A startup called June left stealth mode this week with 20 million dollars in early funding, backed by Marc Benioff's investment firm along with Michael Dell, Aaron Levie, and George Kurtz. Its founders spent years inside Salesforce after the tech giant bought their earlier company, Bonobo AI, in 2019, and they say that experience taught them something uncomfortable: buying AI is easy, but making it work inside a real business is not.
The core problem June is targeting is not really about AI at all. It is about decades of messy company data. Business records get duplicated across different departments, definitions of the same customer field don't match between systems, and workflows have been patched together for years. An AI agent dropped into that mess often cannot tell which version of the data to trust, so it fails quietly instead of saving anyone time.
This is not a small or rare issue. A widely cited MIT study found that 95 percent of company AI pilot projects fail to deliver any measurable financial payoff, mostly because of exactly this kind of plumbing problem rather than the AI models being bad. Separate industry research has put failure rates for AI agents in production even higher when you include projects that never make it past testing.
Businesses have mostly responded to this by hiring people, specifically forward deployed engineers who embed inside a company for months to manually connect AI tools to existing software. Demand for these specialists has exploded so fast that one executive search firm estimates fewer than 2,000 people in the United States have proven they can reliably deliver a return on AI spending this way. That scarcity means their time is expensive and their calendars are full, which slows down every company waiting in line behind you.
June's bet is that a lot of this work can be automated instead of hired out. Its software scans a company's systems, maps out the technical debt getting in the way, and generates a step by step cleanup list, remove these duplicate fields, connect this data source, before building the actual AI agent. One early customer, a mortgage lender called CMG, says the tool got a Salesforce integration project moving after weeks of stalling with human consultants.
The bigger story here is not really about one startup. It is a sign that the AI industry is quietly shifting its attention away from building smarter models and toward the unglamorous work of making AI actually function inside real companies. For any business leader who has already bought AI tools and found them harder to use than promised, that shift matters more than the next model release. The winners in this next phase may not be the companies with the flashiest AI, but the ones that figure out how to plug it into what they already have without hiring an army to do it.