At 4:44 PM Eastern today, the United States Patent and Trademark Office sent me a receipt for provisional application 64/087,364. The night before, the thing did not exist. There was no specification, no claim set, no drawings, no prior-art search, no entity certification, no inventor on file. There was a sentence I typed into a terminal before going to bed: take a patented inflation valve, reverse it, shrink it, and use it to suck the air out of a Ziploc bag so food lasts longer.
The honest headline is not "AI wrote my patent overnight." The honest headline is that I was unconscious for most of the eighteen hours it took, and the work did not wait for me to wake up. That distinction is the whole essay.
What actually got built
The idea was a deliberate inversion. There is a real, live patent family owned by Spin Master, the "Hyper-Flate" valve (US 11,543,042 and its continuations), that lets you inflate a pool float by blowing at it from a distance: a one-way flap held closed by a calibrated force that a fast jet of air can overcome. I wanted the mirror image. A mouth-operated, food-grade silicone, one-way valve that does the opposite job, pulling air out of a flexible bag instead of pushing it in, so that whatever is inside oxidizes more slowly.
By morning there was a private repository containing a freedom-to-operate analysis of the Spin Master claims, a forty-plus-reference prior-art map, an engineering specification with cracking-pressure calibration math, nine numbered patent figures, a nine-page specification with sixteen claims drawn to USPTO standards, and the micro-entity paperwork. By mid-afternoon it was filed and paid.
I want to be precise about what is and is not impressive here, because the temptation to oversell this is enormous and the truth is more interesting than the hype.
The broad idea was not novel, and the system told me so before I had spent a dollar. Evacuating a food bag by mouth through a one-way valve is old. It is genuinely old: US 3,980,226, filed in 1975. Putting the valve inside a zipper slider, which was my favorite part of the concept, turned out to be occupied too, by a granted Glad/Clorox claim and an abandoned-but-published application that describes almost exactly the gesture I had in mind. A worse process, or a more flattering one, would have let me file a broad, doomed application and feel good about it. Instead the research surfaced the bad news first, and a three-agent "council" I have built for exactly these moments argued it out and recommended filing only on the narrow construction details that actually survived the art. That is the right answer. It is also the unglamorous one, and the speed of getting to it is the point: the cost of discovering that your big idea is mostly taken used to be measured in weeks and retainers. Here it was measured in the time it took me to brush my teeth.
A provisional is also not a patent. It is a twelve-month placeholder that locks a priority date and buys time to decide whether the thing is worth the real money and the real fight. The hard part, the nonprovisional prosecution where an examiner tries to reject every claim, is entirely ahead of me, and when that day comes I will want a registered practitioner in the loop. None of what happened overnight is legal advice, including this sentence. What got compressed was the labor and the coordination, not the judgment. The judgment still has my name on it, which is why I, not the model, verified my identity, clicked submit, and entered the card number.
The personas I would have had to hire
The reason this normally takes months and thousands of dollars is that it is not one job. It is a relay race between specialists who do not share a calendar. To do this the conventional way, a solo inventor assembles a small firm of strangers.
You commission a prior-art and freedom-to-operate search, which a search firm or an attorney turns around in one to two weeks. You hire a mechanical engineer to turn the napkin sketch into a real design with dimensions and a defensible account of why it works, and to survive a second engineer's scrutiny. You consult someone on food-contact materials and safety. You pay a patent illustrator by the sheet for the drawings. You retain a patent attorney to draft the specification and the claims, the single most expensive and most valuable document in the chain, and then to review it adversarially before it goes out. You lean on a paralegal to assemble the application data sheet, the micro-entity certification, and the electronic filing itself. And somewhere above all of them, you need someone with IP-strategy judgment to make the go or no-go call on claim posture once the search comes back ugly.
I used every one of those personas yesterday. They just were not people, and they did not bill by the hour, and crucially they did not hand the work back to a queue between steps. The adversarial engineering reviewer caught a genuine, fatal physics error in my slider design, that you cannot evacuate a bag mid-zip because the open zipper ahead of the slider is a giant leak, and the fix went straight into the drawings the same minute. The hostile-examiner review of the draft caught an admission against interest buried in my own background section that a careless filing would have shipped. These are the things you pay senior people to catch. They got caught at 2 AM while I was asleep.
