My AI minions just learned a new trick this week. Anthropic shipped a feature, I rewired three of them around it, and by Friday I was already wondering if I would have to do it again next month. One step closer to world domination, but first, the long-term AI strategy doc needs another rewrite.
If your AI strategy depends on a tool that did not exist 90 days ago, it is not strategy, it is speculation. I sit in AI strategy conversations with client executives on a near-weekly basis (one of the side effects of running a fractional CIO seat across multiple organizations), and some version of this line lands in almost every one. I am also in a couple of peer groups with other CIOs and IT leaders, and the conversation looks identical from the other side of the table. The response is almost always the same. A pause. A small wince. Then a slightly defensive “well, but the roadmap calls for…”
The roadmap calls for a tool that shipped in beta last quarter, has been re-architected once already, and is competing with three other vendors who are all about to ship something similar. That is not a roadmap. That is a bet wearing a roadmap costume. And the gap between those two things is where most long-term AI strategy work falls apart right now.
I am not saying do not bet. I am saying call it what it is, sleeve up the risk appropriately, and stop pretending you have a long-term AI strategy when what you actually have is a tool subscription and a hope.
What 90 Days Really Means in This Cycle
Two years ago I was using ChatGPT to draft training documents and IT policy text (and honestly, mostly treating it like a fancy autocomplete). Today I am building production applications with Claude Code, running a small army of AI agents across my client work, and standing up Azure platforms where AI does most of the actual implementation lift. That is not a slow evolution. That is a category shift in 24 months, and the pace is still accelerating.
Here is what has shipped or materially changed in the last 90 days alone across the tools most of my clients are actively evaluating. Microsoft 365 Copilot picked up a new agent layer and quietly re-priced (again). ChatGPT shipped a new agentic mode. Claude added skills, code execution, and computer use. Google rebranded half their AI surface and merged two product lines. Every one of these vendors has at least one more major announcement queued for next quarter, and that is just the ones we know about.
I run a fractional CIO seat across multiple organizations, and I am watching the same scene play out everywhere. A leadership team asks for a five-year AI strategy. The IT team writes one. Six weeks later, the foundational tool in that strategy has been deprecated, repositioned, eclipsed, or absorbed into something else. The strategy goes back in the drawer. A new strategy gets written. The cycle repeats.
The instinct is to blame the planning team for not being agile enough, or to blame the vendors for moving too fast (they are), or to blame leadership for asking for the wrong artifact. But the actual problem is structural. We are trying to apply a planning horizon that worked fine for ERP migrations and infrastructure refreshes to a tool category that is moving on a quarterly release cycle. The math does not work. It is not a discipline problem. It is a category mismatch.
The Difference Between Strategy and Tool Selection
This is where I have to say something a little uncomfortable. Most of what gets called “AI strategy” in IT planning meetings right now is actually just tool selection wearing a strategy costume. The deck has a few mission slides up front and then forty pages on which vendors we are picking and when. That is procurement with ambition. It is not strategy.
A real strategy answers durable questions. What does the organization want to be able to do? What kind of work do we want humans focused on, and what do we want machines handling? What is our tolerance for risk in data exposure, hallucinated output, or automated decisions reaching customers? Where in the business is AI most likely to create real leverage, and where would it just create new problems with a more expensive license attached? These questions do not change when a new model drops on a Tuesday.
Tool selection answers a different question. What are we buying and deploying right now? Copilot or ChatGPT Enterprise. Claude for Work or Gemini. One agent platform or another. These choices matter, and they matter a lot for the next twelve months. But they have a useful life measured in quarters, not years. Treating them like the foundation of a long-term AI strategy is exactly how you end up with a binder full of dead recommendations.
Here is what actually works. Build the long-term AI strategy around the durable layers, and treat tool selection as a separate, faster-cycle decision that gets revisited on a regular cadence. Quarterly works for most mid-size organizations. Some need monthly. The two artifacts live in different documents, get reviewed at different intervals, and answer different questions. That separation is what makes the long-term piece actually durable.
The Durable Layers Your Long-Term AI Strategy Should Sit On
A few things have not changed materially in the last 90 days, and they will not change materially in the next 90 either. These are what your long-term AI strategy should actually be built on.
Data hygiene and access governance. Every AI tool, regardless of vendor, will only be as good as the data it can reach and as safe as the access controls wrapped around that data. I spent the last two months preparing a Microsoft 365 Copilot readiness assessment for a client, and 80 percent of the work had nothing to do with Copilot itself. It was permissions cleanup, sensitivity labeling, sharing audit, SharePoint hygiene, and oversharing remediation. That work applies just as well if the client ditches Copilot tomorrow and adopts something else. The investment compounds across whatever tool comes next.
Identity and access architecture. If your Entra ID tenant (or whatever you are calling your identity provider this quarter) is not in good shape, no AI tool will be safe to roll out. Conditional access, MFA enforcement, role-based access, just-in-time elevation, and proper service principal hygiene. None of this is AI specific, but all of it is prerequisite to any AI deployment that touches sensitive data. This is the boring foundational work that does not change when the vendor lineup changes.
