Military planners learned this rule the hard way: doctrine is written at the front, not at headquarters. Every army now studying drone warfare studies Ukraine, because that is where the future arrived first, under pressure, at scale. Business leaders need the same discipline with AI. The future of the firm is not in a vendor’s roadmap or an analyst's forecast. It is visible today, in a handful of companies operating at the far edge of adoption.
This week, one of them published its telemetry. Replit, the AI coding platform, released a post called "The Self-Driving Company" describing what happened when it wove agents into the fabric of the business rather than leaving them inside chat windows. The numbers deserve your attention. From January to June, lines of code contributed rose 5.8x. Holding the cohort of engineers constant, output per engineer nearly tripled while the team doubled. Code review latency stayed flat, because an agent now reviews code and calls in a second human only when risk warrants it, saving 30 percent of human review time. Reversion rates and incidents stayed flat. Support closes its hardest escalated tickets 60 percent faster. The company churned a seven-figure SaaS contract because the internal replacement its own agent helped build was better.
Replit is a 4th Stage Firm: Emergent Intelligence
This is no longer an engineering story. At Replit, agents investigate production incidents, answer product questions, analyze business data, enrich sales leads, build account-specific presentations, and triage support tickets. The company's own summary is the best one-line description of the new deal for employees I have read: “People don't feel like they've been automated. They feel like they've been promoted.”
I map enterprise AI adoption on a model I call the 4 phase RISE adoption model: Phase 1: Research & Education, Phase 2: Islands of Innovation, Phase 3: Scaling, and Phase 4: Emergent Intelligence. Most companies we assess sit in the first two stages. (See my Forbes article on the importance of building capabilities.) They are educating their staff, running pilots, funding proofs of concept, and measuring seat licenses. Some have achieved scale implementations like JP Morgan Chase. Check out thi wonderful article by my fellow Forbes contributor Bernard Marr about JP Morgan’s AI efforts. The last phase, Emergent Intelligence is different in kind, not just degree. Intelligence stops living in individual tools and starts living in the connective tissue of the firm: the artifacts, the orchestration layer, and the contracts between humans and machines. Replit’s post is the clearest public description yet of what that stage looks like from the inside.
MORE FOR YOU
Here is the objection you are already forming: "My firm is not a frontier AI company. This does not apply to me." That objection misses the point, the same way a general dismissing Ukraine because his defense needs are unique. I’m not suggesting executives study the leading edge to copy it. One should study it to extract the design principles before your competitors do.
Could Your Firm Become Self-Driving?
So run the exercise of seeing what it would take to make your organization become a self-driving firm. Take your leadership team through one question: what would it take for us to be a self-driving company? Not to become one next quarter. To be able to describe one, in the specifics of your own business.
The question forces answers most executive teams cannot give today. Which of your workflows produce verifiable outcomes an agent could be checked against? Replit started there. What would agents need to see to do real work: your data warehouse, your ticketing system, your CRM? Replit built a semantic layer over its data so agents knew which tables were sources of truth. What guardrails come first? Replit locked everything behind access policies, token proxies, and audit logging before granting access. Who holds authority when the agent escalates? And what happens to your build-versus-buy calculus when your internal agent replaces a seven-figure SaaS product and matches specialized vendor tools at a tenth of the cost, as Replit reports it did?
Notice that not one of those questions is about picking a model. They are questions about the design of your firm: its decision rights, its data contracts, its accountability structure. That is why the self-driving company works as a design exercise even for a bank, an insurer, or a manufacturer. The technology will keep changing. The organizational questions will not.
The firms at the Emergent Intelligence stage are running the experiment in public and publishing the results. The cost of studying them is an afternoon with your executive team. The cost of ignoring them is discovering, two years from now, that a competitor ran the exercise and then executed it.
The front line is visible. Go look at it. %!s()
Image: Valve
Image: Valve