1 hour ago
7
SaaS
AI coding agents may make software cheaper to build, but switching off major platforms remains expensive, risky, and deeply annoying
Salesforce CEO Marc Benioff appeared on the VC podcast All-In last week to share his vision for the future of AI and software. Among the usual bonhomie – he skipped Trump's trip to China; he's not a Democrat or Republican, just "here to support the country" – he revealed the CRM giant he co-founded may spend circa $300 million with Anthropic in 2026 to harness its coding agents.
"Everything's gonna be cheaper to make, it's more efficient," he told the show. "I can do things that I just could not do before. I can go faster than ever before. I can implement my software and sell it at the same time. I can break through obstacles that I've had because I have coding agents and humans working together."
There are other consequences of this rush to the LLM-guzzling future. Despite its growth, Salesforce did not hire any more software engineers in 2025. It has also cut around 4,000 support staff, although it has been hiring elsewhere.
So what's Salesforce doing with the money it's saving? It’s not going to the customer, it seems. In 2025, Benioff told investors he saw a "very high margin opportunity" in users adopting its AI agent platforms. Later, Miguel Milano, president and chief revenue officer, said he was happy to make a loss on a capped-price deal for AI agents in the short term, as the company had "20 years to monetize that customer."
Gartner later warned Salesforce users that a capped enterprise agreement for its AI and data platforms may not be available when they come to renew, potentially meaning customers could struggle to predict costs and understand value.
Salesforce responded: "The claim that we are moving away from capped agreements is inaccurate. Renewals remain flexible, and because AI compute costs may actually shift over time, we focus on tailoring terms so each customer can get the maximum value from their usage."
Regardless of the commercials, Benioff argued on the All-In podcast that customers would welcome the business outcomes from adopting its AI agent-based CRM and sales platforms. "What I can do for our customers is unprecedented; it's impossible to describe what we're gonna be able to do for customers. It's gonna be awesome."
Such confidence in AI coding agents' ability to help Salesforce build Salesforce raises the question of whether users could just get coding agents to build Salesforce, and cut out the middleman. The idea is the foundation of the so-called SaaS-pocalypse, exemplified by a market-rattling report from Citrini Research, which imagines the impact of LLM-based AI on the market for SaaS is such that by June 2028, US unemployment will hit 10 percent and former staffers at application vendors may have to work in other professions
The sentiment was echoed in a Reddit discussion based on the All-In podcast. One participant claimed to know of people who have already vibe-coded CRM systems to avoid paying Salesforce.
Even assuming that's viable, it wouldn't have the impact many expect. Users already have an alternative to paying big vendors. There have been FOSS alternatives to Microsoft's commercial bear traps office applications and PC operating systems for more than 20 years.
The reason customers don't move off these products is that although software is expensive, and some vendors enjoy eye-watering margins, most organizations don't spend that much on software as a proportion of their revenue. IT spending is 3-10 percent of total revenue in most organizations. Most of that is salaries. Software is way down the list.
It would be a brave CIO who was prepared to face the wrath of users by changing their work environment – users always hate change – for a mere fraction of a percent of company revenue. Add the cost of moving data and the difficulty of replicating security, and the threat to the big SaaS application vendors looks minimal.
Some small, specialist tech companies might try it. And vibe coding might have value on the periphery, building extensions or interfaces for existing applications. But whether it actually offers customers value commensurate with application vendors' increased margins is another question. ®
English (US) ·