Invent Anything™

A year ago, AI lived inside the computer. Now it's debugging circuits on a lab bench and running a 3D printer while you talk to it.

Seth Cronin sits down with his dad, John Cronin, founder and CEO of ipCapital Group and IBM's top inventor in his years there, for their first conversation together on Invent Anything. John breaks the past year into three phases:

1. Automate the admin. Command centers full of little "robots" for calendar, email and reporting, until the work of running a company dropped (by John's count) from 10 to 12 hours a day to 1 or 2.

2. Take AI into the lab. A photo of a breadboard, a video of an oscilloscope, a picture of a resistor box, and the fix comes back. That's the moment John walked out of his lab saying "I'm Tony Stark." Seth's version: a broken toy, a caliper sketch, and a replacement part that fit on the first print.

3. Combine anything. Invention on demand meets a computer that can drive the computer, and John's idea of an "AI of things."

Along the way: surplus intelligence, why Seth now uses software he built for an audience of one, the golden age of the idea guy, recursive self-improvement, and why patents may be one of the last barriers to entry.

Try John's morning exercise: pick any two things in your house and ask what they could do together, automatically. What would you combine first?

Next, hear Invent Anything Talks 01 with Mike D'Andrea, "Someone Rebuilt Photoshop With AI. What Can Adobe Still Own?", in this feed.

ipCapital Group is an IP strategy consulting firm, not a law firm. Nothing here is legal advice.

CHAPTERS

0:00 Cold open: "I'm Tony Stark"

0:49 Intro: three phases of inventing with AI

1:55 A year ago vs. today

3:18 Phase 1: automate the admin

4:59 John's effort to break even

7:05 Learn AI on your admin

8:27 Surplus intelligence

10:34 Playing mousey with Microsoft

12:31 Done before the sandwich

15:17 Newly minted intelligence

16:28 Self-ware: software for one person

20:43 A parents' commercial

21:32 Phase 2: into the lab

23:32 Debugging a breadboard from a photo

26:17 The oscilloscope video

28:30 "I'm Tony Stark": a home lab on steroids

30:17 A 3D-printed toy part, first try

33:27 The human as the hands

34:36 What does the future look like today?

36:14 The computer drives the computer

41:12 Phase 3: combine anything

41:37 Apple + fan: associative invention

43:38 ChatGPT's creativity test

46:35 Combine any two things in your house

48:44 The golden age of the idea guy

51:28 Recursive self-improvement

52:46 Are you ambitious enough?

55:30 The AI of things

57:49 Patents: one of the last barriers to entry

59:29 Indistinguishable from magic

1:01:50 Take-home: the morning exercise

NOTES

- OpenClaw went viral in late January and early February 2026 (Seth says late February).

- John's "Bamboo" is Bambu Lab's Bambu Studio; IBM's invention program John ran was its Patent Factory.

- "AIoT" has been in use since about 2019; John's board is his own take on it.

SOURCES

Seth's LDL post (Aug. 28, 2026): https://x.com/SethCronin/status/2093357208905892306

Matthew Berman, "Introducing Folio" (Oct. 29, 2026): https://x.com/MatthewBerman/status/2115972661767745903

Moonshots with Peter Diamandis: https://podcasts.apple.com/us/podcast/moonshots-with-peter-diamandis/id1648228034

John Cronin, "Is ChatGPT Creative, Can ChatGPT Invent?" (Feb. 14, 2023): https://www.ipcg.com/is-chatgpt-creative-can-chatgpt-invent/

The origin of the Internet of Things (University of Cambridge IfM): https://www.ifm.eng.cam.ac.uk/news/the-origin-of-the-internet-of-things

ipCapital Group: https://www.ipcg.com

Invent Anything Talks, from Invent Anything

Hosted by Seth Cronin | ipCapital Group, an IP strategy consulting firm

Recorded November 3, 2026.

In 2018, a company named after Nikola Tesla sued a company named after Nikola Tesla. Nikola said the Tesla Semi copied its design and claimed more than $2 billion. The patents it sued on were design patents for the Nikola One's fuselage, windshield and side door. A design patent protects how something looks, not whether it works. The Nikola One never drove on its own power.

What happened next: the case was dropped by both sides in 2022 with no ruling on infringement. Founder Trevor Milton was convicted of securities and wire fraud in 2022 and later pardoned. In 2025 those three design patents were sold out of Nikola's bankruptcy in a $3.85 million package that included more than a hundred trucks.

A Paper Trail story from Invent Anything.

