The gap between AI perception and reality

There is a widening gap between what the market says about AI and what we actually hear from customers. The media, the VCs, the AI ​​labs, and influencers have all talked about AI replacing humans, tearing out trusted software, and token maxxing as goals worth pursuing. But leaders of real companies are increasingly asking the right questions. How do I make my employees better with AI? Which systems can I trust? How can I measure the ROI of this spend? 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 action, the AI ​​prospects we’re most confident about at HubSpot are things almost no one else says out loud.

Here are six of them.

AI activities are not AI results.

The industry has confused efforts to progress. Write emails, create summaries, conduct research. These are activities that have become significantly easier thanks to AI. These are useful features that we provide at HubSpot. But activity is the input, not the result. Activity without results is theater.

The companies that win with AI are those that start backwards from a business problem rather than forwards from a model demo. For example, customers using the Customer Agent respond to tickets 25% faster, while customers using the Prospecting Agent generate 76% more leads.

That’s why we transitioned the customer agent and prospecting agent to results-based pricing in April. The AI ​​results are crucial. And we help growing companies achieve them. We set our pricing where our point of view is.

AI is necessary. It’s not enough.

Generating code is certainly easier now. Anyone can build a prototype in a weekend, but it is brittle and falls apart under real use. Lowering the minimum code generation limit does not raise the shipping value upper limit because the things that actually run a growing business have become harder, not easier.

You still need clean data, not another silo. You still need to integrate dozens of applications. You still need a complete customer view across marketing, sales and service that is actually based on context.

The industry sells you a model or single-purpose broker. But you are not sold the system in between: the data hygiene, the workflow design, the change management. That is left to the customer. And the more unconnected point 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 have a common context. This is what we are working on at HubSpot. AI is a new level and not a replacement for the foundation.

AI must be developed for the Future 5000, not just the Fortune 500.

Today’s AI roadmap is written for companies that can afford to make it work. Frontier Labs says it is spending billions of dollars on predictive engineers to make AI work in large companies.

This model works if you are a large company. This isn’t working for the millions of growing companies that will drive growth in the next decade. A small or medium-sized business cannot hire forward-thinking engineers, rebuild its data pipeline, or build the contextual platform to make everything work.

So if the consensus is “AI is for everyone,” it’s worth asking who it actually works for today. In practice, it’s the customers who can already afford to make it work, with armies of engineers and developers behind them. This is not democratization.

We optimize for results per token, not tokens per task.

There is a business model conflict in the AI ​​industry that customers have not yet fully recognized. The providers who benefit most from AI use have no incentive to make AI cheaper or more efficient. You are incentivized to keep the meter running. Therefore, customers are asked to pay for the activity and are told that they are purchasing a transformation.

The honest version of AI economics is the opposite: get clear on what outcome the customer wants to achieve, and then find the most cost-effective way to achieve it. That is the customer’s job. It should also come from the provider. At the moment it is not.

Token maxxing is the seller’s game. Maximizing the result is up to the customer. The providers that align with the customer will win. This may not be the case for those providers that adapt to the meter.

AI is supposed to make people more efficient. No longer replaceable.

The loudest AI narrative is autonomy: agents are replacing humans, the number of employees is decreasing, the future has fewer people. This narrative is intended for Wall Street, not Main Street. We reject this formulation.

We build for the person who does the work, not the person who gets deducted from the budget. The agent closes more deals. The marketer sends more campaigns. The service employee solves more complex problems. The owner runs much of the business himself. The AI’s job is to make it more powerful, not to make it disappear.

Yes, we dispatch autonomous agents. But autonomy is a skill, not a mandate. Customers decide where to delegate, where humans are included in the workflow and where the AI ​​makes suggestions. Our default settings are designed to serve the operator, not disrupt the organizational chart.

We believe in human authenticity and AI efficiency. The things that AI cannot replace—trust, judgment, taste, relationships—will only become more valuable as the things AI can do become ubiquitous. The companies that bet against humans will lose the customer, the employee and ultimately the public, 57% of whom already believe that the risks of AI outweigh its benefits.

Trust is more than a privacy policy.

Every AI provider claims trust. But most define it as a security posture: We don’t train on your data, we’re SOC 2 compliant, we offer enterprise SSO. These things are important. They are also table stakes. None of these are a differentiated claim. You keep your promises.

What you prove is something else. True trust is a comprehensive business mindset: how you choose the model and manage costs, reliability and governance for your agents. This is what customers actually demand. Can I trust the model choice? Can I trust the costs? Can I trust the reliability? Can I trust the governance?

Privacy answers what we won’t do. Trust answers what we want. Most of the industry is still trying to answer the first question. The second is what customers need.

What this all boils down to

The AI ​​consensus remained as long as no one in the room had to account for it. Reduce the number of employees. Rip out the old stack. Let the counter run. Trust us.

Growing companies can’t waste time distinguishing between hype and reality. They don’t have forward-thinking engineers to help them implement it. You cannot accept a pricing model that charges for activities and calls it transformation. You can’t build on a stack that treats people as exceptions.

They need AI that is built on a foundation that works for them, that is designed to empower their people rather than eliminate them, and that is delivered by a vendor whose business model is aligned with their business model, not against it.

This is what we are building at HubSpot.