A listed conglomerate in the news has set up an AI project office, earmarked a 20 million budget, brought in three consulting firms, and put out a press release.
Then you look down at your own company. Six people, maybe eight. The accountant still works in spreadsheets, half the customer data is sitting in LINE chat histories, and last year you talked once about whether to bring in a CRM, and nothing came of it. The conclusion comes naturally: once they've finished doing this, we'll follow along.
That conclusion points the wrong way.
01The one being held back is the big ship
In May 2026, the Silicon Valley 101 podcast interviewed the three founders of CreaoAI, a Silicon Valley AI company. Most of the conversation was about harness engineering, AI-first organizations, and development speed, but the most counterintuitive line came near the very end.
Asked which kind of company this approach suits best, they didn't say the Fortune 500. They said small and mid-sized businesses. The reasoning was plain: big enterprises carry a lot of compliance issues, a lot of legacy systems, and a lot of people factors, so transformation is hard going; small businesses have a light compliance burden and no old systems to tear out, which makes them the first and easiest companies to complete the transformation.
This isn't meant as consolation. It's a cost calculation.
The cost of adopting AI has long been misread as a procurement cost: buy a model, buy tools, find a vendor. What really eats up the time is a few other line items. Reconciling data scattered across different systems, getting everyone involved to agree to change how things are done, having someone take responsibility when something goes wrong. Every one of these grows with the size of the organization, and grows out of all proportion.
A fifty-year-old manufacturer has six systems that don't talk to each other, three departments with veto power over the whole thing, and an information security policy written in 2019. An eight-person company has one spreadsheet and one owner.
02Advantage one: the decision chain is only one layer deep
CreaoAI did something that would border on offensive in a traditional organization: it removed the product manager role.
Not because PMs have no value. They found that a large share of their alignment cost landed precisely on that role, and once the role was gone, alignment cost actually went down. The judgment that used to belong to the PM was split back out to every engineer. Their day now looks like this: write a feature at 10 a.m., run an A/B test at noon, cut part of it based on the data at 3 p.m., and rewrite a better version by 5. That one day equals six weeks of a traditional development process.
The point of this example isn't the multiplier, "six weeks squeezed into a day." The point is why the multiplier holds: between "having the idea" and "making the call," nobody has to be convinced.
And that is the default state of a small business.
A ten-person company that wants to try a new quoting process doesn't need a cross-department meeting, doesn't need to go through an approval chain, doesn't need to prove to anyone that the proposal deserves a slot next quarter. The owner nods, and it runs tomorrow. The same thing has to become a project at a big company. At a small one, it's just a decision.
In the past, this was called "having no formal process." In this round of AI, its name is iteration speed.
03Advantage two: no legacy systems to tear out
There's a widely shared figure about enterprise AI: nine out of ten projects die at the implementation stage. The five obstacles usually listed are vague requirements, poor data quality, compute costs, system integration, and nobody driving it, and the one in the middle is especially critical. The same summary concludes: the hard part is integration, not the interface.
Anyone can build an interface. The hard part is letting AI read your ERP, your customer service records, your inventory sheet — three things that, at a big company, belong to three departments, three permission setups, and three formats, with an old database in between that nobody wants to touch.
A small company is in exactly the opposite position. Little data, few systems, formats that are messy but readable. At an eight-person company, all the customer data might be one spreadsheet plus one group chat. That sounds like being behind, but when it comes to adopting AI, it means this: data governance, a project that takes a whole department and half a year at a big enterprise, is an afternoon of tidying up here.
No legacy, no legacy baggage. When companies take stock of what they have, this one almost never makes the list.
04Advantage three: only a few people to convince
That interview split the use of AI into two mindsets, with an order of magnitude between their efficiency ceilings.
The first treats AI as an efficiency tool: people still do the work, and AI helps speed it up. The ceiling on that path is roughly ten times, because a person only has twenty-four hours in a day. The second makes AI the driver of production: AI leads the work, AI assigns tasks to people, and people review the results. Co-founder Kai believes that path is a hundred to a thousand times.
And all three founders agreed that the hardest part of crossing that line isn't the technology. It's trust: getting every employee to believe that the decisions, plans, and execution AI produces are worth adopting. That takes guardrails, and it takes time.
Now put that problem on the scale at companies of different sizes.
At a three-hundred-person company, "getting everyone to trust AI" is an organizational change that needs an internal communications strategy, training, demo projects, and revised KPIs. Middle managers will instinctively resist, because part of their job is relaying information, and that is exactly the part that gets taken over first.
At an eight-person company, it's one conversation over lunch.
05The price of being small: three things that really do trip up small firms
If this piece ended here, it would be a cheap pep talk. So the other side needs saying too.
Nobody drives it. The last of the five obstacles does the most damage, and it hits small companies most often. A big company will at least assign someone to own it. A small company usually doesn't have that person, because everyone has their own job to keep up with, and AI is forever "let's talk about it next week." This is the real cost of being small: decisions are fast, but nothing keeps the momentum going.
The owner is the bottleneck. The flip side of a one-layer decision chain is that if that one layer doesn't move, nothing moves. At a big company, middle management can push a project forward before the owner fully understands it. At a small company, that can't happen. The owner's ceiling of understanding is the company's ceiling.
Nobody builds the guardrails. Letting AI lead decisions presupposes a mechanism that can stop it when it gets things wrong. Big companies have compliance, audit, and information security. Small companies have none of it, and "having none of it" is an advantage early in adoption. Once you reach the point of letting AI execute on its own, it becomes a risk.
One more thing needs to be stated honestly. CreaoAI is a Silicon Valley AI company: twenty-five people, fewer than ten engineers, a product that is itself an agent platform, and 99% of its code written by AI. That ratio doesn't mean a Taiwanese trading company or design firm can replicate it. What they proved is that the path works and where it gets stuck, not a set of numbers you can copy.
The timeline needs a discount too. Reviewing the nearly 80,000 tech-industry layoffs in the first quarter of 2026, Babak Hodjat, Chief AI Officer at RationalFX, pointed out that most of those decisions were "expectation-driven," not a sign that productivity gains had already materialized; the real benefits usually take six to twelve months to show up. A small boat turns easily, but what it turns is its heading, not gravity.
Where to start
The test for a first project is actually simple: low-sensitivity data × high frequency × something a person can check. The job where all three conditions hold at once is the real work best suited to hand to AI first. Turning a meeting recording into decisions, to-dos, and owners is a textbook example.
The checking part is the one most easily skipped, and the one you can least afford to skip. The test of a real rollout isn't "it went live." It's "people are still using it." And people keep using a tool only because they can check its work.
The order after that has already been laid out in earlier pieces in this series: first confirm which step you're standing on, then decide what to buy. The advantage of a small boat is that it turns quickly, but not a single step on the staircase can be skipped.
Let AI move fast. Let people hold the anchor. At an eight-person company, that's far easier than at an eight-hundred-person one.
We're fortunate to live in this era of computers, the internet, and AI.
Before It took ten people to keep a company running
Now A sharp, focused team with AI behind it is enough to reshape a brand
The premise of that line was never that small companies are better. It's that this round's barrier happens to sit exactly where big companies can't tear things down: legacy systems, approval chains, departmental walls. And those are things you happen not to have.
The window won't stay open forever. Big enterprises need time to tear down their walls, but they will finish eventually. In the time before that, the companies that move first will open up the gap.
One team, from strategy to launch.
Rooted in craft. Built for the new wild.