Something unusual is happening in the business formation data — and if you run a service firm or you're thinking about going independent, it matters more than almost anything else you'll read this week.
Bank of America published a chart last week (via a16z) that shows two lines diverging in a way they never have before.
The orange line is new business applications. It's elevated — well above pre-2020 norms. People are still starting businesses at rates we haven't historically seen.
The green line is "high propensity to hire" applications. These are the businesses that indicate, at the time of filing, that they plan to bring on employees. That line just fell off a cliff.
People are starting businesses. They are not planning to hire anyone.
a16z titled the chart "AI-Native Solopreneuring?" — with a question mark. I'd remove the question mark.
For most of modern business history, starting a company and hiring people were the same decision. You couldn't run an accounting practice without staff accountants. You couldn't run an architecture firm without junior designers. You couldn't run a law practice without paralegals. The operational overhead — scheduling, intake, document processing, billing, research, compliance — required bodies.
That link is breaking.
The people in that orange line — the ones starting businesses without planning to hire — aren't building lifestyle businesses or side hustles. Many of them are trained professionals. Former associates. Former staff accountants. Former project coordinators. People with deep domain expertise who are discovering that AI agents can handle the operational work that used to require a team.
A single person with a Mac Studio on their desk running open-weight AI models privately can now process, route, and act on information at a volume that used to require three to five support staff. Intake. Scheduling. First drafts. Data entry. Follow-ups. Status updates. Billing reconciliation. The information logistics that consume the majority of time in any service business — agents handle that now.
The person provides the judgment, the client relationship, and the expertise. The agents handle everything else.
The green line falling isn't just about solos. It's a leading indicator for what's happening inside established firms too.
If new entrants to your market are operating without staff, they have a structural cost advantage you can't match by working harder. Their overhead is a $4,000 machine that runs 24/7 with zero ongoing costs. Your overhead is payroll, benefits, office space, and management time.
This doesn't mean your firm is dying. It means the math on which roles justify their cost is changing — fast. The firms that will thrive are the ones that figure out which work is information logistics (and hand it to agents) and which work is genuine expertise (and protect the conditions for doing it well).
Most firms, when they're honest about it, will find that the vast majority of staff time goes to preparing for work rather than doing the work itself. Research before the brief. Drafting before the review. Formatting before the delivery. Scheduling before the meeting.
Agents handle that entire layer. The question isn't whether to deploy them. It's how fast you can identify which of your information bottlenecks have the highest return when automated — and whether you find the simple wins before your new, leaner competitors do.
If you're one of the people in that orange line — or thinking about it — the data is telling you something encouraging.
You're not crazy. You're not reckless. You're reading the same structural shift that Bank of America's research team is charting and a16z is publishing. The infrastructure to run a one-person practice that competes with a 20-person firm exists right now. It wasn't available two years ago. It is today.
Here's what the practical stack looks like:
A Mac Studio sitting under your desk runs open-weight AI models entirely locally. Your client data never leaves the machine. No cloud APIs. No per-token costs. No third party ever sees your work product. For anyone bound by privilege, NDAs, or regulatory obligations, this isn't just convenient — it's the only architecture that satisfies your ethical duties by design rather than by policy.
On top of that hardware, autonomous agent frameworks — like OpenClaw, which NVIDIA just gave enterprise-grade security at their GTC conference two weeks ago — let you deploy agents that handle scheduling, intake, document processing, research, drafting, billing, and client communication. Not chatbots. Agents that observe, decide, and act in continuous loops.
The person who used to need a paralegal, an admin, and a bookkeeper now needs a machine and the knowledge of how to configure it.
That knowledge is what separates the people in the orange line who will succeed from the ones who won't.
Here's something that doesn't show up in the a16z chart but matters enormously for professional services.
Most AI adoption discussions assume cloud-based tools — ChatGPT, Copilot, Gemini. For a tech startup, that's fine. For a lawyer, architect, or accountant handling client-confidential information, it's a non-starter.
A recent federal court order requires OpenAI to preserve all user conversations indefinitely — including deleted ones. OpenAI now scans private conversations and reports flagged users to law enforcement. The data security concern among legal professionals is the number one barrier to AI adoption, cited by 46% of respondents in the most recent industry survey.
Private, local AI isn't a premium feature. For professional services, it's the baseline requirement. The fact that a $4,000 Mac Studio now meets that requirement — running models capable enough to power autonomous agents — is what makes this entire shift possible.
The people starting businesses without planning to hire aren't just cutting costs. They're building practices that are private by architecture, not by policy. That's a competitive advantage that cloud-dependent firms cannot match.