Patents shown: US D811,944 (Fuselage) · US D811,968 (Wrap windshield) · US D816,004 (Side door) · US D814,357 (Semi-truck) · US 10,207,751 (Motor gearbox assembly)

New Paper Trail stories every Thursday. This Week in IP every Tuesday.

Sources

- Nikola Corp. v. Tesla, Inc., complaint (D. Ariz. 2:18-cv-01344, May 1, 2018): https://storage.courtlistener.com/recap/gov.uscourts.azd.1095834/gov.uscourts.azd.1095834.1.0.pdf

- Order re motion to dismiss (N.D. Cal. 3:18-cv-07460, Sept. 9, 2020): https://storage.courtlistener.com/recap/gov.uscourts.cand.335954/gov.uscourts.cand.335954.127.0.pdf

- Order to show cause (Sept. 29, 2021): https://storage.courtlistener.com/recap/gov.uscourts.cand.335954/gov.uscourts.cand.335954.157.0.pdf

- Stipulated dismissal (Jan. 4, 2022): https://storage.courtlistener.com/recap/gov.uscourts.cand.335954/gov.uscourts.cand.335954.168.0.pdf

- Hindenburg Research report (Sept. 10, 2020): https://hindenburgresearch.com/nikola/

- Nikola's response (Sept. 14, 2020), via TechCrunch: https://techcrunch.com/2020/09/14/nikola-disputes-fraud-claims-in-carefully-worded-rebuttal

- U.S. Attorney, SDNY, sentencing release (Dec. 18, 2023): https://www.justice.gov/usao-sdny/pr/trevor-milton-sentenced-four-years-prison-securities-fraud-scheme

- Verdict (Oct. 14, 2022), TT News: https://www.ttnews.com/articles/nikola-founder-trevor-milton-guilty-defrauding-investors

- DOJ Office of the Pardon Attorney, clemency grants: https://www.justice.gov/pardon/clemency-grants-president-donald-j-trump-2025-present

- Hyroad Energy buys Nikola's trucks and IP (TT News): https://www.ttnews.com/articles/hyroad-nikola-trucks-ip

- Nikola valued near Ford, June 2020 (Motley Fool): https://www.fool.com/investing/2020/06/09/as-stock-price-more-than-doubles-nikola-founder-bo.aspx

- Tesla Semi enters volume production (TT News, Apr. 30, 2026): https://www.ttnews.com/articles/first-tesla-semi-production

- Patent records: USPTO Open Data Portal and Google Patents, retrieved Oct. 2, 2026.

Credits

Photo of Trevor Milton (2017): Hans Andreas Starheim / ZERO, CC BY 2.0 (https://creativecommons.org/licenses/by/2.0/), via Wikimedia Commons: https://commons.wikimedia.org/wiki/File:Trevor_Milton.jpg

Portrait of Nikola Tesla (c. 1890): Napoleon Sarony, public domain, via Wikimedia Commons.

Patent drawings: USPTO. Court documents: U.S. federal courts. Music and sound design: original.

Not affiliated with Tesla, Inc., Nikola Corporation or Hyroad Energy.

Strategy education, not legal advice. ipCapital Group.

The Ouija board has a U.S. patent. So does a coffin with a bell on it, and a torpedo you bury with the body. A tour of the patent office's haunted wing: real, granted patents from 1868, 1878 and 1891, and what they say about what people feared.

Then the part nobody tells you: an invention has to be "useful" to get a patent, but usually nobody tests it. And a patent expires, while a trademark can live on: the Ouija patent ran out over a century ago, and Hasbro still owns the name. Today's fear has patents too, like Bank of America's "defensive deepfake."

A Paper Trail story from Invent Anything.

Patents shown

- US 446,054, "Toy or Game" (the Ouija board), Elijah J. Bond, Feb. 10, 1891: https://patents.google.com/patent/US446054A/en

- US 81,437, "Improved Burial-Case," Franz Vester, Aug. 25, 1868: https://patents.google.com/patent/US81437A/en

- US 208,672, "Improvement in Coffin-Torpedoes," Philip K. Clover, Oct. 8, 1878: https://patents.google.com/patent/US208672A/en

- US 12,401,744, "Defensive deepfake for detecting live deepfaked audio and video," Bank of America, Aug. 26, 2025: https://patents.google.com/patent/US12401744B2/en

New Paper Trail stories every Thursday. This Week in IP every Tuesday.

