For Frack’s Sake, It’s Not Too Late to Ship an Epic AI Agent. And Get Back to Real Growth.

So I spent the other day writing a fairly dark post about how most traditional B2B leaders will slowly atrophy to nothingness. Maybe seven survivors, everyone else decays, Medallia as the cautionary tale, but just one of many:

That’s one side of the coin. Here’s the other.

It’s not too late. Not even close.

You can still ship an epic AI agent for your category, charge real money for it, and get back to growth. Some of you reading this can do it in 90 days. And if you do, you might be shocked at how fast the narrative around your company changes.

As tough as it may seem to capture the incredible growth of AI native leaders, the reality is … the LLMs are open to everyone. 95% of the AI B2B startups with explosive growth are using LLMs and shipping agents you could build too. In fact, you can even copy them now.

But let’s be honest about both sides of this.

First, Be Honest. There Are No Excuses at This Point. Everyone Has Had Time.

The LLMs have been good enough to build real products for 15 months. Claude 3.5 Sonnet shipped in June 2024. The October 2024 update was when everything got great for coding and agentic workflows.

By Claude 3.7 in February 2025 and the subsequent Opus 4 and Sonnet 4 releases in late 2025, the models weren’t just good. They were genuinely great for production use cases. Reasoning, tool use, long-context document analysis, multi-step agentic workflows. All there. All available via API for anyone willing to build.

And then everything got truly epic around Opus 4.5 at the end of 2025 and just went vertical.

So if you’re a B2B CEO reading this and your company hasn’t shipped a real, category-defining AI agent yet, I’m not going to pretend that’s okay. It’s not. You’ve had 15 months to work on it and at least since late 2025 to ship something truly magical. You’ve had access to the same APIs every AI-native startup used. And you have more customers, more data, more revenue, more engineers. And all you shipped … was a basic chatbot? Or a summarization feature? Or a copilot nobody uses?

That’s on you. That’s on your leadership team. That’s on your product org. Own it. It’s time to 100% own it.

But … 95%+ Of Your Installed Base Hasn’t Actually Gone Anywhere Yet

Not yet at least. Not 95%-99% of them.

If you have 1,000 customers, 5,000 customers, 10,000, 100,000+ customers… most of them are still just beginning their own AI journey. They’re not ahead of you. Most have not already deployed an AI-native alternative, outside of coding and support. Most of them have maybe experimented with a couple of AI agents, maybe had one or two pilot projects.

Look at the actual data. Redpoint’s 2026 outlook called it directly: 2025 was a year of unprecedented pulled-forward pilots. 2026 is when those pilots convert or quietly disappear. Gartner shows that even at the largest enterprises, 40% of budgets are being reallocated to AI, but the spending is still fragmented across pilots, infrastructure, and early deployments. The agentic AI market is growing 38-fold over the next decade, but most of that spending hasn’t happened yet.

95% of your customers are not further along than you are. They’re still figuring it out. They’re asking their vendors which agents to trust. They’re looking at their installed software and asking: “Why aren’t you helping me with this?”

That is a massive opportunity. It’s not over. It’s just beginning. The window is still open.

Why HubSpot AEO Product and Figma Make Are So Frustrating

Let me give you two specific examples of what I’m talking about. My frustration with these products isn’t just that they’re not great. It’s that they could be. Their customer bases are literally dying to buy great agentic products from them. And they’re fumbling the opportunity.

Take HubSpot new AEO tool. HubSpot has 300,000 customers. Every single one of them can benefit to some extent from AEO. Many of them would rather just buy from HubSpot. It’s one vendor. The data is already there. The CRM integration is native.

But it’s at best, 60% of what the start-ups in AEO are doing.

Now take Figma Make. I asked Figma Make to redesign the SaaStr AI site recently. It’s worse than anything any leading vibe coding platform would produce today. Generic AI gradients. Trite content. Purple hero sections that say “The Future of SaaS is AI-Powered.” It wouldn’t even pass muster at a demo day.

But Figma has the same opportunity as HubSpot, maybe bigger. They have millions of designers and PMs and engineers who already use Figma daily. Every single one of them wants to generate real, usable designs and code from prompts. They’re all using Lovable and v0 and Replit for this. Figma should own this market. It’s their customers. It’s their canvas. It’s their entire product category.

Instead, Figma Make is generic vibes-to-HTML with no understanding of what the product actually does. So Figma customers use Replit’s design mode, which at least has context because it’s working with real code and real components. And every day they use Replit instead of Figma is another day Figma’s moat erodes.

This is the pattern across so many incumbents right now. Massive customer bases. Real distribution. Unique data. And products that are 60% of the way there. Their customers would pay them first. Their customers want to pay them first. But the products aren’t good enough, so the budget goes to the AI-native alternative.

Why would you buy an AI SDR from a top startup if you could truly get the same today, for real, from HubSpot? You wouldn’t. You’d buy from HubSpot. But HubSpot isn’t giving you that option. So you go elsewhere. And once you go elsewhere, you’re gone.

Build the Best AI Agent for Your Category. Now.

Here’s what you should actually do, starting Monday:

Stop trying to ship 15 AI features across your product. Stop adding AI tags to existing functionality. Stop announcing your “AI-first strategy” on earnings calls.

Pick one agent. The single most valuable, workflow-changing AI agent that specifically serves your customers in the way your category demands. The one that automates the highest-pain, highest-volume work your customers are still doing manually. Build that one agent. Make it genuinely great.

Not 60% as good as what an AI-native startup would ship. Better. You have advantages they don’t: millions of rows of real customer data, years of workflow context, direct integrations into systems of record, a distribution channel that costs them millions to replicate. Use all of it. Pour your best engineers, your best product people, your best designers into this one agent. If that means pulling them off the core product roadmap for six months, do it. The antibodies will fight you. Fight back.

And charge for it. Real money. Usage-based. Outcome-based. Per-agent pricing. Whatever model fits, but charge for it as a first-class product, not a $10/month “AI tier” add-on. Customers will pay if the agent is great. They will not pay if it’s mediocre. 60% as good as Claude gets ignored. 110% as good as anything they could buy elsewhere gets bought.

Your Installed Base Is a Distribution Advantage, Not a Defensive Position

Here’s what AI-native startups would kill for: a list of 5,000 customers who already trust you, already have their data in your system, already use your product daily, and are actively asking how to use AI in their workflow.

You have that. They don’t.

When Salesforce rolled out Agentforce, they didn’t have to acquire a single new customer to hit 18,500 Agentforce customers and $540M+ in ARR. They just sold to the base. When HubSpot shipped Customer Agent and Prospecting Agent, same thing. When Shopify shipped AI merchandising and storefronts, the merchants were already there.

This is the play. Build the category-defining agent. Then activate it across your installed base. If 20% of your customers buy the new agent at even a modest uplift, you’ve added 10-20 points to your NRR in one cycle. That’s not a