Marc Benioff Just Put $20M Into a Problem You Already Know How to Solve
A startup called June raised $20M to automate AI deployment. Read the fine print and it's the strongest argument yet that twenty years in one industry is about to get more valuable, not less.
Paul Akinmade runs technology at CMG, a mortgage lender. He spent weeks trying to connect an AI coding tool to his company's Salesforce instance. Not days. Weeks. It never worked.
When a vendor finally came to him with a fix, he told them something that should be printed on the wall of every AI company in the country: if the product requires forward-deployed engineers, he doesn't want the product.
That one sentence explains more about the current state of enterprise AI than any earnings call you'll read this quarter. And buried inside it is the best news senior professionals have gotten in about two years — though you have to read past the headline to find it.
The job title nobody had four years ago
"Forward-deployed engineer" — FDE — is the hottest role in enterprise software right now. If you haven't run into the term, here's what it means in practice: a vendor sells a company an AI system, and then also rents that company a human being to sit in their office and make the thing actually function.
The role exists because enterprise AI doesn't work the way the demos suggest. A demo runs against clean data in a controlled environment. A real company runs on four systems that were bought in different decades by different leadership teams, half-integrated by consultants who left in 2019, and documented nowhere.
So the vendor sends a person. That person spends six weeks crawling through schemas, mapping tables, and figuring out which of the four fields called "customer ID" the business actually uses. Then they write the integration. Then they leave, and the company discovers nobody internally knows how any of it works.
Akinmade's objection wasn't that he didn't want help. It's that he didn't want a black box plus a permanent babysitter. He wanted a tool his own team could operate.
The $20 million bet
This morning, a company called June came out of stealth to solve exactly that problem. Twenty million dollars, pre-seed — an enormous number for a first round. Marc Benioff's Time Ventures led it. Michael Dell, Aaron Levie, and George Kurtz came in alongside.
The founder is Efrat Rapoport. She and her co-founders sold their previous company to Salesforce in 2019 and spent the following years inside Salesforce watching enterprise deployments succeed and fail at close range. Reporting suggests they raised this round without a formal pitch deck, which tells you what the people writing the checks thought of the thesis.
The product scans a company's existing systems — Salesforce, ServiceNow, Databricks, Workday — and maps what's actually in there. Which databases exist. Which fields are duplicated. Where the workflows break down and why. Then it produces a step-by-step roadmap and drafts some of the implementation code.
Now here's the part worth sitting with.
The founder's own argument is the story
Rapoport's stated rationale for why her company should exist is not that AI reduces a company's need for outside expertise. It's the opposite. Her argument is that AI increases demand for professional services. The messier and faster this all gets, the more companies have to pay someone from outside to untangle it.
Read that again, because it's counterintuitive and it's coming from a source with no incentive to flatter anyone reading this.
A founder who just raised twenty million dollars from Marc Benioff — a man who has watched more enterprise software cycles than nearly anyone alive — is arguing that demand for outside expertise is going up, not down. That is not a warning. That's a market, described by someone betting nine figures of other people's credibility on it being real.
But it only pays off for you if you understand precisely which expertise is going up in value and which is going to zero. So let's be exact about what June does and doesn't do, because most of the coverage is sloppy about it.
What actually happens in an enterprise deployment
Break the forward-deployed engineer's job into its real components. There are five, and they're not equally hard.
One: discovery. Crawl the systems. Inventory what's there. Find the tables, the fields, the integrations, the dead code nobody deleted. This is grinding, unglamorous work and it eats weeks.
Two: interpretation. Figure out why it looks like that. Which of those four customer ID fields does the underwriting team actually trust? Which workflow is technically legacy but still load-bearing for a process nobody documented? Which duplicate exists because of a compliance ruling in 2019 that everyone has forgotten but that still binds?
Three: design. Decide what to automate, in what order, and what to leave alone.
Four: build. Write the integration code.
Five: change management. Get actual humans to adopt it. Handle the political fallout when a workflow changes and someone's territory shrinks.
June automates step one and drafts steps three and four. That's genuinely valuable — discovery is the single grindiest part of any deployment, and compressing six weeks into three days is a real result.