Change management capacity. Can your organization actually absorb a new tool, train people on it, and shift workflows around it? I have watched organizations stand up Copilot licenses for 200 users and have maybe 12 of them actually use it productively six months later. That is not a Copilot problem. That is a change management capacity problem. AI adoption is bottlenecked by how fast humans can change, not by how fast vendors can ship. Investing in the human side keeps paying off no matter which tool wins.
Decision frameworks for evaluation. When the next tool drops (and it will, probably this week), how does your organization decide whether to adopt, pilot, ignore, or wait? Having a documented evaluation framework matters far more than any specific evaluation result. The framework is the durable asset. The decision it produces this quarter will likely be obsolete next quarter. Build the framework once and reuse it forever.
Skills and AI literacy. People who can think clearly about what AI is good at, what it is bad at, where to apply it, and how to verify its output. This is the most underinvested layer in nearly every organization I work with. Tools change constantly. The person who can evaluate the next tool intelligently and verify its output critically is still useful in three years. Train the humans. The tools will sort themselves out.
The Quarterly Rebase
The other half of this is admitting that tool decisions need to be rebased on a much faster cycle than traditional IT planning allows for. I run a formal quarterly AI review with every client now. Not an annual strategic plan. A quarterly check on which tools are in play, which are working, which need to be reevaluated, and which new ones deserve a pilot slot. The strategic plan stays stable. The tool layer gets refreshed.
The deliverable from a quarterly rebase is short. It is not a fifty page document. It is a one page summary of what changed in the market, what changed in our environment, what is working and what is not, and what we are doing differently as a result. Most of the content is recycled from the previous quarter, because most of the durable strategy genuinely does not change. The tool decisions update. The skills investments roll forward. The governance posture adjusts at the margins.
This pattern (long-term AI strategy on durable layers, tool decisions on quarterly rebase) is what lets you actually have a long-term plan without it being obsolete by Q2. The strategic anchor is stable. The tactical execution stays fluid. Both things can be true at the same time, and the organizations that get this right will outperform the ones that keep rewriting their five-year plan every six weeks.
What This Looks Like in Practice
A real example, with the names sanded off. One of my clients was about to commit to a three-year roadmap centered on a specific agentic AI platform that had launched six months earlier. The plan called for full deployment by year two and organization-wide adoption by year three. The vendor was hot. The demos were great. The board was enthusiastic. I read the draft and asked one question. What does this plan look like if the vendor pivots, gets acquired, or is eclipsed by something better in eighteen months?
Silence. Then a really good conversation.
The revised plan kept the same destination (an organization that can deploy and govern agentic AI well) and decoupled it from the specific platform. The durable layers got most of the investment dollars. Data classification, governance maturity, evaluation framework, internal training, and a proper AI use policy. The platform decision became a quarterly pilot that could be expanded or pulled without blowing up the broader strategy. That client is in a much better position now than they would have been if they had bet the entire roadmap on a single tool that may or may not still be the right answer in two years. The strategy survived. The tool choice stayed honest.
The Honest Caveat
This approach is harder to sell to leadership than the clean tool-centric roadmap. Boards want a confident answer with a clear vendor name attached. Telling them that the specific tool is a quarterly decision sounds less impressive than committing to a three-year platform strategy with logos on the slides. There is a real political cost to the right answer here, and pretending otherwise is not honest.
But the cost of the wrong answer is bigger. I have watched too many organizations burn budget on a deployment that became irrelevant before it finished rolling out. That waste shows up in next year’s IT budget conversation, and the IT leader is the one who gets to explain it. Better to take the political hit upfront, frame the approach correctly, and be right in eighteen months than to look confident now and be wrong later (with a paper trail).
The way I frame it for boards: we are committing to outcomes, not to vendors. The outcomes are durable. The vendors are not. This framing usually lands once leadership sees the second or third pivot from a competitor who locked themselves in.
Where to Start This Week
If your current AI strategy is built around a specific tool that did not exist a year ago, do not throw the document out. Just take it apart. Pull the durable strategy out of it (the why, the where, the governance posture, the skills investment) and let those become the long-term plan. Take the tool selection out and make it a quarterly review item. You will end up with two documents instead of one, and they will both age much better than the original.
If you have not built any AI strategy yet, start with the durable layers and skip the tool selection entirely for now. Get your data house in order, your identity layer solid, your evaluation framework drafted, and your people trained to think clearly about AI. The first tool decision will be a lot better when you make it from that foundation instead of from a vendor pitch deck. Reading Microsoft’s own Copilot adoption guidance is also worth an afternoon, because even the vendor admits the prerequisite work matters more than the license itself.
The vendors will keep shipping. The landscape will keep moving. Your long-term AI strategy gets to be stable anyway, but only if you stop confusing it with whatever you are buying this quarter.
I am building a quarterly AI rebase template that pulls from a client’s environment data, recent vendor announcements, and the prior quarter’s decisions to produce a one-page rebase document on demand. It will land under the AdaptoIT product line when it is ready. If you want to know when it ships, or you want to share what your own rebase process looks like, drop me a note.