In Brief
AI agents are becoming a way to operate business software, raising questions about what companies should pay for access, usage, and completed work.
HubSpot's pricing experiments and Paul Roetzer's own workflow show why leaders need to understand the cost of an entire automated process.
What Happened
HubSpot is exploring a future where customers work through ChatGPT or Claude while HubSpot holds their data and powers the underlying tasks. This approach means users can ask AI to operate the software without navigating its screens. The company is also testing pricing that combines user seats with usage credits.
SmarterX founder and CEO Paul Roetzer is already using ChatGPT to retrieve webinar registration data from HubSpot. While making coffee, he asked for registration totals, then scheduled a daily report. He said the setup took about 30 seconds.
That experience sits alongside a much larger spending shift. Menlo Ventures estimates enterprise AI spending has reached a $177 billion annualized pace.
Roetzer explored what these changes mean for software budgets and company operations, on Episode 250 of The Artificial Intelligence Show.
The Key Numbers
$177 billion - Menlo's estimated annualized enterprise AI spending pace
$33 billion - Menlo's estimate of enterprise AI spending last year
4.7x - Average spending relative to last year among Menlo's surveyed companies
Up to 71% - Cost reduction claimed by Zapier's CEO for combining AI with repeatable automation
AI Changes the Economics of Software
The interface is becoming optional. Roetzer's workflow to produce his reports replaced the need to locate a dashboard or ask a colleague for an update. He then imagined extending it into a campaign: identify an audience, draft an email, and send the draft to the marketing team for review.
“I just have a different vision right now for what a company looks like,” says Roetzer. As agents take more steps on a user's behalf, the relationship between human logins and completed work becomes harder to price.
The cheapest workflow may combine several tools. The Information reported that LinkedIn kept its Hiring Assistant's operating costs flat as usage grew by replacing some OpenAI models with smaller models developed internally. Separately, Zapier CEO Wade Foster argued that AI should handle judgment while ordinary code runs predictable steps. An agent builds the automation and helps repair it when something breaks.
Roetzer sees a potential role here: an AI value engineer who compares ways to achieve the same result at lower cost. “That's the value engineer role. Somebody needs to do that,” he says.
Usage pricing creates another budgeting problem. HubSpot's credits can cover activities such as marketing emails, support resolutions, and customer data enrichment. Roetzer questioned how software vendors will charge when agents perform work previously spread across human users. Charging for outcomes or work performed could make bills harder to predict.
“Welcome to variable cost hell.”
— Paul Roetzer, founder and CEO of SmarterX, Episode 250 of The Artificial Intelligence Show
SmarterX Take
“We're going to build analysts for everything,” Roetzer says of the possibilities he now sees. That ambition needs someone responsible for comparing model usage, automation costs, and the value of the results. Determining where AI belongs in a process becomes an operating decision.
That also changes how leaders evaluate existing software. A platform may remain valuable as the place that stores data and executes tasks even when employees spend less time inside it. Menlo found limited evidence of companies replacing core systems such as Salesforce or Workday with AI. The immediate shift may be in how those systems are used and paid for.
What to Watch
Pricing experiments will test customers budgeting. HubSpot's combination of seats and credits offers one approach. As agents perform more work, the question is whether customers can anticipate their total bill and connect it to useful results.
Cost design could become a distinct business responsibility. Roetzer said he might hire for the value-engineering role. LinkedIn's model choices and Zapier's automation approach illustrate different decisions that this person would need to evaluate.
Is Your Organization Ready to Scale AI Workflows?
Only 25% of organizations represented in SmarterX's 2026 State of AI for Business Report have reached the Scaling phase of AI adoption. That finding puts the software-pricing discussion in context: Many organizations are still building the foundations for broader AI use. A successful individual workflow can reveal what is possible, while expanding it across a company raises further questions about costs, responsibilities, and readiness.
Based on more than 2,100 responses across roles, functions, and industries, the report provides benchmarks on adoption, training, governance, and tooling to help leaders identify what needs attention as they move beyond experiments. Read the full report →
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.