Agentic AI. Everywhere.
That’s the headline from Zendesk Relate 2026, if you couldn’t guess from the incessant signage screaming “Hard launch agentic AI.” Look, I’ve spent twenty years watching these tech parades, and frankly, the same old song and dance gets tiresome. But sometimes, buried beneath the glitter and the buzzwords, there’s a kernel of truth – or at least, a clue about who’s actually lining their pockets.
This year, the big reveal was Zendesk scooping up Local Measure, a company that essentially built a full contact center as a service (CCaaS) layer on top of Amazon Connect. The promise? Bringing that native channel depth directly into Zendesk’s ecosystem. Translation: Making their existing customer service platform play nicer with Amazon’s cloud telephony. Seems sensible, if not exactly world-shattering.
The Slow Death of Overlays
One of the more interesting — and I use that term loosely — points from the conference was the pronouncement that “Overlays are dying.” For the uninitiated, an overlay is basically a custom-built agent softphone that bolts onto Amazon Connect, adding features that AWS itself either hasn’t bothered to implement or has lagged on. Think of it as a fancy Band-Aid for the Amazon Connect Contact Control Panel (CCP). For years, these custom solutions were a cash cow for integrators, especially for companies that didn’t have a deep CRM integration and needed specific bells and whistles.
But here’s the kicker: AWS, in its infinite wisdom, hasn’t kept feature parity between its Streams SDK and its Agent Workspaces. This means those custom overlays are constantly playing catch-up. When you factor in how deeply CRMs like Salesforce are embedding their own contact center capabilities, the writing is on the wall. Why build a Frankenstein’s monster of an agent desktop when your CRM can already do most of it natively? The ROI on these overlays is dwindling faster than a free donut at a morning stand-up.
Average Handle Time: A Misleading Metric?
The conference also coughed up the chestnut that Average Handle Time (AHT) is actually going up, not down. The explanation? Bots are hogging all the easy queries, leaving human agents to deal with the messy, complex stuff. So, sure, your AHT might climb from 4 minutes to 7 minutes. But if your bot containment rate has also jumped 40%, then that longer human interaction might just be a sign of success, not failure. The real problem, as one speaker (probably trying to sell something) pointed out, is that most contact center managers are still using AHT as a blunt performance instrument, missing the forest for the trees.
This brings up another gap: how do you even measure bot performance? Traditional metrics like AHT, Average Speed of Answer (ASA), and First Contact Resolution (FCR) simply don’t map well onto automated interactions. A bot that takes 45 seconds to resolve an issue and then transfers the customer isn’t the same as a bot that fully resolves it. FCR becomes a fuzzy concept when an interaction is split between an IVR, an IVA, and a human. The industry, as usual, is scrambling for answers while the tech marches ahead.
Queue Management Chaos
Queue management, apparently, is still a festering wound in the contact center world, and Amazon Connect is no exception. With its expanding routing capabilities – things like proficiency-based routing (scoring agents on skills) and agents picking their own calls from queues – things get complicated. The more granular you make your routing logic, the harder it is to predict what’s actually happening in the queue. And, surprise surprise, more complexity means more opportunities for inefficiencies and, dare I say it, abuse. Who’d have thought?
And then there’s the whole QA (Quality Assurance) issue for automated systems. As bots handle more volume, evaluating their performance becomes as critical as evaluating humans. Was the right intent recognized? Was the bot’s resolution path appropriate? Was the handoff to a human agent smooth? Most existing QA tools and training were built for humans, not for AI. It’s an emerging need that most companies are blissfully unprepared for.
Post go-live support is underserved – real opportunity here. Going live on a CRM is a different problem than going live on a channel that supports real-time communications. A CRM without a live channel integrated has human latency baked in – agents are switching tabs, copying notes, managing context manually.
This quote perfectly encapsulates the often-overlooked friction point when integrating real-time communication channels. Suddenly, all those little inefficiencies you didn’t notice with a siloed CRM become glaringly obvious. Customers often blame the new integration, but in reality, it’s just shining a spotlight on pre-existing problems.
The Real Value: Where’s the Money?
So, where’s the actual money changing hands here? It’s in smoothing out these rough edges. Zendesk’s acquisition of Local Measure is about consolidating its position by offering a more integrated experience with a major cloud player. The persistent theme of agentic AI, while heavily marketed, is really about pushing customers towards more sophisticated (and expensive) automation that, in turn, requires better integration and management. Who benefits? Zendesk, by selling more licenses and services. Amazon, by getting more customers locked into its cloud ecosystem. And the integrators who can navigate this increasingly complex landscape.
What struck me, though, was a brief mention of knowledge quality driving IVR/IVA retention. One customer apparently saw 80–90% containment rates solely due to the quality of their knowledge base. Not the bot’s sophistication, not the AI model. The knowledge. It’s a stark reminder that for all the AI wizardry, sometimes the most effective solution is the simplest: good, clean data. Companies are spending fortunes on AI, but are they investing enough in the foundational knowledge that makes it work? My bet is no.
Is Agentic AI Just Hype?
From where I’m sitting, agentic AI is the latest iteration of a perennial promise: make customer service cheaper and more efficient. The ‘agentic’ part implies a level of autonomy that we’re likely still a long way from achieving at scale without significant human oversight. The real innovation here, from a business perspective, isn’t some magical AI agent; it’s the integration, the data plumbing, and the ongoing services required to make these systems hum. The revenue is in the complexity and the ongoing support contracts, not necessarily in the AI itself suddenly becoming sentient and solving all our problems.
Zendesk’s moves, and the trends at Relate 2026, signal a clear push towards deeper integration and more strong automation. But for all the talk of AI, it’s the messy reality of contact center operations – the metrics, the queues, the human element – that still defines the market. And where there’s messy reality, there’s opportunity… for someone to sell you a solution.
🧬 Related Insights
- Read more: GitHub’s Stacked PRs: Phabricator Nostalgia or Just Better Dev?
- Read more: Daily Briefing: April 25, 2026
Frequently Asked Questions
What does Zendesk Relate 2026 mean for Amazon Connect users?
It signals a continued push for deeper integration and potentially higher-value services built on top of Amazon Connect, especially for users prioritizing native CRM experiences over custom overlay solutions.
Will agentic AI replace human customer service agents?
Not entirely, at least not soon. While AI will handle more routine tasks, human agents will likely focus on more complex issues, with AI assisting them rather than outright replacing them across the board.
How important is knowledge base quality for AI retention?
Extremely important. One example shared indicated that a high-quality knowledge base was directly responsible for 80-90% containment rates, highlighting its foundational role for AI effectiveness.
