When Every Company Races to the Same AI Support Stack, the Differentiator Is What Happens When It Fails

This week, the AI-and-support story stopped being theoretical. Bloomberg reported that Microsoft, Uber, and Commonwealth Bank of Australia have all publicly tied customer service job cuts to AI — Microsoft’s support workforce reportedly shrinking from 50,000 to 40,000, and Uber cutting 10% of its customer service jobs while citing its “embrace” of AI. Forrester’s Kate Leggett told The Verge that almost half of customer service roles will be impacted by 2030 — Forrester’s own forecast puts it at 49% of current customer service jobs gone.

And Wall Street loved it. The same week, Salesforce jumped 7%, ServiceNow 8%, and Workday 10% as investors priced in “AI monetization at scale.” Translation: the market now believes every large company will buy roughly the same AI support stack, and that the vendors selling it will print money.

The market is probably right about the vendors. But if you run a support organization, there’s a strategic problem buried in that rally that almost nobody is talking about.

When everyone buys the same stack, the stack stops being an advantage

Think about what actually happened this week. Three companies in three completely different industries — a software giant, a rideshare platform, and a bank — all deployed AI support and all got the same result: fewer humans, faster handling of routine tickets. That’s not three companies innovating. That’s three companies converging.

Your competitors are buying the same agentic platforms, fine-tuning on the same kinds of help-center content, and deflecting the same categories of tickets: password resets, order status, refund policy, billing questions. Within a couple of years, an AI that instantly resolves the easy 70% of tickets will be as differentiating as having a website. Table stakes. CSAT parity on easy tickets is guaranteed, because everyone’s easy tickets are being handled by functionally identical software.

When a capability commoditizes, competition doesn’t end — it moves. And in support, it moves to exactly one place: what happens when the AI fails.

The residual is where the customer decides who you are

Here’s the uncomfortable math of deflection. When AI absorbs the routine tickets, the queue that reaches your remaining humans isn’t a smaller version of the old queue — it’s a concentrated distillation of your hardest problems. I’ve written before about how when AI takes the easy calls, what’s left gets a lot harder. The escalation queue in 2026 is the failure set of a very capable machine: the ambiguous, the emotional, the multi-system, and — critically — the physical.

The customer who reaches a human in this world has already lost a round with your bot. They’re not calling to ask a question; they’re calling because the question resisted automation. Their frustration is pre-loaded. And this is the moment when your brand is actually decided — not on the 70% the bot handled identically to your competitor’s bot, but on the 30% where the outcome depends entirely on how you built for failure.

Companies that treat the escalation as an afterthought — a shrinking cost center staffed by whoever survived the layoffs — will discover that they’ve automated their way into indistinguishability on the easy stuff and inferiority on the hard stuff. That’s a worse competitive position than they started with.

Cutting 10% of agents is easy. Deciding what the other 90% can do is the strategy.

Notice what none of this week’s announcements included: any detail about how the remaining humans are equipped. Uber said it cut 10% of customer service jobs. Fine — but the strategic question is what the other 90% can now do that they couldn’t before. If the answer is “the same job with more difficult tickets,” that’s not a transformation. That’s a slow-motion CX failure with a good quarter attached.

The winning version of this transition treats the human tier as the product, not the overhead. That means three things:

• Speed to the right human. If the bot fails and the customer then waits 40 minutes for a person, you’ve compounded the failure. Escalation paths need to be as engineered as deflection flows.
• Context that survives the handoff. The human should never make the customer repeat what the bot already knows. This is table stakes, and most companies still fail it.
• Resolution tools that match the residual problems. This is the one almost everyone misses — because the residual problems increasingly aren’t text-shaped.

The hardest escalations are physical, and you can’t type your way through them

Look at what actually survives deflection: the router with the blinking amber light, the appliance making a noise the customer can’t describe, the medical device that won’t pair, the installation that “doesn’t look like the diagram.” These problems escalate precisely because they live in the physical world, where a customer’s written description is the lossiest possible interface. The customer isn’t a trained observer. They don’t know which of the four cables matters. The bot failed for the same reason a phone agent will struggle: nobody can see the problem.

The traditional answer is to dispatch a technician — which is why I’ve argued that the first decision in any support escalation should not be “send a tech.” A truck roll for a problem that a competent agent could resolve in five minutes of looking is the most expensive apology in business.

The better answer is to close the visual gap at the moment of escalation. This is the layer we built Viewabo for: the agent sends a link, the customer taps it, and their smartphone camera becomes the agent’s eyes — no app download, no friction stacked on top of an already-frustrated customer. The agent sees the blinking light, the cable, the error screen, the physical thing the bot could only ask about. The escalation that was heading toward a truck roll or a third callback becomes a single resolved conversation.

That’s not a chatbot alternative. It’s the piece the chatbot stack structurally cannot provide, sitting exactly where the competition is moving.

The differentiation window is open right now

Here’s the irony of this week’s headlines: the more aggressively the market rewards AI support adoption, the faster it commoditizes, and the more valuable the failure-handling layer becomes. Everyone racing to deploy the same stack is collectively guaranteeing that the stack won’t be anyone’s edge.

Forrester says half of support jobs disappear by 2030. Maybe. But the half that remains will handle the tickets that decide renewals, reviews, and reputations. The companies that win won’t be the ones with the best bot — everyone will have a good bot. They’ll be the ones whose humans, at the exact moment the bot gives up, can see the problem, resolve it fast, and turn the AI’s failure into the best interaction the customer had all year.

Your AI stack is your competitor’s AI stack. Your escalation is yours alone. Invest accordingly.