Igor Nestorovic, CCXP

Head of Customer Operations | CX & AI-Enabled Service Operations | Operating Models, Automation & VoC | CCXP

[Igor Nestorovic, CCXP @inestorovic]

AI is removing some of the routine work through which people used to learn the job.

That means new hires can encounter the hard cases earlier, with fewer low-risk repetitions behind them.

The career ladder hasn’t disappeared. It starts higher. https://t.co/2gCE26Cwz4

[Igor Nestorovic, CCXP @inestorovic]

Banks are introducing AI FinOps teams as agent-to-agent workflows drive up computing costs.

Before translating automation into staffing savings, measure the full cost of resolution, including retries, supervision and escalations.

Via @Reuters:

https://t.co/Ug2xsvV64c

[Igor Nestorovic, CCXP @inestorovic]

.@Stanford's data suggests the first rung is thinning fastest where AI substitutes for junior work.

But where AI complements people, the picture looks different.

That distinction matters.

The question isn’t only what AI can do.

It’s what people still get to learn by doing.

https://t.co/Kh6qUKJK5V

[Igor Nestorovic, CCXP @inestorovic]

AI is removing some of the routine work through which people used to learn the job.

That means new hires can encounter the hard cases earlier, with fewer low-risk repetitions behind them.

The career ladder hasn’t disappeared. It starts higher. https://t.co/2gCE26Cwz4

[Igor Nestorovic, CCXP @inestorovic]

@simonw This feels like a useful design principle for AI-assisted work: keep the evidence available, but put the human decision surface first. More detail is not the same as more clarity.

[Igor Nestorovic, CCXP @inestorovic]

@swyx Interesting implication for AI at work: if the scratchpad is part of thinking, automating the first draft may save time while removing the place where judgment develops. The best workflow may keep the messy reasoning visible, not just the polished answer.

[Igor Nestorovic, CCXP @inestorovic]

Hello World from @fedora 45 #Linux

[Igor Nestorovic, CCXP @inestorovic]

.@Gartner_inc : only 4% of surveyed customer service leaders achieved their expected AI ROI.

Automation creates capacity. Capturing its value requires decisions about exceptions, rework, staffing and where that capacity goes.

The business case has to follow the work.

https://t.co/HIxAsRZ1d9

[Igor Nestorovic, CCXP @inestorovic]

@DhruvJain08 That's the critical question. Someone experienced still needs to catch that first hard case, but we can't rely on informal escalation forever. Coaching, supervised decisions, and structured feedback need to become part of the operating model.

[Igor Nestorovic, CCXP @inestorovic]

AI is removing some of the routine work through which people used to learn the job.

That means new hires can encounter the hard cases earlier, with fewer low-risk repetitions behind them.

The career ladder hasn’t disappeared. It starts higher. https://t.co/2gCE26Cwz4

[Igor Nestorovic, CCXP @inestorovic]

@emollick The task horizon is impressive, but the 50% success rate is the operational reality. Longer-running agents don’t remove supervision; they change it. Someone still has to verify outcomes, handle exceptions, and remain accountable across days rather than minutes.

[Igor Nestorovic, CCXP @inestorovic]

.@Gartner_inc finds only 19% of organizations effectively account for AI in workforce planning.

The risk is deciding headcount before understanding the new work: exceptions, oversight, quality, and the judgment people still need to develop.

Plan the work first.

https://t.co/TfRsCHi6xg

[Igor Nestorovic, CCXP @inestorovic]

And once we automate the repetitions that built it, we may discover that judgment has a training cost we never priced.

[Igor Nestorovic, CCXP @inestorovic]

After automation, I’d add four questions to the dashboard:

Where did human effort increase?
What happened to case complexity?
What happened to customer effort?
What new work did automation create?

A smaller queue is useful.

It just isn’t a map of where all the work went.

https://t.co/u4nqc4rbL9

[Igor Nestorovic, CCXP @inestorovic]

AI handled 1,000+ support tickets at a fraction of the human cost.

Great result.

But the more interesting question came afterwards:

Where did the rest of the work go?

October’s Support Matters. https://t.co/DhUQNpDxYo

[Igor Nestorovic, CCXP @inestorovic]

@boardyai That may be the harder question. If the routine work disappears, we need another way for people to build the judgment the exceptions will still require.

[Igor Nestorovic, CCXP @inestorovic]

The interesting part is that this is not just an AI adoption story. They had to change what the company was willing to stop doing, who could make that call, and how the operating model supported it. The existing business made the transformation harder, and the result more meaningful.

[Igor Nestorovic, CCXP @inestorovic]

Visibility can expand almost immediately; decision rights and escalation paths usually do not. Senior leaders may see more of the firm without the organization becoming better at acting on what they see. That gap is where AI-enabled span of control can quietly become a new governance risk.

[Studenti_u_blokadi @studentblokade]

Студенти су победили.

[Igor Nestorovic, CCXP @inestorovic]

An AI agent can have a customer's password without having clear authority to act.

Banks are already encountering that gap.

Read-only access buys time. The harder work is proving intent, tracing actions and resolving disputes when something goes wrong.

https://t.co/dpcWcSWR78

[Igor Nestorovic, CCXP @inestorovic]

AI work doesn’t always reappear as another queue.

Sometimes it becomes:

exception reviews,
calibration,
policy decisions,
QA,
cross-functional coordination.

Ten minutes here. Fifteen there.

Eventually those ten minutes become a job.

https://t.co/MfyL5AoPUf

[Igor Nestorovic, CCXP @inestorovic]

AI handled 1,000+ support tickets at a fraction of the human cost.

Great result.

But the more interesting question came afterwards:

Where did the rest of the work go?

October’s Support Matters. https://t.co/DhUQNpDxYo

[Igor Nestorovic, CCXP @inestorovic]

@somi_ai @mlpowered @lennysan This is a useful reminder that what looks like hallucination can sometimes be faithful optimization against a bad signal. In operations, we’d call that a control-design failure: the system did what it was rewarded to do, just not what we intended.

[Igor Nestorovic, CCXP @inestorovic]

@chrisalbon @ramez The diminishing-returns point part I find most useful. Even if model capability keeps improving, real-world outcomes are still constrained by data quality, authorization, integration and an organization’s capacity to absorb change. Intelligence is only one bottleneck.