Building a Better Mousetrap Doesn't Build a Business

The proverb says, "Build a better mousetrap, and the world will beat a path to your door." It sounds inspiring. It also turns out to be wrong in a very specific, very expensive way.

The US Patent and Trademark Office has issued more than 4,400 patents on mousetrap designs, per Jack Hope's classic American Heritage piece and Vox's 2015 follow-up. Spring traps, glue traps, electric traps, CO2-powered traps, Bluetooth traps that text you when they catch a mouse. Roughly 40 new mousetrap patents are granted every year across 39 official subclasses. Fewer than two dozen of them have ever earned their inventors a cent. None have displaced the $1 wood-and-spring snap trap that William C. Hooker patented back in 1894. The world did not, in fact, beat a path to anyone's door. The world bought the cheapest thing that worked.

A patent fences off a specific mechanism. It does not, by itself, fence off a market, an audience, or a business — and that gap is why so many smart inventors end up holding valid patents and no customers. That gap is the part most founders, including a lot of engineers I've talked to, get wrong on the way in.

So in this post I wanted to look at the US patent system the way I think about anything else I'd consider building — not as a legal exercise, but as a product decision. What does a patent actually protect? What does it cost? When is it a moat, and when is it just an expensive piece of paper? And why has the mousetrap defeated every "better" version that's tried to replace it?

A patent is a fence around a mechanism, not a market

As the primer sketched, a patent is a 20-year exchange — full disclosure for the right to exclude. The part that took me longer to internalize is what that exchange doesn't buy you.

The thing the patent does not do is the thing every founder wants it to do. It doesn't make people buy your product. It doesn't stop competitors from solving the same problem with a different mechanism. It doesn't guarantee anyone cares.

That's why the mousetrap example is so useful. The 1894 Hooker snap trap is a spring-loaded bar that snaps when the mouse touches the bait pedal. Every "better" mousetrap since then has had to invent a different mechanism — electrocution, CO2, glue, motion sensing, Bluetooth — because the spring mechanism is taken. So the patent worked. It protected the spring. What it didn't do is protect the category. Anyone can still sell mousetraps. They just can't sell that one.

A patent is a fence around a specific way of doing something. The market is the land on the other side of the fence. You still have to plant something there.

The three filters every patent has to pass

Before any of this matters, your invention has to actually be patentable. US patent law (codified in 35 U.S.C. §§ 101, 102, and 103) puts every application through three filters.

Useful is the easy one. Almost anything that does something physical or computational passes. Abstract ideas and laws of nature don't count, but a working trap does, and so does a working algorithm.

Novel means nobody, anywhere in the world, has already disclosed this. "Disclosed" is broader than people realize — it includes earlier patents, academic papers, products on shelves, conference talks, and even your own blog post if you wrote it more than a year ago. The US gives you a 1-year grace period for your own public disclosures; most other countries don't, so a single tweet can destroy your novelty in Europe.

Non-obvious is where almost every battle is fought. The legal test asks: would a "person having ordinary skill in this technology" find this change obvious based on what already existed? About 85% of patent applications get a first-round rejection, and most of them are 103 rejections — obviousness rejections — not novelty ones.

This is where the mousetrap teaches the lesson again. Picture the inventor who comes in saying, "My mousetrap is novel — it has Bluetooth!" The examiner has read thousands of these. The examiner says: "Mousetraps exist. Bluetooth exists. Pressure sensors exist. A person skilled in IoT would find it obvious to combine them."

To get past obviousness, you don't need a totally new idea. You need a combination that produces an unexpected result. "We added Bluetooth" is obvious. "We added Bluetooth and discovered the timing of notifications shifts rodent behavior in a way that lets us predict the next catch within 24 hours" — that's the kind of unexpected result examiners actually allow.

For software engineers, this is the part worth internalizing. "I used machine learning to do X" almost always gets rejected as obvious. "I used a specific architecture to achieve a measurable result that prior art could not" has a real shot. The detail and the metric are the difference.

