Chat inefficiencies

There was another uncomfortable truth unearthed in Lendi’s agentic journey: Chat is not the future of most customer experiences. Lendi needed to redesign Guardian to be more like a guided cockpit, not a conversational free-for-all.

Stewart explains Guardian begins with a dashboard precisely because customers should not be “thrown straight into an AI journey that they’re not familiar with or prepared for”. The real agentic layer operates beyond the dashboard, where the system scans, triggers, suggests and orchestrates.

Lendi’s Rate Radar, for example, “scans the customer’s rate 24/7… looks at our product catalogue and determines if there’s a better offering available. You can link that with open banking so it’s to the minute accurate,” says Stewart.

Instead of AI chat, this is what Hyman would characterise as agentic motion by design, where the system watches and acts by default. It is a system of nudges, widgets, dashboards, suggested actions and proactive interventions. In short, the UI becomes the agentic system. But Lendi also built the internal machinery to continuously rewrite it.

During a demonstration, Stewart showed how “as a product manager … I can come in here, change the model, change the tools, change everything without having to do any code deployments.” 

“You can see a customer behaviour in production, realise you need to address it and have it updated within 10 minutes,” he says.

Agentic motion

“Agentic motion” is the strategic endgame, according to Hyman,  yet most companies live in “human motion”.

“Humans basically are required to move everything along … whether it’s answering emails, making calls, progressing support tickets, or pushing revenue forward. Nothing progresses until someone nudges it,” he says.

Companies respond by bolting AI onto that motion. They get 5 per cent or 10 per cent improvement and call it a transformation.

“But really the main game,” Hyman says, “is this shift from human motion to agentic motion.” 

Per Hyman, agentic motion is where “agents … are watching; they’re acting and they’re reacting by default.” Watching content creation. Watching the customer pipeline. Watching issues. Driving revenue. Doing the busy work. 

“What that allows people to do… they’re there for judgment, they’re there for empathy… relationships… high value work,” says Hyman. “This isn’t just a technology problem. This is really a people problem,” Hyman says. And the winners will be those who “lift the gaze … raise the ambition.”

In an agentic world, marketing stops behaving like a series of moments and starts behaving like a system in motion. Customer engagement doesn’t arrive in neat, episodic bursts anymore such as a launch here, a promo there. Instead, it becomes continuous, always-on, and responsive to what the customer is doing right now.

And that shift changes what conversion is. It’s no longer something you “run” as a campaign with a beginning, middle and end. It becomes more like a behavioural system with an ongoing pattern of nudges, interventions and learning loops that adapt as signals change and intent flickers.

Indeed, that’s a theme that repeatedly emerged last year as Mi3 was preparing its Agentic AI report called Inside the Tornado.

As the report noted, “For the better part of a century, marketing has moved to a familiar beat: Plan it, buy it, blast it, measure it. Rinse. Repeat. The campaign cycle wasn’t just a process; it was muscle memory. Now the tech sector is taking huge bets that the rhythm is breaking. And the replacement they are building doesn’t tick. It pulses. Agentic martech, its advocate says, will observe, adapt, and act in real time. Every interaction will be a signal. Every signal will be a lesson. And the learning will never stop.”

Even the brand promise stops living in language alone. Instead of being something you declare in a message, it becomes something you perform, like a loop of actions that repeatedly proves (or breaks) trust in real time.

A radical lesson

There’s one more episode worth lingering on because it contains the human heartbeat of the story. When Lendi’s first agent “employee” failed, Lendi didn’t hire a prompt engineer to reprogram it, they asked a 23-year-old sales gun to train it.

“We actually took one of our 23-year-olds [and asked]: ‘teach us how to sell, teach the agent how to sell,” Tyler says. “And he’s now our best prompter in the business.” 

In the agentic era, competitive advantage required codifying excellence and distilling frontline compulsion into repeatable behavioural patterns, according to Lendi’s experience

This is the real opportunity revealed from the firm’s dance on the cutting edge: AI needs to scale craft as much as content. And not just volume, but effectiveness.

So what, ultimately, did Lendi learn from its early foray and that 270-tool monster?

That agents should not be overloaded. That chat should not be worshipped. Guardian only started to move when Lendi wrapped the conversation in deterministic scaffolding, such as widgets, follow-ups, tools, and stopped asking the model to do dumb machine work like database writes. That shows deterministic software still matters, and that sales instincts must be embedded.

To that, you can add the idea that governance is non-negotiable, and organisations must unlearn before they can build. And that sometimes, the best prompter in the company isn’t the one with 20 years of experience, it’s the 23-year-old who knows how to close the deal.