Platform Trust and the AI Pricing Reset: Lessons from Microsoft Ignite
If AI Becomes the Front Door to Work, Can You Still Get In Without It?
A cheaper AI model won’t help the finance team if nobody can reach the queue to approve a supplier payment. That sounds like an edge case until you draw the architecture and realise how many supposedly separate tools rely on the same platform.
On the GeekWire Podcast, Todd Bishop and Mary Jo Foley discussed Microsoft’s apology for a global Outlook outage alongside its pitch to make Copilot the “OS for work.” The company was also competing on price with AI model providers whose models it sells. The combination raises a question that goes beyond Microsoft: how much of an enterprise’s ability to operate should depend on any one platform?
A convenient front door can become a single point of failure
There’s a good reason to put AI where people already work. An assistant that can find a policy, summarise a request and route it for approval saves people from moving between applications. But once the same platform handles identity, notifications and workflow execution, those conveniences can fail together. A second AI model won’t help if it relies on the same unavailable identity service or can’t reach the system that records the approval.
I would start with one ordinary transaction: a purchase order. If email and the AI assistant are unavailable, can an authorised person still find the order, check the spending limit and record a decision? Can the team reconcile that decision when normal service returns without approving the order twice?
Not every workflow needs a parallel system. For some, waiting an hour is perfectly reasonable. For critical ones, a tested fallback is worth more than a second platform that nobody knows how to use. Make that distinction before designing for redundancy.
Measure the cost of a completed task
Model prices matter, but the lowest price per token doesn’t necessarily produce the lowest cost per completed task. Retries, human review and the work of correcting a bad recommendation all show up somewhere, even if they don’t appear on the AI invoice.
The same applies to switching models. An API abstraction can make the switch look easy in a diagram. It won’t tell you whether the replacement follows approval rules or asks for human review in the right cases. A small evaluation set drawn from real workflows-especially the messy exceptions-is a better test of portability than a successful demo.
There’s a practical parallel in Northeast India. For a district office designing a citizen service, a last-mile connectivity problem and a cloud outage have different causes, but the citizen can be left with the same uncertainty: Was my application received? A service that clearly distinguishes a locally saved draft from a confirmed submission, and can resume without duplicate filings, solves a more immediate problem than adding an AI interface.
Three questions for the next architecture review
- What must keep moving? Pick the workflows that cannot wait and set a target for resuming them, not just for restoring the application.
- What can we replace? Trial an alternative model on representative tasks and measure the total cost per correctly completed task, including review.
- What have we actually tested? Run a drill without the primary workspace. Check that staff can still act within their authority and that decisions can be reconciled and audited afterward.
The best measure of an AI-enabled workplace may be what its people can still get done on the day the AI isn’t there.
About the Author: Sanjeev Sarma is the Founder Director and Chief Software Architect at Webx Technologies. With a core focus on Generative AI integration, Cloud-Native Scalability, and Enterprise Software Architecture, he has spent over two decades driving digital transformation across Northeast India and beyond. Beyond his corporate leadership, Sanjeev is deeply invested in shaping the future of the IT industry. He serves as an Industry Expert on the Board of Studies for Assam Don Bosco University’s School of Technology, advises state technology committees, and actively mentors emerging tech startups at STPI. He brings a unique, dual perspective of high-level enterprise execution and future-ready academic curriculum development.