A Marc Benioff-backed startup thinks AI can solve the AI deployment problem

The paradox Efrat Rapoport describes is brutal: AI was supposed to reduce headcount. Instead it has spawned a new class of highly paid specialists — forward-deployed engineers — whose sole job is wrestling models into production inside sprawling enterprises. The market's response to the deployment crisis has been labor arbitrage. Throw more bodies at the integration layer. Rapoport and her three cofounders think that answer is wrong, and they just convinced Marc Benioff, Michael Dell, Aaron Levie, and George Kurtz to back a $20 million pre-seed bet that software can automate the grunt work.

This team knows the terrain. They built Bonobo AI, a pre-transformer language model company, launched a voice-to-text service in 2017, and sold to Salesforce two years later. They spent the next several years inside the tech giant watching customers hit the same wall repeatedly. The models worked fine in isolation. Put them against a live Salesforce instance layered over ServiceNow, DataBricks, Workday, and a dozen other systems accumulated across decades, and the project stalled. Duplicate database fields. Conflicting workflows. Technical debt measured in engineer-years. The model was never the bottleneck. The plumbing was.

June's pitch is that the plumbing can be mapped and fixed by code. The platform ingests a company's existing systems, reverse-engineers the actual business processes — not the org chart fantasy, the reality — identifies bottlenecks, and outputs a concrete, ordered remediation plan. Remove these duplicate fields. Connect this orphaned data source. Standardize that naming convention. Each step carries a "build" button. Click it and June writes the integration logic, deploys the agent, notifies the relevant teams through their existing comms channels. The human stays in the loop but stops being the pack mule.

Paul Akinmade, chief strategy officer at CMG, a major U.S. mortgage lender, moved his engineering organization onto Claude Code fast. The model layer was trivial. Then the deployment reality hit. He needed the AI to operate inside a mortgage origination stack that had accreted complexity since the Bush administration. June's approach would have handed him a sequenced punch list instead of an open-ended engineering project. That difference compounds across every Fortune 500 company currently burning budget on FDE squads.

The SaaSpocalypse narrative — that AI will eat traditional software vendors — assumes clean greenfield deployments. Enterprise reality is brownfield. No one is vibe-coding a CRM replacement for a global logistics conglomerate. The incumbents — Salesforce, ServiceNow, Workday — stay because ripping them out costs more than the predicted AI savings. June accepts that constraint. Its agents live beside the legacy stack, not atop a fantasy replacement. The value unlock comes from making the existing mess navigable to automation.

Investors bought in before a deck existed. That signals either extraordinary conviction or a valley echo chamber. The track record argues conviction. Bonobo exited to Salesforce. The founders then ran Salesforce AI initiatives. They have seen the failure modes at scale. They also know the buyer: the same CIOs who signed off on Bonobo's acquisition now approve June's pilot. Distribution risk is low. Execution risk remains high. Auto-generating a remediation roadmap is a hard ML problem. Executing the builds safely across heterogeneous environments is a harder engineering problem. One hallucinated "remove duplicate field" instruction cascades into data loss.

Rapoport's framing is precise: "Before AI can create value, someone has to deal with legacy systems." The industry has hired that someone. June wants to automate them. If the platform delivers, the FDE boom becomes a historical footnote — a labor-intensive bridge to a software-defined future. If it stumbles, the forward-deployed engineers keep their jobs. The $20 million buys the proof.