Kieran Flanagan on Why Growth Teams Will Disappear and AI Will Own GTM

The Growth Team Is Dead (And AI Killed It)

Kieran Flanagan believes the growth function as we know it is headed for extinction. Not because growth doesn't matter, but because AI is collapsing the artificial boundaries between product-led and sales-led motions so quickly that the traditional division of labor no longer makes sense.

At HubSpot, where Flanagan runs growth and AI implementation, he's watched his own playbook start to dissolve. Growth teams spent years perfecting tooltips, onboarding flows, and upgrade paths—the mechanics of touchless conversion. Sales teams owned the human-led pipeline. Customer success managed retention. Support handled tickets. Clean lines, clear owners.

Then multimodal agents arrived. An AI that can converse via text or voice, see your screen, guide you through onboarding, answer support questions, and close a deal doesn't respect org charts. "Who owns that experience?" Flanagan asks. "What's the handover? What part does growth own? What part does customer success own? What part does customer support own? What part does sales own?"

An AI innovation pod should be the pod that's trying to integrate that experience across your go-to-market.

— Kieran Flanagan

Flanagan's answer: a single team—call it an AI innovation pod—that thinks across the entire go-to-market, from first website visit to renewal. Growth teams traditionally avoided the sales-assist layer. They optimized the self-serve funnel and left the humans alone. But when your onboarding agent is also your sales agent is also your support agent, that division crumbles. The skill set required is the same: ingest data, build prompts, ship experiments, measure outcomes.

The Data Layer Is the New Moat

Flanagan is blunt about what separates AI GTM that works from AI GTM that's vaporware: data. Not just structured CRM fields, but the messy, unstructured stuff—sales call transcripts, chat logs, support tickets, developer docs. HubSpot's AI chat agent now books meetings and deflects support inquiries. The step-change improvements came when the team fed it development documentation and external API references.

The same pattern showed up in email personalization. HubSpot ran experiments where AI-generated emails delivered triple-digit conversion rate lifts. The secret wasn't a single magic prompt. Flanagan's team broke each email into discrete components—subject line, intro, body, close—and wrote separate prompts for each. That modularity allowed rapid iteration. Every new data source they plugged in—behavioral signals, firmographic context, past interaction history—produced another spike in performance.

If you had told me 12 months ago that we could use AI to 5x the conversion rate on email, I'm like, oh look, there's only so much personalization you can do in an email.

— Kieran Flanagan

The implication: off-the-shelf AI SDR tools struggle because they're trying to write prompts that work for hundreds of customers. Custom prompts, trained on proprietary data, win every time. Flanagan wonders aloud whether third-party AI tooling can ever match the performance of in-house builds when the competitive advantage lies in how well you know your own customer.

He recommends a two-layer architecture. First, an AI ops team that ingests and normalizes all your structured and unstructured data. Then, a team that sits on top of that data layer and ships AI experiences across the full customer journey. The foundation is everything.

Where AI GTM Actually Works (And Where It's Still Smoke)

Flanagan has run enough pilots to know which AI GTM use cases are real and which are hype cycles. Customer support agents work. They're reliable, they scale, and they free human support reps to do white-glove onboarding for high-value accounts. Email personalization works. HubSpot is now automating prospecting that SDRs used to do manually, with higher conversion rates.

Chat-based selling works, too. HubSpot's AI chat agent handles both support deflection and sales qualification. The key is intent switching—the agent detects whether you're asking a technical question or evaluating a purchase, then adjusts its behavior. That means the team that used to staff live chat for support can now be redeployed as a sales function, because the AI handles first-touch triage.

The hype? Autonomous multitask agents. Flanagan is skeptical. Agents need data, tools, and context to function, and even the best models—he singles out Claude Opus as the most promising—are still inconsistent and unreliable when asked to complete complex, multi-step goals without human oversight. He's spent time with external agent-first startups and seen the same problem: they work in demos, but they don't work predictably enough to run unsupervised in production.

I don't think that I have spent or seen a lot of great agent-first companies be as reliable and consistent as they need to be.

— Kieran Flanagan

The lesson: narrow, single-task agents that do one thing well will ship first. Broad, autonomous agents that replace entire workflows are still a few model generations away.

The Labor Budget Question No One Wants to Answer

Flanagan tries to be optimistic when he talks about AI and headcount. He offers the standard line: AI gives you time back, and you can redeploy humans to higher-value work. Creative storytelling instead of clip production. Strategic account management instead of tier-one support tickets. But when pressed, he admits the uncomfortable truth. "If we are being really honest with each other, teams are going to be much, much smaller."

HubSpot hasn't cut its support or SDR teams. It's redeployed them. Support reps move up-market to onboard enterprise customers. SDRs focus on accounts the AI can't close. But Flanagan believes AI tools will eventually migrate from software budgets to labor budgets. Companies will price agents against the cost of a human doing the same work. That's when the real displacement begins.

He frames it as a choice: use AI to improve unit economics and shrink teams, or use AI to expand coverage and grow faster with the same headcount. Smart companies will choose growth. But not every company will make that choice, and not every role will survive the transition.

I think we're going to price agents against human labor costs, outcomes.

— Kieran Flanagan

The venture capital calculus is simple. If AI tools eat into labor budgets, software multiples stay high. If they just shuffle software spend around, the market rerates. Flanagan sees the former happening. That's why he thinks the growth team of the future looks less like a team and more like a pod of AI infrastructure builders.

What to Do If You're Starting from Zero

For early-stage founders building GTM teams today, Flanagan's advice is clear: don't start with separate growth, sales, and support functions. Start with an AI ops layer that can ingest and structure all your data. Then build a single team on top of that layer, focused on deploying AI across every customer touchpoint. Whether you call it growth or AI innovation or something else doesn't matter. What matters is that the team can think across the entire funnel, from acquisition to support, without handoffs.

He acknowledges this is easier to say than to execute at a company like HubSpot, where growth teams have years of institutional knowledge and strong product-led muscle. But for teams starting fresh, without organizational debt, the path is obvious. The growth playbook of the last decade—optimize onboarding, reduce friction, instrument everything—still applies. The difference is that the team running that playbook needs to be fluent in prompt engineering, data pipelines, and agent reliability, not just JavaScript and SQL.

Flanagan hasn't disbanded HubSpot's growth team. But he can see the horizon. As multimodal agents get better, as AI chat becomes the default interface, as the line between self-serve and sales-assist blurs into irrelevance, the organizational structure that made sense in 2018 will stop making sense. The companies that adapt fastest will be the ones that stop asking which team owns the AI experience and start asking how to make the AI experience seamless across every stage of the customer journey.

Growth teams could evolve into this AI innovation pod. If we just remove names, if we think about AI and how AI is going to benefit go-to-markets, the ability to ingest all of this data and have a team that sits on top of that, that is able to innovate on how they actually use that data to make using your product much easier, to make extracting value much easier, to make selling your product much easier, to make supporting your product much easier.

— Kieran Flanagan

The growth team is dying. Long live the AI innovation pod.