But notice what's untouched. Steps two and five. Interpretation and adoption. Those are not data problems. Those are organization problems, and no scanner reads them.
The thing a scanner can't see
June can tell you a company has four different fields called "customer ID."
It cannot tell you which one the underwriting team actually trusts.
It cannot tell you the third one exists because of a compliance ruling nobody has looked at since 2019 but that still constrains what you're allowed to automate.
It cannot tell you that the VP who owns that workflow will quietly kill the entire project if you touch it without asking him first — not in a meeting, not on the record, but by simply never freeing up his team.
That's not a gap in the technology. It's a category difference. The input required to answer those questions is two decades of watching how one specific industry actually operates, including the parts that were never written down because everyone involved already knew them. There's no dataset for that. You can't scrape it. You can't prompt your way to it.
The work being automated is the junior work
Here's the shift, stated plainly.
Crawling the schema. Mapping the tables. Writing the first pass at the integration. That was a twenty-six-year-old's job. A tool just did it in three days instead of six weeks, and it will keep getting faster.
What survives is the judgment layer. Knowing what the data means. Knowing which process is politically untouchable. Knowing the difference between a system that will get adopted and one that will sit unused for a year while everyone in the status meeting pretends it's working.
Nobody has automated that, and the reason isn't that the models aren't good enough yet. It's that the training data is a career.
This is why the fear framing around this story gets it backwards. The headline reads like AI is coming for the deployment consultant. What's actually happening is that the rented body model is compressing — the model where a vendor bills you for six weeks of a junior engineer's discovery time — while the judgment work becomes more concentrated and more valuable per hour. Fewer hours, higher stakes, harder to replace.
If you have twenty years inside mortgage operations, or claims processing, or clinical trial logistics, or supply chain for a specific regulated industry, you have been sitting on the part of this that doesn't compress. You've just never been asked to price it separately, because it always came bundled with a job title.
What this means for how you position yourself
Here's the practical version.
If you walk into a room and position yourself as "I know AI," you are competing with every twenty-six-year-old who watched the same videos you did — and with tools like June that are getting better at that exact layer every quarter. It's a commodity claim and it prices like one.
If you walk in as "I know mortgage operations, and I know AI," you are frequently the only person in the room who can do the job at all.
That's the whole game, and most people have it inverted. They spend their transition trying to catch up on the technical layer — certifications, courses, side projects — while treating their twenty years of industry knowledge as the boring part of the resume, the part they're trying to get past. It's the reverse. The tools got dramatically better at the part that was never actually hard. The part that was always hard is now the entire value.
The companies in this market will keep paying for deployment help. Rapoport just raised twenty million dollars on precisely that bet, and the AI Compute Funding Index we maintain tracks 450+ funded companies across twelve domains who are, every one of them, headed for the same wall CMG just hit. Fast-growing, well-capitalized, and about to discover that their AI works beautifully right up until it meets a legacy system somebody built in 2014.
The only open question is whether they pay you, or whether they pay someone who has never seen the inside of your industry and will spend the first two months learning what you already know.
What to actually do with this
Three things.
Stop leading with the technology. Your headline, your opening line on a call, your first paragraph — none of them should start with AI fluency. Start with the domain. The AI is the multiplier on the noun, not the noun.
Name the specific systems you know. Not "process improvement." The actual stack. The actual regulatory constraint. The actual reason your industry's data is shaped the way it is. Specificity is the proof of the thing that can't be automated.
Look at the funded companies, not the household names. The organizations with this problem right now, urgently, are the ones that just raised and are scaling fast into legacy-integrated enterprise customers. They're hiring for it and they're not household names, which is exactly why the opening is still there.
The market didn't get worse this morning. It got clearer about what it's willing to pay for.
See Where You Fit
You've read who's getting funded and who they need. The real question is the personal one: where does your twenty years plug into this?
I'll build you a free, personalized Opportunity Map + Executive SWOT — the specific companies hiring for a background like yours, and an honest read on how you show up right now — back in your inbox within 24 hours. No cost, no pitch.
Show me where I fit → build my free Opportunity Map + SWOT
Prefer to talk it through first?
Written by
Bill Heilmann