Americans Hate AI. Competitors Worldwide Are Already Agentic.

Seventy-one percent of Americans don't want a new AI data center built anywhere near them. Two-thirds of small-business ChatGPT output is already running agentically worldwide — up from a third in April. Both of those are true at the same time, and only one of them is going to decide who's still in business in two years.

I say this to prospects more than I'd like to: the public mood on AI and the private adoption curve inside actual businesses have almost nothing to do with each other. People can hate the idea of AI in the abstract and still have it quietly doing real work inside their company by Tuesday. If you're waiting for the vibes to improve before you touch any of this, you're not being cautious. You're just watching competitors who didn't wait pull ahead while you argue about whether it's trustworthy.

The trust gap is real

Gallup's numbers, via MIT Technology Review, aren't close [1]. 71% of US adults oppose building new AI data centers near them. For comparison, only 53% oppose new nuclear plants — AI infrastructure polls worse than nuclear infrastructure.

Source: MIT Technology Review — People really hate AI. So why can't they get enough?

And yet ChatGPT crossed a billion monthly users in May. Half of American adults now use a chatbot; a quarter use one daily [1]. Distrust and daily use are not in tension. They're the same population, doing both at once.

Two-thirds of small-business ChatGPT output is now agentic worldwide

Here's the number that should actually change how you plan Q4. OpenAI says that in August, agentic output — tasks completed by ChatGPT Work and Codex, not just chat replies — made up two-thirds of all small-business output tokens on its platform worldwide. In April, that share was one-third [2]. That's not steady growth. That's a category shift inside five months.

The usage behind it is already wide, not niche. In the week of September 9–15, roughly 4 million employees worldwide at companies with fewer than 500 people used OpenAI's products. Nearly one in five of those users worked at a business with fewer than 10 employees [2]. This isn't an enterprise-only trend wearing an SMB costume. Solo operators and five-person shops are already in it.

Chat vs. agentic — the line that actually matters

Most people still describe all of this as "using ChatGPT," as if asking a question and handing off a multi-step task are the same action. They're not, and the difference is the whole story:

The risk in agentic use isn't that it's unreliable. It's that it's reliable enough that you stop checking, and the one time it's wrong, nobody catches it until a client does.

OpenAI just put a clock on the head start

On September 30, OpenAI and America's SBDC — the network behind roughly 1,000 small business development centers nationwide — announced a partnership to train about 150 SBDC advisors [2] and get hands-on AI workshops in front of at least 1,000 small businesses [3]. Advisors and clients who finish the in-person sessions get free ChatGPT Plus. Rollout is phased over the coming months, not everywhere at once [3].

Strip the press-release framing and what's left is a limited-run, free-to-claim training pipeline rolling out on a schedule you don't control. I've written before about what happens when SMBs treat a vendor's free tier as a permanent feature instead of a window — the terms change under you. This one's the inverse problem: the free seats and the trained advisor in your area are a window too, and it closes region by region as the phases roll out. Whoever in your market books the workshop first gets the SBDC's attention and the free Plus seats. Whoever shows up in month four gets whatever's left.

What this looks like inside a 5–20 person shop

Most of the SMB teams I audit are stuck at stage two below — using AI to speed up one task at a time, never handing it the whole job.

flowchart LR
  A["Stage 1 — Chat Q&A<br/>ask it, copy the answer"] --> B["Stage 2 — Single-task automation<br/>one prompt, one deliverable"]
  B --> C["Stage 3 — Agentic workflow<br/>multi-step, tool use, memory"]
  C --> D["Stage 4 — Outcome ownership<br/>agent finishes the job, a human reviews"]
  Note["Most SMBs I audit are stuck here"] -.-> B

Getting from stage two to stage three isn't a tooling problem — Claude, ChatGPT Work, and half a dozen agent frameworks can already do the multi-step part. It's a trust-and-process problem: you have to decide which outcomes you'll let an agent own, and build the review step for the ones you won't. That's the same work I did at Systemic Digital, where the real fix wasn't a new tool — it was mapping which parts of their delivery process were actually deterministic enough to hand off, and which still needed a human call. The framework born on that engagement is the one I still run on every new client.

Nobody has to trust AI to lose to a competitor who's already using it.

What to actually do this week

If your business is still entirely at chat-and-copy, you don't need a philosophy about AI. You need one task — invoicing, reporting, ticket triage, whatever eats the most unglamorous hours — moved from "someone asks ChatGPT and copies the answer" to "the agent does the task end to end and a human signs off." Not everything. One thing, done honestly, with a real review step attached. That's the actual service I sell, and it's a smaller, cheaper, less dramatic engagement than most people expect — because the goal isn't to automate your whole business, it's to find the one task where an agent earns outcome ownership and prove it before you touch anything else.

The short version

• 71% of Americans oppose new AI data centers near them — AI polls worse than nuclear power [1]
• ChatGPT crossed 1 billion monthly users anyway; half of US adults use a chatbot, a quarter daily [1]
• Two-thirds of small-business ChatGPT output tokens are now agentic worldwide — up from one-third in April [2]
• ~4 million employees worldwide at sub-500-person companies used OpenAI's products in one September week; 1 in 5 of those at businesses under 10 people [2]
• OpenAI + America's SBDC are training 150 advisors [2] and targeting 1,000+ small businesses with free ChatGPT Plus access, rolling out in phases starting now [3]
• The real decision isn't whether to trust AI — it's which single task you let an agent finish end to end first

Sources

[1] MIT Technology Review — People really hate AI. So why can't they get enough? — technologyreview.com [2] OpenAI — Helping small businesses put AI to work — openai.com [3] America's SBDC (PR Newswire) — America's SBDC Announces Partnership with OpenAI to Bring AI Training to Small Businesses Nationwide — prnewswire.com