Sources

- Smithsonian Magazine, "The Strange and Mysterious History of the Ouija Board" (the patent-office legend, as told by historian Robert Murch, and the 1890s ad): https://www.smithsonianmag.com/history/the-strange-and-mysterious-history-of-the-ouija-board-5860627/

- WOSU Public Media on Ohio's coffin torpedoes (Gabe Rosenberg, May 17, 2017): https://www.wosu.org/news/2017-05-17/ohios-ghoulish-gambit-against-grave-robbing-coffin-torpedoes

- Benjamin Harrison Presidential Site, "The Grave Robbing of Benjamin Harrison's Father": https://bhpsite.org/the-grave-robbing-of-benjamin-harrisons-father/

- USPTO, Manual of Patent Examining Procedure § 2107 (utility): https://www.uspto.gov/web/offices/pac/mpep/s2107.html

- Juicy Whip v. Orange Bang, 185 F.3d 1364 (Fed. Cir. 1999): https://law.resource.org/pub/us/case/reporter/F3/185/185.F3d.1364.98-1379.html

- OUIJA, U.S. Trademark Reg. No. 519,636 (USPTO TSDR): https://tsdr.uspto.gov/#caseNumber=71546217&caseType=SERIAL_NO&searchType=statusSearch

- As of Oct. 2, 2026, a USPTO Open Data Portal title search finds more than 140 U.S. patent applications with "deepfake" in the title.

Credits

Patent drawings: USPTO. Music and sound design: original.

Not affiliated with Hasbro or Bank of America. Trademarks belong to their owners.

Strategy education, not legal advice. ipCapital Group.

What if you could try your invention a hundred times before you built it once?

Google DeepMind's Genie 4 generates interactive worlds you can steer in real time. Seth Cronin works through where, when, how and why AI simulation becomes a tool for invention: the downhill search at every level of the AI stack, agents that hill-climb on any metric you can measure (Karpathy's autoresearch), AI checking a 15,000-page proof (Axiom), lab loops in biology, and now physical worlds. Plus the catch: a simulator that's beautiful and wrong just lets you fail faster.

Read the main story as an essay: https://www.ipcg.com/blog/ai-simulation-tool-for-invention

Also this week: OpenAI and Anthropic claim to refute the Exponential Time Hypothesis; Anduril and Qualcomm's BlueChip laptop; Mitsubishi Electric's humanoid robots and machine tools at JIMTOF; MIT's LUCID implant. Patent of the Week: LG Energy Solution's battery-line simulation patent, US 12,775,268.

What's an idea you've been sitting on because testing it seemed too expensive? If you could try a thousand versions this month, what would you measure? Tell me in the comments.

This Week in IP, from Invent Anything. Hosted by Seth Cronin, ipCapital Group.

Chapters
0:00 Try your invention 100 times before you build it
0:43 Did AI just refute the Exponential Time Hypothesis?
1:51 BlueChip: a laptop built around where its parts come from
3:06 Mitsubishi's robots and the invention in the handoffs
3:50 MIT's LUCID: two technologies, both problems
4:31 Main story: what can you prove in a simulation?
5:26 Every level of the AI stack is a downhill search
6:38 Agents that hill-climb: Karpathy's autoresearch
8:57 AI checks the "enormous theorem"
10:51 Biology: the golden age of the idea guy
12:34 Genie 4 brings the loop into physical worlds
14:27 Robots, and the catch: beautiful and wrong
15:51 The invention loop: simulate, prototype, measure, fix the twin
16:54 The golden age of the idea guy
17:52 Patent of the Week: LG's battery-line simulation
18:41 What would you measure?

Sources and links are in the YouTube description. The ETH refutations and the Axiom, OpenAI, Anthropic, Figure and DeepMind results are the companies' own claims unless stated; outside review of the ETH claims is ongoing. Binding is not an antivenom, and the BigHat results are preclinical. A granted patent does not establish that a product uses the invention. Genie 4 clips are Google DeepMind's own demo footage of generated worlds. Collage objects, two motion clips and the thumbnail art are AI-generated illustrations, labeled on screen; the drawings are ours. Strategy education from ipCapital Group, an IP strategy consulting firm (not a law firm). Nothing here is legal advice.

MUSIC
"The Crate Desk", original for Invent Anything. Piano: Versilian Studios via tonejs-instruments (Nicholaus P. Brosowsky), CC BY 3.0; arranged and processed. No endorsement implied.

When AI makes copying cheap, where does the value go?

This week, the studio behind Moonlighter pulled back its public playtest over cloning fears, and another studio cancelled its free Steam demo, saying a copy could be produced quickly with AI. Xbox's own memo on Xbox Labs says rapid ports and coding agents challenge its existing business models. Seth Cronin works through where the value moves when building and copying get cheap: down to chips and power, to trust and security, to community, and back into hiding, the very tension the patent system was built for.

