Why AI’s Second Act Is About Scaling Operations

Last year every board wanted to know one thing: did we adopt AI. This year the question got harder. Are we actually scaling with it, and what’s quietly falling behind in the parts of the business AI hasn’t touched yet.

That distinction matters. Adoption is a procurement checkbox. Scaling is a real operating decision. It touches headcount, compliance, and the muscle memory of teams who’ve run the same process for years.

McKinsey’s 2026 survey of 1,719 execs found 44% of organizations now say AI is scaling across the enterprise, up from 38% the year before. Here’s the number that actually matters: 80% of people say AI made them personally faster. Only 37% of companies can point to any real EBIT impact.

Translation: individuals are flying. Companies are not. I think about that gap constantly because closing it is basically my job.

Heading into 2027, ops is where AI scaling gets real. Four things I keep coming back to.

The COO’s Job Just Got Busier

People assume the COO role is pure execution. It’s not. The COO sits between the vision and the reality, and AI just made that space a lot more crowded.

Every team now has its own AI roadmap. Someone has to sit above all of it and ask if it actually adds up to something coherent, or if it’s just five separate bets running in parallel.

Misalignment was always expensive. AI makes it compound faster. My job is to catch friction before it hits a customer and keep execution honest when the easy move is to let excitement stand in for a plan.

This isn’t a brake on speed.

Design for Scale Before You Need It

Smaller companies love to push off process design because moving fast feels better. AI makes that excuse even easier to reach for.

A chatbot bolted onto a support queue looks like progress. It’s not. It’s just a faster version of the same undocumented mess. The real work is going in early and finding which 10-step process can become 3 steps without cutting anything important.

An agent handling renewals is only as good as the handoff logic underneath it. If the process was fuzzy when a human owned it, it stays fuzzy no matter what tech you throw on top.

Go slow to go fast. It’s a cliché because it’s true.

That doesn’t mean red tape. It means every workflow step has a clear owner and documentation tight enough that a model could follow it without guessing. And it means teams actually talking to each other instead of building in silos.

Get this right early and growth stops requiring headcount to grow at the same rate as customers.

Blindness Is the Real Risk, Not Risk Itself

Every growing company carries risk. The danger is not seeing it clearly, and AI changed what “seeing clearly” even means. McKinsey’s 2026 AI Trust Maturity Survey found nearly two-thirds of execs name security and risk as the top blocker to fully scaling agentic AI.

Part of my job is spotting operational risk before it becomes a headline. An autonomous agent is now one of those risks I have to think about like a person. It’s not a tool someone flipped on. It’s a decision-maker, and I’m accountable for its output even when I never personally reviewed it.

That means building approval frameworks and controls before an agent touches live customer data, not after the first incident forces the conversation. It means more cross-team conversations, especially around failure modes that are brand new to agentic systems.

Risk management was never about slowing down. It’s about making sure the business survives its own growth.

Automate the Boring Stuff First. CX Follows from That, It Isn’t Built Around It

Every process we evaluate for AI gets the same three questions I’d ask a new hire: who owns the output, what happens when it breaks, and how do we find out before the customer does.

That third question is the one most companies skip. It’s also the one MIT’s research quietly backs up. The highest ROI deployments weren’t in sales and marketing, where more than half of enterprise AI budget currently goes. They were in back-office work. Eligibility checks, reconciliation, documentation. The stuff that never makes it into an investor demo.

Customer experience gets treated like a support problem. It’s actually the sum of dozens of decisions the customer never sees, and that’s getting more true as AI moves into the service layer.

CX comes down to whether expectation matches the promise you made. The processes behind the scenes decide whether AI becomes an edge or a liability.

Leadership Is an Operating Rhythm, Not a Title

As companies scale, complexity outgrows headcount, and AI widens that gap even faster.

My job includes creating clarity inside that bigger surface area. Who reviews what an agent produces. Who has the authority to pull it back if a workflow starts spitting out results nobody approved.

That shows up in the operating reviews we already run, now with a standing line for what’s automated versus what’s human-reviewed. It also shows up in the harder conversations, like when one team wants to ship an agent faster than another team is ready for.

A solid operating rhythm always kept fast-growing companies from outrunning their own judgment. AI just raised the stakes of skipping it.

None of this argues against moving fast with AI. It argues against confusing motion with progress.

The companies that actually scaled with AI, twelve months from now, won’t be the ones that adopted first. They’ll be the ones that went slow enough to go fast, stayed honest about the impact on customers, and automated the unglamorous stuff before chasing the shiny stuff.

Scaling agentic AI is about more than technology. Operational excellence is what transforms its potential into measurable business impact.