US-based firm Curative says it replaced Salesforce and cut a $600,000 bill to zero

What exactly was claimed, and who claimed it?

The claim is worth reporting precisely, because the precise version is less tidy than the viral one.

The words came from a post shared in July 2026 by Harry Stebbings, founder of the technology podcast 20VC. Presented as a quote, it read: ‘We replaced Salesforce with a vibe-coded CRM built for our own workflows. The custom system integrated our AI agents more effectively, worked better for the team, and made Salesforce unnecessary. That decision cut a $600,000 annual software bill to zero.’ Stebbings did not present it as settled proof. He asked his followers whether it was an anomaly or the start of a much larger trend.

Two plain-language notes help here. A CRM is the software a company uses to keep track of its customers and its sales. 'Vibe coding' is the new shorthand for building software by describing what you want to an AI tool in ordinary language and letting it write most of the code.

Two things are worth adding to the version most people saw. First, the speaker. The words are those of Fred Turner, co-founder and chief executive of Curative, a US health insurer, speaking on Stebbings' 20VC podcast. Turner said the internal system was built in about two months, and that Curative intends to cut roughly 80 per cent of its software spending this year. Second, the saving is self-reported. There is no independent audit of the $600,000 figure, the timeline, or the claim that the new system works better.

‍What do the studies of AI at work actually find?

When researchers count the failures as well as the successes, the picture changes sharply, and it has stayed remarkably steady for more than a year.

The pattern was first put on record in July 2025, when the Massachusetts Institute of Technology published a study called The GenAI Divide: State of AI in Business 2025. It examined 300 public AI deployments and surveyed and interviewed hundreds of managers and staff, and found that around 95 per cent of enterprise AI pilots delivered no measurable impact on profit or loss. Only about 5 per cent produced real, countable value. The researchers were careful to say the problem was rarely the technology itself. Most projects stalled because the tools could not hold on to context, adapt to how a team actually worked, or survive contact with daily routines.

A year of newer research has not softened that finding. It has confirmed it. In September 2025, Boston Consulting Group surveyed 1,250 senior executives across more than twenty industries and found that only 5 per cent of companies had reached the level it calls 'future-built', the small group reliably turning AI into value at scale, while about 60 per cent were still generating no material return despite continued spending. In April 2026, the analyst firm Gartner reported a survey of 782 technology leaders in which only 28 per cent of AI use cases were judged to have fully succeeded and met their return-on-investment expectations, one in five had failed outright, and well over half of those leaders had at least one failed project behind them.

Three independent studies, spread across a year and using different methods, keep landing on the same uncomfortable point. The technology can be powerful, but most attempts to put it to work still do not pay off, and the reasons are human and organisational rather than magical.

So did anyone actually replace Salesforce?

The most useful evidence comes from the people who have gone furthest and are honest about the details.

Jason Lemkin, who runs the software community SaaStr, is exactly the kind of person the viral claim describes. His small team runs more than twenty AI agents in daily use, and they have closed real revenue. Yet he is blunt about what they did and did not do. 'We did not vibe code our own CRM. We didn't rebuild Salesforce. We went headless on top of it.' In other words, they kept the underlying Salesforce system that stores their data and dropped only the screens people used to log in, running their own agents on top. His warning to anyone tempted to rip the whole thing out is practical. Every connection to every other tool becomes your problem, permanently, and the AI that built the system will not be the one woken at eleven at night when a silent failure quietly corrupts the sales forecast.

The most famous earlier example tells the same cautionary story. The Swedish company Klarna was widely reported to have replaced Salesforce with AI. Its chief executive later clarified, as reported by TechCrunch, that this was not what happened. The company had consolidated scattered data onto its own internal stack while still using Slack, which Salesforce owns, and he said plainly that he did not expect other companies to follow the same path.

This does not mean the pressure on software is imaginary. Real, if more modest, savings are being reported. A partner at Greenleaf Management, a firm that had used Salesforce, said it had saved roughly $100,000 by rebuilding a custom application with AI coding tools. A company called Oplign said it had replaced a rival tool over a single weekend at a fraction of the cost. The honest reading, offered by several analysts, is that AI has genuinely lowered the cost of building software, which puts real pressure on weak, bloated or overpriced products. The leap from that to every company building its own core systems has no evidence behind it. Even Salesforce has responded not by disappearing but by letting customers point AI coding tools at their data while the Salesforce system keeps running underneath.

What does this mean for the future of work?

The durable pattern is not replacement but rearrangement. The parts of a job that AI handles well, drafting, summarising, moving data between systems and running defined tasks, are being handed to it. The parts that stay stubbornly human, deciding what is worth building, judging when a slick demonstration will fall apart in real use, and owning the consequences when something breaks, are becoming more valuable rather than less. The 'headless' model is a good picture of the future of work as a whole. The engine keeps running. What changes is who sits at the controls and what they choose to spend their attention on.

That reframes the anxiety so many working professionals now carry. The pressing question is not whether AI will take your job in one dramatic stroke, the way the viral clip implies. It is whether you can tell a real capability from a rehearsed one, build with the new tools where they genuinely help, and recognise the places where they add cost and risk. Those are learnable skills, and they reward judgement over panic.

The direction of travel is real. Gartner expects that by 2028 a meaningful share of routine work decisions will be made by autonomous software, and that a third of business applications will include AI agents. But real is not the same as instant, and a genuine trend is not the same as the frictionless miracle in a shareable quote. The future of work will be shaped far more by the slow, unglamorous question of what actually holds up in daily use than by a single story that happened to travel well.

So the sensible response to the next viral claim is neither to dismiss it nor to rebuild your career around it. Treat it as a single survivor. Ask what is not being shown. Learn the tools that make your own work sharper, and keep the judgement that lets you see the planes that did not come back. They are not in the feed, and they are still the larger number.

‍

Sources