Once you file it, you can't change it

This is the rule I find most interesting from a product-thinking angle, because it goes against every instinct a software engineer has.

When you file, the description of your invention is frozen. You can argue with the examiner. You can narrow your claims. You can't add a single new technical detail that wasn't in the original filing. If you discover a better algorithm six months later, the patent doesn't cover it. You file a new one.

The first time I read this, my reaction was "that's an absurd constraint, why would anyone live with it?" After thinking about it from the system's side, I think the rule is actually doing the right job. Patents are public bargains. The deal is "you tell us how it works, we protect it." If you could keep updating the spec after filing, the deal collapses — competitors couldn't read your patent and know what they're up against.

What this means in practice is that filing a patent is more like releasing v1.0 of an open-source library than shipping a product behind a feature flag. You commit to what you've built. Improvements are new releases, with new versions, on new timelines.

The product-thinking move that comes out of this is the continuation strategy: founders who file seriously don't file once. They file the core invention, then file follow-up applications as they improve, each one sharing the original priority date for the parts that were already disclosed. Apple's iPhone touch interface isn't one patent. It's a family of patents, filed across years, layered on top of the original 2007 disclosure.

This is also why the cost question can't be answered with one number. The first filing isn't the real expense. It's the second, third, and fourth applications around it.

The cost is mostly the negotiation, not the form

The primer covered the headline numbers — $8,000 to $20,000 to grant, $25,000 to $40,000 over 20 years. What it didn't say is why most of that money goes to lawyers, not the government.

The actual government fee for a non-provisional filing can drop to $455 for a micro entity. The rest of the bill comes from the negotiation.

After you file, an examiner reviews the application and almost always sends back something called an Office Action — a list of rejections, objections, and requests for clarification. About 85% of applications get a first rejection — the normal part of the process, not a sign your idea is bad.

Each round of arguing with the examiner takes attorney hours. A patent attorney in a major city bills $400 to $800 an hour, and a software patent typically goes through two or three rounds of responses before allowance. That's where the $8,000-to-$20,000 figure comes from.

Then there's the long tail. Maintenance fees at 3.5, 7.5, and 11.5 years escalate sharply — the last one alone runs $7,400 for a large entity. A patent kept alive for the full 20 years racks up $12,000 to $15,000 in maintenance fees on top of everything else.

The actual product question isn't "should I patent this?" It's "is this patent worth $25,000 to $40,000 over its life?" That number gets sober fast.

The patent process is a negotiation, not a verdict

The 85% first-rejection rate is the single most useful statistic for understanding how the patent system actually works. Most founders treat the patent application like a college admissions essay — submit it, hope it gets approved, take a rejection as a final no. That's the wrong model.

The right model is closer to a multi-round negotiation with an expert reviewer who is paid to find prior art you missed. The examiner has tools you don't, sees thousands of applications, and is professionally skeptical. Their first rejection is almost never the last word.

The three rejections you'll actually see are:

• 102 (novelty): "I found a 1992 patent that already describes this." Your move: amend claims to narrow around the prior art, or argue the prior art doesn't actually disclose what the examiner thinks.
• 103 (obviousness): "Combining two existing references makes your invention obvious." Your move: show your invention produces an unexpected result the references don't predict.
• 112 (clarity): "Your claims are vague or aren't supported by your specification." Your move: rewrite the claims with more precise language tied to specific paragraphs in the spec.

Most allowed patents got there by narrowing the claims. Inventors come in wanting broad protection ("a method of using machine learning to predict customer behavior") and walk out with narrower protection that actually exists in the spec ("a method using a gradient boosting model trained on transaction history to predict 30-day customer churn with measurable accuracy improvement over prior art").

Narrower than what you wanted — but real, defensible, and yours.

The reframe that helped me here: the examiner isn't your adversary in the way a courtroom prosecutor is. They're more like a senior code reviewer who's read your PR before. Their objections are almost always reasonable, even when they feel frustrating. A good patent attorney is the person who helps you respond to the review without losing the parts that matter.

