Every so often an industry stops arguing and starts writing checks. Between March and July of 2026, the AI industry did exactly that — and the checks tell you more than any keynote ever will.
Eight weeks that settled the argument
In March, Sequoia Capital published an essay called "Services: The New Software." Its thesis was simple and a little heretical: the next giant AI companies will not sell software tools — they will sell completed work. For every dollar businesses spend on software, they spend six on services. The model that wins is the one that does the job, not the one that helps you do the job.
Then the checks started.
- May: OpenAI launched The Deployment Company — a separately capitalized firm, roughly four billion dollars behind it, whose entire purpose is to embed engineers inside enterprises until AI runs in production. Anthropic followed the same month with its own embedded-engineering venture.
- June: AWS committed one billion dollars to a new Forward Deployed Engineering organization — thousands of engineers placed inside customer companies, priced around, in their words, "shared goals and business results, not billable hours."
- July: Microsoft answered with Frontier Company: two and a half billion dollars, six thousand experts, and a tagline that concedes the whole point — "outcomes, not outputs."
Nine billion dollars, four rivals, one conclusion: models don't transform businesses. Deployment does. The scarce thing was never intelligence — it's the engineer sitting inside your operation, finding the work worth automating, and wiring it in until it holds.
What "forward-deployed" actually means
The term comes from Palantir, which borrowed it from the military. A forward-deployed engineer doesn't work from headquarters and ship you a tool. They embed with you, learn how the work actually flows — not how the org chart says it flows — build against your real systems, and stay until the thing runs in production. The deliverable isn't a recommendation. It's a working system and the keys to it.
If that sounds like common sense, it is. It's also the exact opposite of how most companies have experienced AI so far: a pilot, a demo, a deck, and a renewal notice.
Why the giants stop at the Fortune 500
Here's the part the announcements don't say out loud. A billion-dollar deployment division has a minimum deal size, and it isn't yours. These programs exist for the NFL, for global banks, for companies with procurement departments larger than most contractors' entire payroll. The economics of six thousand embedded experts only work at enterprise scale.
Which leaves a strange and temporary gap: the model has been proven, loudly, at the top of the market — and the businesses that actually run your city (the contractor, the franchisee, the shop, the firm) have no one offering it to them. The work is the same. Invoices don't care about your headcount. Insurance certificates expire on the same schedule at a 33-location franchise as they do at a Fortune 100.
What this means if you own a business
Three things, practically:
First — stop evaluating tools and start pricing work. The question is not "which AI should we subscribe to?" It's "which recurring block of hours could a system take off our people?" Count the hours someone spends re-typing invoices, chasing certificates of insurance, assembling the same Monday report. That number, multiplied by a year, is your budget line. It's usually far bigger than any software subscription you've been debating.
Second — start with the boring problem that costs you the most. Every announcement above quietly agrees: transformation programs fail; single mapped workflows in production succeed. The companies that win the next few years won't have the fanciest AI. They'll have picked the right dull, expensive process and solved it completely.
Third — own what gets built. The enterprise programs compete on trust: your data, your cloud, your keys. Demand the same at your scale. If a vendor's answer to "who owns this when we're done?" is a subscription, you're renting a tool, not buying work.
Where to start
Not with a call, necessarily. Start with an honest inventory: the Workflow Audit is ten yes/no questions about how your operation actually runs — self-scored, no email required. If you'd rather see what this looks like when it's finished, the case library has the receipts: one assistant running thirty-three restaurants, five agents behind a 25-year general contractor, compliance dossiers cut from weeks to minutes.
The giants just spent nine billion dollars agreeing on how AI actually becomes useful. The good news is that the method scales down. The better news is that, at your scale, it's a project measured in weeks — not a division measured in billions.