough. The harder question, as always, is whether the organization is.
AI doesn't fix that. It makes it more visible and more painful. The organizations that invested in cloud-native infrastructure before AI became urgent are genuinely at an advantage right now. The ones that didn't are finding out the hard way that you can't build reliable AI on top of unreliable data.
AI adoption in enterprise is slower than the headlines suggest because security and compliance teams are asking questions that align closely with zero trust security frameworks .
AI adoption in enterprise is slower than the headlines suggest because security and compliance teams are asking questions that align closely with zero trust security frameworks .
The same shift is also influencing modern mobile application development , where AI-powered experiences are quickly becoming the standard expectation.