Tory Evans

Our IT Guy Spent a Week Evaluating 'AI Replatforming' and Came Back Confused

February 20, 2026

Derek came back from his evaluation week looking like a man who had tried to assemble furniture without instructions, found extra pieces, and then discovered the furniture was already discontinued. He dropped a 14-page document on the shared drive titled "AI Replatforming - Notes" and then spent the next two days talking exclusively about the Mandalorian. Nobody brought it up. It just sort of sat there.

And here's my take: Derek's confusion is not a Derek problem. Derek's confusion is the most honest thing anyone in this building has produced about AI this year.

I want to talk about AI replatforming. What it actually is, what the vendors are claiming, and why an intelligent, technically capable person can spend five full days researching it and emerge more confused than when they started. Because I think that confusion is signal, not noise. I think it's telling us something important about where this whole thing actually stands.

What "AI Replatforming" Actually Means (Sort Of)

The term itself is slippery. At its core, replatforming means moving your technology from one platform to another while keeping the fundamental logic mostly intact - you're reshaping the application to fit new infrastructure rather than rebuilding it from scratch. That definition has existed for decades. Airbnb did it with AWS in 2024. Nike did it with Salesforce Commerce Cloud back in 2017. Capital One did it with AWS in 2015. It's a known, understood thing.

What's new is the "AI" in front of it. And that prefix is doing a tremendous amount of work right now.

The version being sold to enterprises today goes roughly like this: Your legacy workflow software - the stuff that runs procurement, supply chain, HR approvals, the things that cost you SaaS licensing fees every year - can be replaced by AI-native custom applications built in days, not quarters. The pitch is efficiency. The pitch is cost reduction. The pitch is liberation from multi-year SaaS contracts that you've been trapped in and quietly resenting.

Mistral AI CEO Arthur Mensch made exactly this argument recently and he was specific about it. "In a couple of days, we can create fully custom applications to run a workflow, to run a procurement workflow, or to run supply chain workflows, in a way where I would say five years ago, you would actually need a vertical SaaS," he said. Mistral has framed this as a genuine exit ramp. Mistral has over 100 enterprise customers actively exploring enterprise AI replatforming - organizations looking to retire legacy systems that have grown costly over time.

That's a real claim from a real company with real customers. I don't dismiss it. But let me tell you what Derek found when he went looking for how to actually do that.

The Gap Between the Pitch and the Walkthrough

I sent Derek into this with a specific brief: find out what AI replatforming means for a company our size, what it would cost, how long it takes, and who we'd hire to do it. He came back with notes that mostly describe a week of vendor calls where every conversation started with a roadmap slide and ended when he asked about data readiness requirements.

Here's the thing the pitch decks don't lead with: the enterprises best positioned to benefit from this change are those that have invested in clean, accessible data infrastructure. AI systems are only as capable as the data they are connected to, and organizations that have deferred data governance work will find themselves unable to capitalize on the speed advantages Mensch describes.

That's not a minor footnote. That's the whole story for most mid-size businesses. Sixty-three percent of organizations either do not have or are unsure if they have the right data management practices for AI, according to a Gartner survey of data management leaders. Sixty-three percent. So two out of three companies getting pitched on AI replatforming are being sold a race car when they don't yet have a road.

And Gartner goes further: through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. Sixty percent abandoned. Not stalled. Not delayed. Abandoned.

Derek's confusion makes complete sense now. He was trying to evaluate a thing that requires significant pre-work that nobody in the sales process was eager to quantify.

The Vendor Noise Is Real and It's Intentional

I've been thinking about this from a coaching angle, which sounds strange, but stay with me. When someone comes to me and says "I want to completely transform my life" and then describes a transformation that requires habits, resources, and self-knowledge they don't currently have - that's not a bad goal. That's just a goal with missing prerequisites. The confusion comes when the person selling you the transformation skips the prerequisites entirely.

That's what's happening in enterprise AI right now. Many vendors are contributing to the hype by engaging in "agent washing" - the rebranding of existing products, such as AI assistants, robotic process automation, and chatbots, without substantial agentic capabilities. Gartner has a name for it. "Agent washing." They've been flagging it all year. Gartner estimates only about 130 of the thousands of agentic AI vendors are real.

One hundred and thirty. Out of thousands.

So when Derek was doing his vendor calls, he was statistically likely to be talking to people selling rebranded automation as revolutionary infrastructure overhaul. That's not me being cynical. That's Gartner's math.

The confusion isn't a sign that Derek missed something. The confusion is the correct response to a market where the vocabulary has been deliberately stretched until it covers things it shouldn't cover. Every vendor wants to be your sole platform. A tenured CxO would chuckle at these AI vendor developments. Why? Vendors have always wanted to consolidate all of your spending with them.

Chris read Derek's notes and said "this seems complicated." Chris is extremely kind about these things. I told him it's not complicated, it's just dishonest marketing dressed up in infrastructure language. Chris nodded and offered to get coffee. Chris is good people.

The Real Version of This Is Happening - Just Not How It's Marketed

Here's where I'll give the trend actual credit, because I think it deserves it when you strip the sales layer off.