Patents become a moat only when you've got something on the land

Here's the part that took me longest to see. Patents matter to companies that already have something to protect. They rarely turn nothing into something.

This is what I wrote about a few weeks back when I was thinking about why building good products means understanding more than tech. A technical innovation that nobody asked for is just expensive practice. Wrap a patent around it and you've made the practice more expensive.

The companies that get real moats out of their patent portfolios — Apple, Qualcomm, IBM — already have the four other things every business needs: a product people want, distribution that gets it to them, a brand that earns trust, and the operational ability to keep delivering. The patents are the last layer, not the first.

A few examples of how this plays out in different shapes:

Apple patented the iPhone's touch interface, gesture recognition, and Retina display. None of these patents made the iPhone successful. The iPhone was successful because it actually worked, the App Store opened a software market, and people loved using it. The patents bought Apple roughly five years before Android caught up on touch interactions — which was enough time to lock in the brand and the ecosystem.

A reader might push back here: "But Apple filed those patents before the iPhone shipped — so didn't the patent come first?" Technically yes, the filings preceded the product launch. The point is that Apple already had everything else in place when they filed: a proven product team that had shipped iPods to hundreds of millions of customers, retail stores, a brand customers trusted to gamble on a $500 phone, and the operational ability to execute. The patents protected a mechanism inside an already-loaded gun. A solo inventor with the same patents and none of those other layers would have ended up exactly where the 4,400 mousetrap inventors ended up. The patent is what protects the moat once you have one. It isn't what creates the moat.

Qualcomm owns roughly 40,000 patents on wireless technology. Every smartphone manufacturer pays Qualcomm licensing fees. The portfolio is so dense — what's called a patent thicket — that designing around it is harder than just paying. That's a real moat, but Qualcomm built it over decades, on top of a long history of actually shipping wireless chips.

IBM files thousands of patents a year and uses most of them defensively. Its strategy is roughly "we won't sue you if you don't sue us" — a kind of mutually-assured-licensing posture. IBM cross-licenses with peers and rarely litigates. The patents are infrastructure, not weapons.

Biotech is the cleanest exception to all of this. In drugs and vaccines, the patent isn't protecting the moat, it is the moat. Stanford's Cohen–Boyer recombinant DNA patent (US 4,237,224, granted 1980) was licensed to 468 companies over its life and brought Stanford and UCSF more than $100 million in royalties — and arguably created the modern biotech industry. Bessen and Meurer's Patent Failure (Princeton, 2008) — the canonical academic case against the modern patent system — concedes the same exception: "only in some sectors, such as the pharmaceutical industry, do patents act as advertised." If you're building a drug, a vaccine, or a novel molecule, flip everything in this post upside down. The mousetrap rule applies to software. The Cohen–Boyer rule applies to biology.

Wendover Productions has a sharp explainer on why the biotech exception is the way it is — drug discovery economics, the FDA approval cliff, and what happens when you try to protect genes themselves:

Zoom out from the industry-specific cases and the bigger frame is even less flattering to the lone-inventor mythology the patent system encodes. Steven Johnson made the case directly:

Johnson argues breakthroughs emerge from networks — slow hunches colliding across people and decades — not from the lone inventor patents are built to reward. The biotech exception is real, but it sits inside a broader truth: in almost every domain, the act of inventing is not the scarce resource. Distribution, networks, and timing are.

And then there's the side of the system that gets less attention from product builders, but matters a lot for anyone shipping software:

Non-practicing entities (NPEs ), often called patent trolls, own patents but don't make or sell anything. They make money by sending demand letters and threatening to sue. According to Unified Patents' 2025 mid-year report, nine of the top ten US district-court patent plaintiffs in the first half of 2025 were NPEs, and the Eastern and Western Districts of Texas alone accounted for 63.7% of all NPE cases filed during that window. NPE litigation is up roughly 20% year-over-year. A typical NPE buys patents in bulk from bankrupt companies or auctions, then identifies operating companies that might be infringing, then sends a letter. Defending against an NPE in court can run $1 million to $5 million even if you win.