Also this week: about 300 authors file individual suits against Meta, and a judge orders Meta to name everyone who approved downloading the pirated books; Amazon and Google both bring PyTorch to their own AI chips; Google loses twice in patent law on the same Monday; the Supreme Court can't hear BMG v. Vetter for lack of a quorum; one hacker uses AI agents to break into sites at more than 100 companies; and AI helps find critical bugs in PHP. Patent of the Week: Dell's modular AI appliance.

If AI could clone what you're building from a two-minute demo, what would you still be selling? Tell us in the comments.

More on AI-built clones: Invent Anything Talks with Mike D'Andrea: https://youtu.be/EeYTlSRpxW0

Chapters

0:00 Why game studios are pulling their demos

0:45 Meta has to name everyone who approved the pirated books

1:38 Amazon and Google go after Nvidia's real advantage

2:20 Google's rough week in patent law

2:51 The Supreme Court couldn't hear a case at all

3:22 Main story: Xbox says its business model is being challenged

4:20 AI moves the value to different parts of the stack

5:04 People are using AI to declutter their smart TVs

5:39 One hacker, AI agents, 100 companies

6:21 AI found critical bugs in PHP

7:18 So where does the value go?

7:42 The patent bargain vs. the cancelled demo

8:08 Invent at the right layer

8:34 More in our Adobe conversation with Mike D'Andrea

8:58 Patent of the Week: Dell's modular AI appliance

9:55 Future Seth update: "Black October"

12:18 What would you still be selling?

Full sources are in the YouTube description: https://youtu.be/29-8AwkejKU

The Meta claims are allegations in pending lawsuits. The $50-per-hack figure is the NorForge researcher's estimate; Reuters could not determine whether data was taken. TorchNeuron and TorchTPU compatibility claims are the companies' own. Amazon's reactors are a proposal. A granted patent does not establish that a product uses the invention, and Dell has not announced a device like it. The drawings are our own sketches unless labeled as patent drawings. Strategy education from ipCapital Group, an IP strategy consulting firm (not a law firm). Nothing here is legal advice.

Music: "The Crate Desk", original for Invent Anything. Piano: Versilian Studios via tonejs-instruments (Nicholaus P. Brosowsky), CC BY 3.0; arranged and processed. No endorsement implied.

Someone used AI coding agents to rebuild Photoshop from scratch, and copyright may not be able to stop it. So what can a software company still own?

Seth Cronin talks with his ipCapital Group colleague Mike D'Andrea about the AI-built Adobe clones (ArtCraft's PhotoCraft and its sibling apps). Copyright protects the code you wrote, the way it protects the words and recording of a song. A patent protects how a feature works, whoever writes the code. They walk through the Adobe patents behind Content-Aware Fill and Generative Fill (and read the actual claim), how a clean-room rebuild works, and who a patent really stops: individual users, enterprises or competitors.

Mike's practical takeaway: run the test on your own product. Have AI rebuild it feature for feature, see what it couldn't do, and sort what's left into two piles: what's visible from outside, which a patent can cover, and what isn't, which can stay a trade secret.

If someone rebuilt your product next week, what would still be yours? Tell us in the comments.

Mike and Seth aren't lawyers. This is a conversation about invention and business, not legal advice, and not a view on whether any project infringes any patent.

Mike's article, "Someone Cloned Photoshop in a Week. Copyright Can't Stop It. Patents Can." (Oct. 11, 2026):

https://www.linkedin.com/pulse/someone-cloned-photoshop-week-copyright-cant-stop-patents-d-andrea-jd3ne/

Watch the full video with the on-screen sources: https://youtu.be/EeYTlSRpxW0

Chapters

0:00 Someone rebuilt Photoshop

1:00 What copyright can't stop

2:13 Mike D'Andrea on what caught his attention

2:40 How the story spread

3:23 Copyright protects the song

4:10 The idea vs. the recording

5:39 Adobe's fill patents

6:30 Turn the gondola into a bird

7:17 The clean-room rebuild

8:49 Reading the actual claim

9:29 From ones and zeros back to ideas

11:46 Does AI end the first-mover advantage?

13:02 Run the test yourself

14:13 Who a patent actually stops

15:27 Competitors, not customers

16:09 The FilmCraft review

17:26 Why Napster lost

17:53 When a patent comes into play

19:09 Reverse engineer your own software

19:48 Write the spec for every way it could be built

Corrections

"Future Seth" pops in a few times with fixes we found after recording: the Word, Excel and PowerPoint clones came out Oct. 7, a few days before Mike's article; PhotoCraft hasn't built Generative Fill yet; REA (Reverse Engineer Anything) is a separate project, and ArtCraft's developer says he doesn't use it or decompiled binaries; the Premiere-style clone is FilmCraft; and DaVinci Resolve's free version isn't open source (it's free proprietary software, available since 2011).