Something genuine is happening in enterprise software. Mensch points to workflow software, the automation and orchestration layer, where AI is making its most direct incursion. These are the tools that manage defined, repeatable business processes - and increasingly, AI agents can handle those processes with greater speed and at lower cost than a SaaS subscription. That's real. The workflow software category is legitimately under pressure. The argument that custom AI-native apps can replace off-the-shelf SaaS for specific, well-defined processes is not fantasy. It's happening in pockets. Klarna dropped Salesforce and Workday in late 2024 and built their own AI-native stack internally. That's not a pilot. That's a decision.

The enterprise numbers back this up in aggregate. Companies spent $37 billion on generative AI in 2025, up from $11.5 billion in 2024, a 3.2x year-over-year increase. That money is not going to nothing. Something is being built. In 2024, 47% of AI solutions were built internally, 53% purchased. Today, 76% of AI use cases are purchased rather than built internally. Enterprises are buying. They're making decisions. The experimental phase is genuinely starting to close.

But - and I need you to really sit with this - most organizations are still in the experimentation or piloting phase: nearly two-thirds of respondents say their organizations have not yet begun scaling AI across the enterprise. And on the bottom-line question: for most organizations, the use of AI has not yet significantly affected enterprise-wide EBIT. Thirty-nine percent of respondents attribute any level of EBIT impact to AI, and most of those respondents say that less than 5% of their organization's EBIT is attributable to AI use.

I spent a week in my car - before it got repossessed, different story, growth experience - going through AI vendor documentation on my phone. Not for replatforming specifically. But for the rhythm of these pitches. You read enough of them and you notice that the ROI claims are always further down the slide deck than the transformation promise. Always. The sequence matters. They want you excited before they get to the hard part.

What the Actual Stakes Look Like for Real Businesses

Here's what I think Derek should have come back with, and why this matters for anyone running a business right now:

The companies that are making this work are not jumping at the category - they're jumping at a specific problem they already understand completely. The ones that are failing spent tens of millions on reputable vendors with executive sponsorship, and they failed because the data they assumed would be available wasn't, the vendor's capabilities worked differently in production than in carefully staged demos, and the processes they'd planned to automate had already been restructured by the time implementation reached them. Dead in 18 months. Not because the technology is fake. Because they skipped the prerequisites that every sales deck quietly buries on slide 22.

And yet - here's where I actually land on this - I don't think the answer is to wait. Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls, according to Gartner. That's not a reason to avoid the space. That's a reason to be the 60% that doesn't get canceled. The difference between those two groups is almost entirely about data readiness and specificity of use case - not about how much you spent or how impressive your vendor is.

Linda asked me last Thursday how the "AI research project" was going. I told her it was going beautifully, we were learning a tremendous amount. She said Gerald had tried to use AI to plan a trip and it booked them a hotel that didn't exist. I told her that was actually a very useful data point. She looked at me the way people look at me a lot lately.

What Derek Should Have Looked For

My take is this: the evaluation framework was wrong from the start. You can't evaluate "AI replatforming" as a category. It's too big and too polluted with marketing language. "Most agentic AI propositions lack significant value or return on investment, as current models don't have the maturity and agency to autonomously achieve complex business goals or follow nuanced instructions over time," according to Gartner. "Many use cases positioned as agentic today don't require agentic implementations."

The right question isn't "should we AI replatform?" The right question is: what specific workflow do we hate paying for, do we have clean data that describes that workflow, and is there a vendor in the legitimate 130 who has actually deployed something like it in production?

That question is answerable in a week. The broader question is not.

If you're thinking about this from a project management angle, mapping your existing workflows in something like Monday.com before you talk to any AI vendor is genuinely useful prep work - not because Monday.com solves the problem, but because you need to know what you're trying to replace before someone can tell you how to replace it. Same logic applies to business intelligence tooling: if you can't currently see your own operational data clearly, you are not ready to let AI act on it.

The confusion Derek came back with is, I'd argue, the appropriate response to a market that has deliberately made itself confusing. Agentic AI grew up, but still isn't what any reasonable executive would call mature. Agentic AI is a heavy lift that includes forward-deployed engineers, an enterprise data strategy that works, services and use case refinement. Simply put, it's easy to build an AI agent. Scaling them with guardrails and building enough trust to let these models run your company is another matter entirely.

Jamie - the boss's son - came by my desk while I was finishing this piece and asked if we'd figured out "the AI platform thing." I said we'd figured out the right questions to ask. He said that wasn't really an answer. I said a lot of the best answers sound that way at first. He walked away. He'll be fine.

The Bottom Line on AI Replatforming

AI replatforming is a real trend with a real structural logic behind it. The idea that AI agents can replace expensive workflow SaaS for specific, defined, well-documented business processes - that's not wrong. The Mistral CEO isn't making it up. The spend numbers aren't fake.

But the way it's being sold right now is ahead of the infrastructure reality for most companies, the vocabulary is being abused by vendors who are rebranding automation as transformation, and the failure rate for companies that jump in without data readiness is projected to be catastrophic.

Derek came back confused because he was evaluating a real thing using a briefing that the vendors wrote. That's a rigged starting point. The right starting point is your own operations - specifically, the processes you understand well enough to describe to a machine. Start there. Be boring about it. The window is real but it isn't closing tomorrow, and sprinting into it without a map is how you end up being part of the 60% that Gartner already predicted would fail.

I sent my 6am motivational text from the parking garage this morning. The wifi still reaches. Some things persist even when the context changes. Clarity about what you actually need is one of them.