6 min read
“There’s very little that we do where we need frontier models.” That is not a line you expect from the CEO of one of the world’s biggest banks, at an event about artificial intelligence. But it is what Goldman Sachs CEO David Solomon told a room of small business owners on October 22, according to Fortune’s report from the launch.
“Frontier models” is industry shorthand for the newest, most powerful, and usually most expensive AI systems, the top tier that each company releases to win benchmark races. Solomon’s point was that most everyday business work does not need them. He said they matter for areas like cybersecurity, defense, and drug discovery, and that those issues are separate from using the basic tools that already exist to run a business.
Here is my take, and I will try to stress-test it rather than cheer it: for most solo owners, Solomon is right. But he is right for a more specific reason than “cheap is good,” and there are a few jobs where paying for the top model really does earn its keep.
The context: a $500 million bet on Main Street AI
The remark came at the launch of Goldman’s new Small Business AI Institute, part of a fresh $500 million commitment to its long-running 10,000 Small Businesses program. As the program announcement describes, the institute brings in Google, Anthropic, OpenAI, and Meta for training, plus research on how small firms use AI and help with access to capital. Google, for example, is offering its AI Professional certificate to nearly 20,000 owners and employees with three months of free Google AI Pro.
The survey Goldman shared alongside it is the more revealing part. In a Babson College survey of program graduates conducted in August, 73% of AI-using owners spend under $100 a month on AI, and 16% use only free tools. Yet 86% now use AI in some form. In other words, small businesses are already getting value without buying the top tier. Solomon was describing what owners are doing, not just giving advice.
Why “you don’t need the frontier” holds up for solo work
Most solo tasks are bounded
Drafting a follow-up email, summarizing a call, rewriting a product description, turning notes into a proposal outline, categorizing expenses. These are well-defined jobs with a clear right answer shape. Mid-tier and even free models handle them well, because the hard part is usually giving the AI your facts and context, not raw reasoning power. A top model that does not know your prices will still guess them.
The bottleneck is setup, not intelligence
The same Babson survey found only 15% of graduates say AI is fully built into their core operations, and 77% want more training. That gap is not caused by weak models. It comes from owners not having a saved prompt, a facts sheet, or a repeatable workflow. Upgrading the model does nothing for a process that does not exist yet.
Expensive tools punish casual use
Solomon also said that “at the moment, this is an expensive technology,” and that the goal is to “get it right, not to get it quicker.” For a one-person business, every subscription competes with ad spend and your own paycheck. Paying for a premium plan you open twice a week is the AI version of a gym membership you do not use. If you are unsure how usage-based plans add up, our beginner’s guide to AI credits walks through the budgeting.
Where the argument breaks down
A skeptical read has to cut both ways. “You rarely need frontier models” is not the same as “you never do.” Here is where I would pay for the top tier without hesitating:
- Long, messy documents with real stakes. Reviewing a 40-page commercial lease, a client contract, or a grant application, where missing one clause costs real money. Stronger models are noticeably better at keeping a long document in mind and spotting contradictions. (They still do not replace a lawyer.)
- Multi-step agent work. When an AI is asked to do a chain of tasks on its own, such as researching, comparing, and drafting across several apps, small errors compound. Stronger models fail less often per step, and that adds up.
- Numbers you will act on. Building a pricing model, a cash flow forecast, or a quote calculator. A cheaper model may produce something that looks right and is subtly wrong.
- Your one highest-value workflow. If one process brings in most of your revenue (say, proposals for a consultant), the quality jump may be worth $20 a month even if everything else runs on free tools.
Notice the pattern: the frontier earns its price when the cost of a mistake is high or the task is long. It does not earn it for volume chores.
The rule I would give a solo owner
Instead of asking “which AI is the best,” ask two questions about each task:
- If the AI gets this slightly wrong, what does it cost me? A clumsy social caption costs nothing. A wrong number on a quote costs a client.
- How long or complex is the input? A paragraph of notes is easy. A stack of contracts or a multi-app task is hard.
Low cost and short input: use the free or cheapest tier, whether that is free ChatGPT, free Claude, or free Gemini (we compared them in our free plan comparison for solo businesses). High cost or long input: use the strongest model you can access, and check its work. That is it. Most owners will find 80% or more of their tasks land in the first bucket.
What this means for the AI companies chasing you
There is a quiet irony in all four frontier labs sponsoring an institute whose host just said the frontier is mostly unnecessary for small business. It also hints at where the market is heading. The labs know that small businesses will not pay premium prices for chores, which is why cheaper, faster models keep arriving, as we saw with the latest round of budget-friendly model releases. Competition for your $20 is pushing good-enough AI toward free.
That is good news, with one warning. As the base tools get cheaper, the advantage shifts to how well you set them up: your prompts, your saved instructions, your facts sheet, your review habit. That work is free, and no model upgrade does it for you.
My bottom line
Solomon’s remark will be read by some as dismissing AI. It is the opposite. It is permission for small businesses to stop chasing the newest model and start building workflows with the tools they already have. Spend your first hour on setup, not shopping. Upgrade only the one or two tasks where a mistake would really hurt.
Do you pay for a premium AI plan, and could you name the specific task that justifies it? Share it in the comments. If you cannot name one, that might be your answer.
