There is an ever-widening gap between what the market says about AI and what we actually hear from customers. The media, VCs, AI labs, and influencers have all talked about AI replacing humans, eliminating trusted software, and token maxxing as ends worth pursuing. But leaders running real companies are increasingly asking the right questions. How can I improve my employees with artificial intelligence? Which systems can I trust? How can I measure the ROI of this expense? We hear these questions every day.
After three and a half years of building, shipping, and watching many of our growing customers put AI into practice, the AI prospects we’re most certain about at HubSpot are the things almost no one else says out loud.
Here are six.
AI activity does not represent AI results.
The industry has confused the movement for progress. Drafting emails, generating summaries, research activities. These are tasks that AI has made much easier. They’re useful features, and we ship them to HubSpot. But the activity is the input, not the outcome. Activity without results is theater.
The companies that win with AI are those that work backwards from a business problem, not forwards from a demonstration model. For example, customers using Customer Agent respond to tickets 25% faster, while those using Prospecting Agent generate 76% more leads.
This is why we moved Client Agent and Prospect Agent to performance-based pricing in April. What matters are the results of artificial intelligence. And they’re what we help growing companies build. We put our prices where our point of view is.
Artificial intelligence is necessary. It’s not enough.
Generating code is definitely easier now. Anyone can build a prototype in a weekend, but it’s fragile and falls apart in real-world use. Lowering the code generation cap doesn’t raise the shipping value cap because the things that actually run a growing business have gotten harder, not easier.
You still need clean data, not another silo. You still have to integrate with dozens of applications. You still need a complete view of the customer across marketing, sales and service, a view that is truly context-based.
The industry will sell you a template or single-use agents. But it won’t sell you the system in between: the data hygiene, the workflow design, the change management. This is left up to the customer. And the more seconded agents accumulate, the more difficult the job becomes.
The future belongs to companies that integrate AI into a coherent system, where data, workflows, agents and people share context. This is what we’re building at HubSpot. AI is a new level, not a replacement for the foundation.
AI needs to be built for the Future 5000, not just the Fortune 500.
Today’s AI roadmap is written for companies that can afford to run it. According to their own disclosures, Frontier Labs are spending billions of dollars on outside engineers to make AI work inside large companies.
This model works if you are a large business. It doesn’t work for the millions of growing companies that will drive growth over the next decade. A small or medium-sized company can’t hire outside engineers, rebuild its data pipeline, or build the context platform to make it all work.
So when the consensus says “AI is for everyone,” it’s worth asking who it actually works for today. In practice, it is the customers who can already afford to make it work, with armies of engineers and developers behind them. This is not democratization.
We are optimizing results per token, not tokens per activity.
There is a conflict between business models in the AI industry that customers have not yet fully seen. Vendors that benefit the most from using AI are not incentivized to make AI cheaper or more efficient. They are incentivized to keep the meter running. Then customers are asked to pay for the business and told that they are purchasing the transformation.
The honest version of AI economics is the reverse: be clear about the outcome the customer is trying to achieve, then find the lowest-cost path to get it. This is the client’s job. It should also be the seller’s. At this moment it is not.
Token-maxxing is the seller’s game. The maximum result belongs to the customer. Suppliers who align with the customer will win. Sellers who align with the counter may not.
Artificial intelligence is supposed to make people more powerful. No longer replaceable.
The strongest narrative about AI is autonomy: agents replace humans, headcount decreases, the future has fewer people. That narrative is built for Wall Street, not Main Street. We reject this classification.
We build for the person doing the work, not the person being taken out of budget. The rep closes more deals. The marketer sends multiple campaigns. The support person who solves more complex problems. The owner runs much of the business himself. AI’s job is to make them more powerful, not make them disappear.
Yes, we ship self-employed agents. But autonomy is a capacity, not a mandate. Customers decide where to delegate, where to keep humans in the workflow, and where the AI suggests. Our default settings are designed to serve the operator, not cut across the org chart.
We believe in human authenticity and the efficiency of artificial intelligence. The things that AI can’t replace: trust, judgment, taste, and relationship, will become increasingly valuable as the things AI can do become ubiquitous. Companies that bet against humans will lose the customer, the employee and… possibly the publicof which 57% already believe that the risks of AI outweigh its benefits.
Trust is more than a privacy policy.
Every AI vendor claims trust. But most define it as a security posture: we won’t train on your data, we’re SOC 2 compliant, we offer enterprise SSO. These things matter. They are also stakes. None of these are differentiated statements. They are what you promise.
What you demonstrate is another thing. True trust is a complete business posture: how you choose the model and manage costs, reliability and governance for your agents. This is what customers actually ask for. Can I trust the choice of model? Can I trust the cost? Can I trust the reliability? Can I trust the governance?
Privacy answers what we won’t do. Trust responds to what we will do. Most of the industry is still answering the first question. The second is what customers need.
What does all this add up to
The AI consensus remained valid until no one in the room had to answer for it. Cut the workforce. Tear out the old battery. Keep the meter running. Trust us.
Growing companies can’t waste time distinguishing what is exaggeration from what is reality. They don’t have detached engineers to dedicate to implementation. They can’t absorb a pricing model that bills for activity and calls it transformation. They can’t build on a stack that treats humans as an exception.
They need AI built on foundations that work for them, designed to empower and not eliminate people, and delivered by a vendor whose business model is aligned with theirs, not contrary to it.
This is what we’re building at HubSpot.