This is the strongest argument for the defensive side of patent strategy: even if you don't plan to enforce a patent, owning one gives you something to put on the table when an NPE shows up. Cross-license, settle without paying cash, or join a defensive aggregator like the LOT Network. The patents you file aren't always there to attack with — sometimes they're there to put down on the table when someone else attacks first.

The most public counterexample I've seen — and the one engineers will recognize — is Cloudflare's fight against Blackbird Technologies. Cloudflare ran a crowd-sourced bounty program called Project Jengo to hunt for prior art across Blackbird's entire portfolio, paid out $50,000+ to 18 submitters, won the case, and effectively put Blackbird out of business. Cloudflare's general counsel documented the whole thing publicly:

📺 Worth reading: How Cloudflare stood up to a patent troll — and won — Cloudflare's GC walks through the Project Jengo crowdsourced prior-art bounty, the win against Blackbird Technologies, and the more recent 2024 trial victory against Sable Networks.

The reason this story matters for the rest of the post is that it shows the system isn't entirely captured by NPEs. With enough resources and enough public will, operating companies can push back. Most startups won't have Cloudflare's budget, but the defensive-aggregator option (LOT Network and others) lets smaller companies pool the same kind of leverage.

When a patent is actually worth filing

Putting this together, the question I'd ask myself before filing isn't "is this patentable?" It's "what does this patent let me do that I can't do without it?"

Three of these branches end at "probably not." That's intentional. The default for most software products should not be "file a patent." It should be "ship the product, build distribution, see if you have something worth protecting first."

The branches that do end in "file" share a pattern. There's a real product, the invention is hard to design around, the cost is small relative to the moat, and there's a concrete reason — fundraising, exit, or active legal threat — that makes the patent earn its keep.

A few situations where filing actually pays off:

• You're a hardware or biotech company. Patents are how the entire industry values you, and investors will look for them.
• Your invention is the kind of mechanism a competitor can copy in a weekend if it's not protected. (Software algorithms with measurable, non-obvious results, hardware-software combinations, novel chemical processes.)
• You expect to be acquired and the buyer will pay more for IP than for the company alone.
• You're at risk of being targeted by an NPE and need patents to cross-license your way out.

A few where it doesn't:

• Your moat is actually distribution, brand, network effects, or speed. Those aren't patentable, and a patent on something else doesn't help you.
• Your invention is easy to redesign around. Most "method of doing X with software" patents fall here.
• You can't afford the maintenance fees, or you're going to forget to pay them. About a third of all granted patents are abandoned for unpaid maintenance.
• You're the kind of company that would never sue anyone over a patent. Filing then becomes purely defensive — cheaper to publish defensively and skip the system.

The interesting move I keep noticing in the engineer-founders I admire is defensive publication . If you've built something clever but don't think it's worth patenting, write it up and publish it — a blog post, a Stack Overflow answer, a paper, a public spec. Once it's public, nobody else can patent it either. You've protected your freedom to operate without paying the patent system a dollar.

Try this: before drafting your next provisional, walk your current invention through the four-branch filter above. Most software ideas land at "probably not yet" — and that's the right answer. Keep building the product, and revisit the patent question after you have customers.

If you do end up at "file", the next two posts get tactical: how to read patents so you know what's already protected, and the international time game so you don't burn $30,000 on day one.

The mousetrap, one more time

So back to the question we started with: why have 4,400 mousetrap patents not displaced the $1 snap trap, and fewer than two dozen earned their inventors a cent? It isn't that the patents are bad. It's that they're the answer to the wrong question.

Every "better mousetrap" inventor has thought about novelty, non-obviousness, and utility. They've drafted claims, paid filing fees, argued with examiners, and gotten patents. The system worked exactly as designed. What none of them solved is the actual product problem: a mousetrap costs $1, customers are price-sensitive, distribution runs through hardware stores and grocery aisles, and the snap trap is good enough. A patent on a Bluetooth mousetrap doesn't change any of that. It just protects the Bluetooth.

