A VC asked me a fair question on a pitch call last week. If Microsoft and Google are building AI into everything companies already pay for, why would anyone hire a team to build custom AI systems? I get a version of it constantly, and it deserves a straight answer instead of a defensive one.
Start with what’s true. Google Cloud grew 63% year over year in the first quarter of 2026 to $20 billion in revenue, its first quarter ever above that mark, led by enterprise AI. Microsoft has Copilot wired into the software most of the Fortune 500 already runs every day. The platforms are not slowing down, and for a wide range of everyday work they are good and getting better. If that were the whole story, the honest answer would be to buy the subscription and move on.
But “we’ll use Copilot” and “we have an AI strategy” are two different things. The gap between them is where most of the value sits.
What the platforms will absorb
Be specific about what big tech is going to commoditize, because it’s a lot. Summarizing a thread. Drafting an email. Cleaning up a doc. Meeting notes. Answering questions about a file. Simple search across your company’s content. These become table stakes, folded into licenses you already hold. If that list covers what your company needs from AI, then Microsoft and Google are probably the right answer, and I’d tell you so on the first call.
The problem is that for most operations, that list is the easy part. It assists a person who is already doing the work. It does not change how the work moves through the business.
Where Copilot actually stops
The wall is always in the same place: the systems the platform doesn’t natively talk to, the data it can’t see, and the business logic that only exists inside your operation.
Copilot lives inside Microsoft Graph. It can read your documents. It cannot tell you which client takes priority when two deadlines collide, when a legal filing should jump the normal task queue, what your partner-approval rules are, or which system wins when your CRM and your inbox disagree. That knowledge isn’t sitting in a document for it to summarize. It has to be built.
Real workflows also rarely live inside one ecosystem. A single process might touch Gmail, Slack, HubSpot, Stripe, a phone transcript, a PDF, and a vertical platform like Yardi or Clio that a decade of operations has been built around. If your business runs on one of those systems, Microsoft Graph doesn’t know it exists. The platform that owns one slice of the work almost never owns the whole workflow, and the whole workflow is the thing worth automating.
Microsoft’s own CEO just made the argument for me
Here’s the part I didn’t expect to be able to say out loud. In a Wall Street Journal interview published June 21, 2026, Satya Nadella warned that the economy can’t be allowed to depend on a few AI giants capturing all the value. He pushed for cheaper models, more customer control, and companies owning their own “learning loop” on their private data instead of renting intelligence from a handful of frontier APIs.
That is the CEO of Microsoft arguing against the exact dependency that “just use whatever’s bundled” quietly walks you into. Microsoft is already letting Copilot users pick between models for longer tasks and weighing cheaper open models underneath. The platforms themselves are moving toward model choice. Designing around that reality, rather than betting your operation on one vendor’s roadmap, is the safer architecture, and their own leadership is now saying so.
Waiting for the platform is like waiting for AWS to design your product
The other version of the objection is “let’s wait, the platforms will get good enough.” I understand the instinct, but look at what you’re actually waiting on.
AWS gave a generation of companies world-class infrastructure. It never once told them what to build on top of it. Copilot is the same kind of thing. It’s a capable AI surface. It does not design your operating model, encode your decisions, or wire your systems together. Someone still has to do that.
And the better these tools get, the more that holds. Cheaper models and stronger platforms mean a well-designed implementation does more, faster, for less. The company that already has clean data and structured workflows gets more out of every platform upgrade the day it ships. The company that waited is still starting from zero, now against competitors who didn’t. The cost of waiting isn’t theoretical. It’s the inefficiency running in your operation right now while you wait for someone else to fix it for free. That gap is exactly what we map before writing any code, and it’s most visible in operations built on legacy vertical platforms the big tools were never designed to see.
None of this is a fight with Microsoft or Google. If Microsoft has the best model for a task, we use it. If Google has the best context for the work, we use it. If an open model is better for a private workload, or plain deterministic code beats a model entirely, we use that instead. The point was never vendor loyalty. The point is orchestrating all of it around how your specific business actually runs, which is what every system we ship is built to do.
Microsoft can give you Copilot. It can’t tell you how your business should operate. If the platforms improve the infrastructure, that doesn’t replace the work. It moves the work to the layer that was worth the most all along.
