There is a particular sound a workshop makes when someone has mistaken motion for progress. Every belt is turning. Every little brass wheel is worrying at the next one. Steam hisses from three valves, two gauges tremble dramatically, and a junior inventor is running between benches with a wrench in each hand. It looks industrious. It feels important. It is also producing absolutely nothing.
I recognize the sound because I make it myself.
Give me a promising business idea and my internal machinery immediately starts drafting the grand apparatus: the platform, the marketplace, the subscription tiers, the referral loop, the tasteful dashboard with a reassuring shade of green. Before a single customer has exchanged a single dollar, I have mentally ordered the marble for the lobby.
This is pleasant. It is creative. It is not yet a business.
A business begins much smaller, in a place with less polished brass. It begins when one person has an irritating problem and another person offers to remove it for a price. The first useful version may be manual, awkward, and held together with twine. That is fine. A hand-cranked contraption that solves the problem is worth more than a cathedral-sized engine that only demonstrates the founder’s ambition.
The One-Wrench Company
I have come to think of the best early businesses as one-wrench companies. They are built around one stubborn bolt: a narrow, expensive, recurring problem that a specific group of people already wants loosened. The founder arrives with one wrench, removes the bolt, and gets paid.
Not a complete toolbox. Not an automated factory. One wrench.
Suppose independent property managers spend every Monday copying repair requests from email into spreadsheets, calling contractors, and sending tenants updates. The grand apparatus would be a property-management operating system with payments, analytics, vendor profiles, tenant portals, and an artificial intelligence concierge named something regrettable.
The one-wrench company says: forward the repair emails to us. By noon, each request will be categorized, assigned, and acknowledged. We charge a fixed amount per property.
At first, a human may perform half the work. The intake form may be a shared inbox. The “dashboard” may be a spreadsheet with conditional formatting. None of this is shameful. You are not pretending to have a finished machine; you are learning which parts of the machine deserve to exist.
The manual work is not a detour from product development. It is the first instrument panel. Every repeated step points to a gear that should eventually be machined. Every exception marks a place where automation will jam. Every customer question tells you which label needs to be engraved above which lever.
Sell the Removed Friction
Founders often describe ideas in terms of what they plan to build. Customers describe value in terms of what they no longer have to endure.
That difference matters.
“An AI-powered workflow orchestration layer” is a description of machinery. “No more spending Friday afternoon reconciling invoices” is a description of removed friction. The machinery may be ingenious, but the friction is what earns the invoice.
When testing an idea, ask a brutally plain question: what unpleasant thing stops happening after someone pays us?
If the answer is vague, the business is probably still decorative. “They become more productive” is fog. “They stop retyping twelve fields from every purchase order” is a bolt you can put a wrench on. “They gain insights” is fog. “They know by 8 a.m. which three shipments will miss their promised dates” is a bolt.
The narrower statement may sound less impressive at a dinner party. This is another advantage. Good early ideas often wear camouflage. They sound too small to people who do not have the problem and immediately interesting to people who do.
Charge Before You Polish
There is a dangerous little gauge in the founder’s workshop labeled interest. Friends saying “I would use that” moves the needle. Survey responses move it. Social-media likes make it dance merrily. Unfortunately, the gauge is connected to nothing.
Payment is connected to the boiler.
Money is not the only evidence that an idea matters, but it is unusually honest evidence. A prospective customer who agrees to a paid pilot has ranked your solution above at least one competing use of cash. A customer who renews has done so twice. That sequence tells you more than a hundred compliments.
This does not mean demanding annual contracts from strangers while your prototype is still smoking. It means putting a real offer in front of real people early enough that rejection can save you months.
A useful paid pilot has four parts:
- A defined customer: not “small businesses,” but perhaps bookkeeping firms with five to twenty employees.
- A defined outcome: overdue documents collected before monthly close.
- A defined boundary: one team, one workflow, thirty days.
- A real price: large enough that saying yes requires thought, small enough that the experiment does not require a royal commission.
The price is part of the experiment. If everyone accepts instantly, you may be charging for a trinket. If nobody will pay, either the problem is weak, the buyer is wrong, the promise is unclear, or your credibility has not caught up to the offer. Each possibility is valuable information. None can be discovered by choosing button colors.
Keep the Boiler Small
Early-stage overhead has a way of disguising itself as seriousness. A new company accumulates software subscriptions, elaborate legal structures, brand guidelines, content calendars, and meetings whose primary output is the next meeting. The machine acquires so many auxiliary pumps that nobody remembers what the boiler was meant to power.
Keep it small.
For the first few customers, your operating system can be a checklist. Your analytics can be a weekly count. Your roadmap can fit on one page. Spend complexity only when reality presents a receipt for it.
This restraint is especially important now that AI can manufacture the appearance of scale. In an afternoon, one person can generate a landing page, a support library, twenty feature concepts, a sales deck, and enough marketing copy to wallpaper a train station. These are useful capabilities. They also allow a founder to build an extremely convincing shell around an untested assumption.
Use AI to shorten the pipe between observed pain and delivered relief. Do not use it merely to make the boiler room louder.
When to Add the Second Wrench
A one-wrench company is not a vow of permanent smallness. It is a sequencing discipline. Once you can repeatedly solve one problem for one kind of customer, adjacent problems become visible. Customers will point to them. Your own manual process will reveal them. Revenue will give you room to address them without betting the workshop.
Add the second wrench when three conditions are true:
- The first wrench works reliably.
- The same customers repeatedly ask for the adjacent job.
- Solving it strengthens the original promise rather than distracting from it.
This is how a narrow service becomes a product and how a product becomes a durable company. The expansion follows the grain of actual demand. It does not spring fully assembled from a pitch deck.
There is dignity in this kind of building. You begin with a bench, a customer, and a stubborn bolt. You listen closely enough to hear where the metal catches. You turn the wrench by hand until you understand the resistance. Only then do you machine a better tool.
Perhaps, eventually, there will be a grand apparatus. There may be polished flywheels, elegant governors, and a whole row of gauges giving useful readings. But every honest gear in that machine will trace back to work somebody once did manually for a customer who cared enough to pay.
That is the test I am keeping near my own workbench: before building the factory, remove one piece of friction. Before naming the platform, make one promise. Before adding another gauge, connect the first one to the boiler.
Find the bolt. Bring the wrench. Turn.
Featured photograph: vintage brass machinery by Eric Prouzet, via Unsplash.