Selling Is Easy. Serving at Scale Is Where Companies Break.

01 /The Concept

What makes today's startup immediately interesting is who founded it: someone who sold his previous company for $1.2 billion in 2021. That means he saw potential in this new space worth at least as much.

Also interesting: he set out to solve his own problem. He wants the headcount at his new company to never exceed 100 people – while generating at least $1 billion in annual revenue.

Before founding the company he later sold, he served as VP of Engineering at HubSpot, where he watched it scale from 600 to – eventually – 7,000 employees. That's precisely the trajectory he wants no part of.

And he doesn't just intend to achieve this ratio himself – he plans to help others do the same. That's what Agency is built for.

The startup's core thesis is counterintuitive: "Closing deals is easy. Serving customers is hard." Most founders would say the opposite.

The difficulty, in the founder's framing, is that closing a deal is a one-time event. But serving that customer afterward is a perpetual, 24/7 obligation.

It's a bit like having a child. The initial act of persuasion might take one conversation. The hard part is the years of raising and developing what comes after.

The problem with customer service at scale has two dimensions:

- First, headcount grows proportionally. More customers means more account managers and support reps.

- Second, the complexity and volume of service-related processes increase. Managing those requires more managers, engineers, and operations staff.

At some point, the growth in support headcount and operational overhead stops being proportional to customer count – and goes exponential.

Giving everyone an AI assistant helps at the margins. Each rep can handle more customers, so headcount growth slows. But the fundamental dependency between customer count and headcount never disappears.

The only real solution to the customer service scaling problem isn't giving humans AI tools. It's having AI own the entire customer service function – autonomously, without humans in the loop at all.

That's what Agency built: an AI agent called Kai, capable of handling the full scope of customer service in a fully autonomous mode.

Kai can:

- Onboard new customers – explaining not just the product in general, but how each specific customer can use it for their particular goals.

- Upsell additional products – proactively, when Kai identifies an opportunity based on how the customer is using the product, not just in response to requests.

- Renew contracts – negotiating price with the customer based on documented value delivered.

- Proactively reach out – when Kai detects signals that a customer is using the product less or may be at risk of churning.

- Deliver real-time customer health reports – synthesizing over 1,000 signals about current and predicted customer behavior, contract status, and renewal probability.

Agency was founded in 2024 and raised $12M immediately. It has now announced the release of Kai and a new $20M round. During the pilot period, Agency's customers have already onboarded 35,000 of their end users onto the platform.

02 /Why It Matters

The Agency founder noticed something striking: technology has actually made customer service worse, not better.

In a small neighborhood shop or a local repair café, you typically get excellent, personalized service. The staff knows you, knows your preferences, and offers you what you're likely to want. If something goes wrong, even the owner steps in.

This pattern plays out across every business. When it's small, the founders personally attend to every customer. As it grows, individualized relationships get replaced by "corporate communications" managed through technology platforms – and personalization becomes little more than mail-merge name insertions.

The paradox: small businesses deliver high-quality customer service without any technology at all. The moment technology enters the picture, customer service becomes impersonal, hollow, and ineffective.

Part of the reason is that today's customer service software is, to borrow a phrase, a horseless carriage. The first automobiles were literally carriage bodies without the horses – all the awkwardness of the old form factor without actually rethinking the design.

Same story in customer service. Companies handed their support teams new tools but left humans in control of everything. The carriage is the same; the engine is slightly newer.

There's a second paradox: companies desperately try to standardize their customer service. Two hundred support reps don't each communicate in their own style – the company forces them into a single template, a single tone, a single process.

So on one hand, customer service is left to humans. On the other, those humans are made to operate like robots executing predefined scripts.

Why not take the obvious next step and hand the whole thing over to actual robots – replacing horseless carriages with actual automobiles.

Leaving, of course, maybe 10% of communications with your largest accounts in human hands – the relationships where genuine personalization and personal touch reinforce the value of the contact.

Agency isn't alone in observing the degradation of customer service at scale. Magic ([related review](/review/v-sfere-uslug-samoe-glavnoe-jeto-otnoshenija)) raised $10M in October to build a personalized customer service platform for physical venues – restaurants, cafes, shops, hotels. Its goal: convert a first-time visitor into a repeat customer, and keep them coming back as long as possible.

03 /Opportunities

The vision of AI fully owning customer service – as Agency paints it – is genuinely appealing. But most of us have also been stuck in infuriating loops with AI support bots that couldn't answer a simple question.

That gap between aspiration and current reality is real. But it's also the best possible moment to enter the market.

Because the technology will keep improving. And if you start building now – signing early customers, establishing a foothold, accumulating domain expertise – you'll be in exactly the right place when the capabilities catch up. If you wait until the technology is good enough, the distribution will already be locked up.

So the first direction: AI platforms for fully automated customer service.

This is the path to companies genuinely achieving billion-dollar revenue with a hundred, ten, or even one employee – a scenario being discussed seriously today, not as a thought experiment but as a near-term timeline question.

And customer service is only the start. A lot of adjacent processes need automation too – and both point toward the same underlying architecture: AI that owns an entire business function end-to-end, not just assists the humans who do.

Lancey ([related review](/review/chtoby-povysit-jeffektivnost-v-10-raz)) launched a new platform in November that analyzes user feedback on software products, suggests what to build next, and then writes the code itself – leaving the human team to simply choose what to ship.

Rocketable ([related review](/review/v-obshhem-sluchae-jeto-poka-fantastika-a-v-chastnom-vozmozhnost-na-milliard)) raised $6.5M to build an entire software holding company where AI autonomously develops and maintains the products. The only human-touch point: the holding acquires products from human developers who built them, proved demand, and demonstrated they could generate revenue.

So the second, broader direction: build startups and platforms where AI handles an entire chosen domain in fully autonomous mode. Even if the technology isn't quite there yet – that's a matter of time.

What else could be fully automated with AI?