The people who killed the form are coming for the rest of your funnel.

In 2017, a company called Drift told B2B marketers to do something uncomfortable: kill the lead form, stop gating content, stop making buyers wait, and let them talk to you in real time or lose them. Drift hit a billion-dollar valuation by 2021, put conversational marketing on 150,000 websites, and sold to Vista Equity Partners for $1.2 billion. The form didn’t actually die, but the idea that a buyer should fill out seven fields and wait two days for a follow-up email started to feel broken in a way it hadn’t before. 

Now the people who built that are back, and they’re not building chatbots this time. They’re building autonomous AI agents, and they just landed inside one of the largest marketing platforms in the world. 

On August 5, Klaviyo announced it is acquiring Agency, a three-year-old AI startup with 25 employees and $32 million raised from Sequoia, Menlo Ventures, and Felicis. Agency built a product that automates the onboarding, follow-ups, upselling, and support work that human customer success managers do, and its founder, Elias Torres, becomes Klaviyo’s Chief Product Officer as part of the deal. 

Torres’s resume is the reason this isn’t just another acqui-hire press release. He co-founded Performable, which HubSpot acquired in 2011, then co-founded Drift and ran its engineering for eight years through the Vista exit. After that he consulted for OpenAI, working with enterprise clients like the NBA and Live Nation, and that experience led him to start Agency in 2023. The man has built for the gap between “someone showed interest” and “someone from sales got to them” twice now, and the third time around, he doesn’t want a human in the loop at all. 

The personal thread makes the story stranger and better. Torres hired Klaviyo CEO Andrew Bialecki at Performable around 2010, right out of Harvard, and mentored him through his first startup job. Bialecki went on to bootstrap Klaviyo into a $9.2 billion IPO, and Torres was an angel investor in the 2015 seed round. Bialecki’s framing of the reunion: “It’s the next Big Tech revolution: agents. Let’s get the band back together, and let’s go build.” 

Marketing platforms acquire AI startups every week now, and most of those deals are talent grabs in nicer packaging. This one read differently because Klaviyo already has the products. Earlier this year it launched Composer, an AI agent that builds complete marketing campaigns from a plain-language prompt (audiences, copy, channel coordination, send timing), and Customer Agent, which handles post-sale support across web chat, email, SMS, WhatsApp, and RCS. Customer Agent is resolving 65 percent of queries automatically for some brands, and one customer, Naked Wardrobe, hit 86 percent autonomous resolution within 90 days. Torres isn’t arriving to build version one; he’s arriving to run a product line that already works, with a mandate to scale it across 200,000 businesses sitting on 9 billion customer profiles and 250 trillion quarterly data points. As Torres told Klaviyo’s newsroom: “By joining Klaviyo, I get the other half, an understanding of consumer behaviour that nobody else has, and the ability to scale faster.” 

What changed most, though, is Torres’s own thinking about what he’s building and why. Drift was a very good tool that changed a category, but it was still fundamentally a tool: it sat on your website, engaged visitors, and routed them to sales, and a human still had to show up and close. His position now is explicitly post-tool. In a Sequoia podcast he said, “I just want to tell people what to do one time and they do it 10,000 times,” and he dismisses the current wave of AI assistants that summarise your meetings, because if the output still requires a human to interpret it and decide what to do, the bottleneck hasn’t actually moved. Agency’s product wasn’t a chatbot or a dashboard. It was an agent that analysed emails, CRM data, chat logs, and call transcripts, then took action on its own: scheduling, following up, flagging churn risk, without waiting for someone to review a report first. 

His standard for AI accuracy is equally telling: “If it’s doing 10 percent more than I wasn’t doing before, I’m extremely happy.” Not perfection, just whether it does something that previously wasn’t getting done at all. That reframe is worth sitting with if you run demand gen, because the debate in most marketing teams is still whether AI tools are accurate enough to trust, and Torres is arguing the question itself is wrong. The right question is whether the alternative, a human who doesn’t have the bandwidth to do the work, is actually better. 

Klaviyo is a B2C platform today, and its customers are mostly ecommerce brands, but the agent architecture Torres is building doesn’t particularly care about that distinction. An autonomous agent that manages post-sale relationships, spots expansion signals, and runs follow-up sequences without being asked is exactly what B2B customer success and account-based marketing teams need, and currently stitch together from three or four tools that don’t share data. The gap to watch is the middle of the funnel: Composer already builds campaigns, Customer Agent already handles what comes after the sale, and the qualification, nurturing, and handoff that demand gen teams spend most of their time managing sits right between them. Torres has built for that exact surface before, with chatbots at Drift, except the thesis now is to have an agent handle the whole thing, thousands of times simultaneously, without a human managing the queue. 

The people who built the last generation of B2B marketing infrastructure (conversational marketing, marketing automation, intent data, revenue intelligence) are rebuilding it around agents instead of workflows. The shift isn’t AI-assisted, it’s AI-executed: not tools that help your team work faster, but agents that do the work your team never gets to. Whether that vision lands in full by 2027 or 2029, the direction is already set, and the man who told you to kill the form is now building the thing that replaces what came after it.