The lesson, I think, is one I keep running into in different shapes. The hard part of building something is almost never the technical novelty. It's understanding the customer, the price they'll pay, the distribution that reaches them, and the brand they'll trust. Patents protect mechanisms. Markets reward all the other things.

For software builders specifically, that means treating a patent the way you'd treat any other tool in the stack — useful when it solves a problem you actually have, expensive when you bolt it on because it sounds important.

What the mousetrap taught me

• A patent is a fence around a specific mechanism, not a fence around a market — anyone can solve the same problem with a different mechanism
• The three filters are useful, novel, and non-obvious; almost every battle in the patent process is fought on the obviousness rejection
• "I added ML" is obvious; "I used a specific architecture to get a measurable, unexpected result" has a real shot
• Once you file, the spec is frozen — improvements need new applications, which is why serious patent strategy is a family of filings, not a single one
• The real cost of a patent isn't the filing fee, it's the attorney hours during prosecution and the maintenance fees over 20 years — call it $25,000 to $40,000 all-in
• The 85% first-rejection rate is normal — patent prosecution is a multi-round negotiation, not a pass/fail test
• Patents become a moat only when you already have a product, customers, and distribution — they're the last layer, not the first
• Patent trolls are a real cost of doing business in the US — owning a small defensive portfolio (or joining a defensive aggregator) gives you a seat at the negotiating table
• Defensive publication is a free, underused alternative — if you've built something clever and don't want to patent it, publish it and lock it out of the patent system entirely
• The default answer for most software products should be "don't file yet" — ship first, see what's actually worth protecting, and let the patent decision follow the product

Build something the market wants, then fence it off. Most founders, including most engineers, want it in the reverse order — build the fence first and hope the market shows up. The mousetrap has 4,400 fences and nobody can find anything worth fencing in. The market doesn't reward the fence. It rewards the thing on the land.

Where these numbers come from

• 35 U.S.C. §§ 101, 102, 103 — the statutory definitions of utility, novelty, and non-obviousness. Used for: the three patentability filters.
• USPTO Fee Schedule (uspto.gov) — Used for: filing, search, examination, issue, and maintenance fee numbers.
• USPTO Manual of Patent Examining Procedure (MPEP) — Used for: how examiners evaluate Office Actions, the 102/103/112 framework, and the role of claim narrowing.
• Michael Meyer Law, "How Much Does a Patent Cost" — Used for: realistic attorney drafting and prosecution cost ranges.
• Jack Hope, "A Better Mousetrap," American Heritage (October 1996) — Used for: the 4,400+ mousetrap patent count, the ~40 new patents per year across 39 subclasses, and the 95%-amateur-inventor breakdown.
• Vox — "7 horrifying attempts at building a better mousetrap" (Phil Edwards, July 2015) — Used for: the "fewer than two dozen have earned their inventors a cent" stat and modern photo examples.
• Wendover Productions — "The Gene Patent Question" — Used for: the biotech-exception explainer embedded in the post.
• Google Patents (US10440944B2) — Used for: the Victor Smart-Kill Bluetooth trap as a real example of an IoT mousetrap patent.
• Gifford Pinchot III, Intrapreneuring — Used for: the framing that a successful innovation needs more than a clever mechanism, drawn through earlier posts in this series.
• Unified Patents, "Patent Dispute Report: 2025 Mid-Year Report" — Used for: the "9 of the top 10 US district-court plaintiffs are NPEs" figure, the 63.7% Texas concentration, and the ~20% year-over-year increase in NPE litigation.
• Cloudflare Blog, "The Project Jengo Saga" and "Cloudflare defeats patent troll Sable at trial" — Used for: the Project Jengo bounty program, the $50,000+ paid to prior-art submitters, the Blackbird Technologies win, and the 2024 Sable trial victory.
• Reports on non-practicing entities (NPEs) — Used for: the typical cost of defending against NPE litigation in court.