Big Technology

Tony Bates on why AI-to-AI service still needs human rules

Tony Bates· Chairman & CEO at Genesys
·~35 min·English·Big Technology
AgentsBusiness StrategyAI Company
TL;DR

Genesys CEO Tony Bates argues that AI-to-AI customer service is inevitable, but it will run on the same governance, guardrails, and human hand-offs that already govern human agents — the operating model, not the model, is the hard part.

01Core Mental Model

The bar just moved to your own AI

Consumer chatbots now remember context and preferences, so customers expect every brand to know them the same way — and most brands' systems are too fragmented to deliver it.

if I can get that from my personal AI that's, you know, maybe sitting right on your laptop there, why can't I get that from the brand?

— Tony Bates, Big Technology
Key Insight
The threat to incumbents is not a competitor — it is the customer's own assistant. Once a personal AI sets the reference experience, a brand is judged against a tool the customer already trusts, not against its industry peers.
02The Framework

The road to AI talking to AI has five levels

Bates maps customer experience onto a self-driving-style ladder — from predefined phone trees, through today's agentic orchestration, up to “universal orchestration” where humans and AIs interact in every combination, including AI to AI.

you're going to see AI to AI. It's absolutely going to happen.

— Tony Bates, Big Technology
Key Insight
Framing progress as levels, like autonomy, lets Bates place today's product at “level four” and cast the top rung not as machines replacing people but as orchestration — human-to-human, human-to-AI, and AI-to-AI at once — a positioning that keeps a human in the loop and Genesys in the coordinating seat.
03Governance

AI agents play by the same rules as humans

An enterprise agent must sit behind the same guardrails a human rep does — what it can and cannot say, what it can see, when to escalate, and a full audit trail — so a customer's agent shouting “disregard all previous instructions” gets the same refusal a person would.

the same set of standards the same set of operating models will have to apply in exactly the same way that they do today when it's a human

— Tony Bates, Big Technology
Key Insight
By insisting the rules are unchanged, Bates reframes AI safety as an enterprise discipline regulated industries already run, not a new frontier: the hard part is not the model, it is the operating model around it.
04Operating Model

AI is a digital teammate you onboard and manage

Bates urges companies to treat AI as one more member of a single talent system — onboarded, given tools, and performance-managed exactly like a human information worker — rather than as a switch that replaces people.

AI is not your enemy. AI is your digital teammate, right? And the way that folks need to think about it inside their businesses is that it's one talent system. It's one set of resources.

— Tony Bates, Big Technology
Key Insight
Casting AI as a teammate you manage, not software you buy, turns onboarding, evaluation, and governance into ongoing management tasks rather than a one-time purchase — which quietly makes the durable product the orchestration layer above any single model.
05The End Goal

Personalization is a customer of one, not a persona

Bates draws a hard line between targeting — grouping people into personas and treating them the same — and personalization, which is understanding why one person wants something and handing off to the right human or AI at the right moment.

Personalization is understanding why you want something and giving it to you exactly the right moment and doing it in the right way.

— Tony Bates, Big Technology
Key Insight
The distinction changes the unit of optimization: targeting improves how you treat a whole segment, while personalization means choosing the next best action for one person from their context — including which channel or agent should answer, and when NOT to automate at all.
06The Interface

We are still in the MS-DOS moment

Typing or talking to a chatbot is, in Bates's view, a primitive “terminal expanded” built for a developer mindset — the friction-free, multimodal, embedded interface that would make universal orchestration feel natural has not arrived.

we're kind of in our MS DOS moment right now. We haven't had our Windows moment.

— Tony Bates, Big Technology
Key Insight
Admitting the interface is still primitive is also a convenient hedge: if AI-to-AI service is inevitable but the natural interface for it has not arrived, the disruption stays over the horizon — and today's incumbents keep their lead while it does.
07Humans in the Loop

Software cannot hold the moments that matter most

Bereavement calls, a crisis hotline, a nervous 83-year-old resetting a password — Bates argues these high-empathy moments should never be outsourced to software, so a hybrid workforce persists for a long time.

I don't think you want to hand that moment of truth to a piece of software.

— Tony Bates, Big Technology
Key Insight
Bates reframes the jobs question as a mix shift, not a headcount cut: routine contacts loom large today only because experience is bad, and fixing that frees people for the loyalty-driving moments that actually generate revenue.
08The Contrarian Take

The “SaaS apocalypse” misreads how enterprises work

Bates dismisses the idea that one model will ingest all data and replace software: large regulated customers will not hand their IP to a foundation model, and they demand governance, auditability, and resiliency a single ingesting model cannot provide.

they're not about to hand over their data to their foundation models. This is their IP.

— Tony Bates, Big Technology
Key Insight
His rebuttal mixes two kinds of evidence: a capability claim — no model can grasp every part of an enterprise — and a commercial one, that his own numbers show no seat contraction and AI as a tailwind. The commercial half is the load-bearing one: the moat is enterprise IP, governance, and trust, not only what models cannot yet do.