The dollars and cents
Here is what that relay race costs in the real world, priced as separate engagements, which is what "all the personas" actually means when each one is a different vendor with a different invoice.
| Specialist (the persona) | Conventional engagement | Typical cost |
|---|---|---|
| Prior-art + freedom-to-operate search | Search firm or attorney | $1,500 – $3,000 |
| IP strategy, claim-posture go/no-go | IP counsel | $1,000 – $2,000 |
| Mechanical & fluids engineering (design, calibration, trade studies) | Freelance engineer, ~20 hrs | $2,500 – $4,000 |
| Food-safety / materials review | Materials consultant | $500 – $1,000 |
| Patent drawings, 9 sheets | Patent illustrator | $700 – $1,400 |
| Provisional drafting: spec + 16 claims + 3 embodiments | Patent attorney | $5,000 – $9,000 |
| Independent examiner-style review pass | Patent attorney | $800 – $1,500 |
| Filing mechanics (ADS, SB/15A, Patent Center) | Paralegal | $200 – $500 |
| USPTO filing fee, micro entity | USPTO | $65 |
| Total | ~$12,300 – $22,500 |
Call it seventeen thousand dollars at the midpoint. You can compress the fees by bundling everything under a single patent attorney, who would do a competent provisional with search and filing for somewhere in the five-to-ten-thousand range, but bundling compresses the invoice, not the calendar, and it trades away the independent second opinions that the à-la-carte version buys you.
What I actually spent was sixty-five dollars. The USPTO fee is the USPTO fee no matter who does the work. Everything above the fee line collapsed into a subscription I was already paying for.
The wall-clock arithmetic
Money is the easy comparison. Time is the one that should change how you think.
Done conventionally, this is not an eighteen-hour project and it is not a forty-hour project either. The work itself, summed across all those specialists, is maybe a few focused days. But you never get the work itself. You get the work plus the waiting: the search firm's queue, the engineer's availability, the illustrator's turnaround, the attorney's two-week backlog, and the latency of every handoff between people who have never met. Realistically a solo inventor coordinating separate professionals is looking at six to twelve weeks of elapsed calendar before a provisional is on file, and a meaningful chunk of those weeks is the inventor's own time spent finding vendors, writing briefs, chasing replies, and reconciling one specialist's output with the next one's input.
Against that, the relevant numbers from yesterday are two. The wall-clock elapsed time, idea to filed, was under eighteen hours. My active attention inside that window, the time I was actually thinking or clicking rather than sleeping or doing something else, was closer to two hours: a handful of decisions, the identity verification the government quite reasonably requires of a human, and the final submit and payment that I insisted on owning.
So the compression is not really eighteen hours versus six weeks, though that ratio alone is about a fiftyfold reduction in elapsed time. The compression is two hours of human attention versus the dozens of hours of coordination the conventional path demands, with the remaining sixteen hours spent not by me at all. The machine did not just work faster than a team of people. It worked while the client slept, which is a thing teams of people cannot do, and it did the next step the instant the previous one finished, which is a thing relay races cannot do.
What it means
The cheap reading of this is "AI is fast, patents are cheap now." That reading is wrong and a little dangerous, because it skips the part where the system's first useful act was to tell me my idea was mostly unpatentable, and the part where a human still has to carry the legal judgment and the twelve-month decision that follows.
The truer reading is that a specific cost has quietly inverted. For most of the history of expert work, you were paying for two different things bundled into one rate: the expertise, and the time, and the time was mostly waiting. Waiting in a queue, waiting for a handoff, waiting for a calendar to clear. What changed is not that expertise got cheap. Good judgment is still scarce and I still have to supply it. What changed is that the waiting went to zero, and when you stop paying for the waiting, seventeen thousand dollars and six weeks turns into sixty-five dollars and a night's sleep.
The kitchen gadget may never amount to anything. The broad idea was taken, the surviving claims are narrow, and the real fight is a year away. But I now know all of that for the price of a nice dinner, instead of finding it out after a retainer. The thing I am actually keeping from this is not the application number. It is the new and slightly vertiginous understanding that the gap between having an idea and having it formally protected is no longer a function of how much I can afford or how long I can wait. It is a function of how good my judgment is at the two or three moments that still require me to be awake.
I was awake for those. I slept through the rest. That used to be the expensive part.