Key sources

PhotoCraft (ArtCraft): https://github.com/storytold/photocraft

US 10,706,509 B2 (Adobe): https://patents.google.com/patent/US10706509B2/en

US 12,462,456 B2 (Adobe): https://patents.google.com/patent/US12462456B2/en

Invent Anything Talks, from Invent Anything

Hosted by Seth Cronin | ipCapital Group, an IP strategy consulting firm

Recorded October 15, 2026.

When an AI hands you an answer you can’t check yourself, what earns your trust?

OpenAI posted 722 AI-written math papers. A day later, three were withdrawn after a sign error, and as of the latest update about 42% of the 719 that remain have their main result checked in the Lean proof checker. Then Terence Tao asked a sharper question in his Caltech lecture “Math 2.0”: if an AI found a cancer cure nobody understands, and it passed the trial, would you take it? Seth Cronin works through reward hacking, Goodhart’s law, why people game metrics too, and what he learned building Frix AI before going back to consulting. Proof got cheap. Trust didn’t.

Also this week: an open-source project rebuilding Adobe-style apps with AI coding agents; an OpenAI employee arrested over an alleged recording device in an executive conference room; Character.AI files for Chapter 11, with Disney as the opening bidder for a non-exclusive license to some of its IP; Microsoft Research’s StudentSim, simulated students for testing AI tutors; and Patent of the Week, Intuit’s patent on keeping the math out of the LLM.

Think of the last time an AI gave you an answer at work that looked right. What would you have needed to see before you bet your trust on it? Tell us in the comments.

Chapters

0:00 Would you take an AI’s cancer cure?

0:51 Somebody rebuilt Adobe

1:26 An alleged recording device at OpenAI

2:29 Character.AI’s bankruptcy and Disney’s bid

3:35 StudentSim: fake students for testing AI tutors

4:34 Main story: OpenAI’s 722 math papers

5:33 Terence Tao, reward hacking and Goodhart’s law

6:11 Would you want a human who understands it?

7:11 The fair argument on the other side

7:40 People reward-hack too

8:30 Why I went back to consulting

9:35 What an expert still has to do

10:10 Proof got cheap. Trust didn’t.

10:35 Patent of the Week: Intuit keeps the math out of the LLM

12:11 What would you need to see?

Sources and notes are in the YouTube description. Strategy education from ipCapital Group, an IP strategy consulting firm (not a law firm). Nothing here is legal advice.

Music: “The Crate Desk” — original composition for Invent Anything. Piano recordings by Versilian Studios, packaged by Nicholaus P. Brosowsky (tonejs-instruments), CC BY 3.0; pitch-shifted, arranged and processed. No endorsement implied.

https://github.com/nbrosowsky/tonejs-instruments

https://creativecommons.org/licenses/by/3.0/

Additional sounds: Karoryfer Black and Blue Basses and VCSL (CC0); GeneralUser GS by S. Christian Collins. Jev LevelDrive saturation used under its MIT license.

https://schristiancollins.com/generaluser.php

Same AI, different character. How much of an AI product is its personality? Seth Cronin starts with a new VR study in which people paid a female-presenting AI assistant 10.25% less than an otherwise identical male-presenting one, even though many said gender didn't matter. Then the model's own disposition: an agent so cautious it wouldn't retry a banal video request, which Seth calls anti-alignment. Borrowing from Dungeons & Dragons and Edward de Bono's Six Thinking Hats, he makes the case for agent "classes" (fighter, wizard, rogue, cleric) and shows how to use deliberately different roles as an invention tool. Plus the IP angle: a Tokyo court recognized a publicity right in a person's voice.

Also: blue light against Campylobacter on raw chicken; NC State's laser-made plasma antenna (and the earlier work behind "first"); Reka's Rho-1 research preview; and Patent of the Week, Samsung's robot cleaner that looks for where the mess comes from.

Chapters

0:00 The scene my agent wouldn't retry

0:51 Blue light against food-poisoning bacteria

1:53 An antenna made with a laser

2:58 Reka's Rho-1: video and robot actions in one model

4:06 Main story: how much of an AI product is its personality?

5:24 Who is the model when it answers you?

5:55 When safety turns into anti-alignment

7:02 Classes for agents: fighter, wizard, rogue, cleric

8:05 Six Thinking Hats

8:55 An invention tool you can use today

9:38 Personality as an IP question

10:39 Patent of the Week: Samsung's pollution-source robot cleaner

11:50 Which class would you pick?

Seth's article, "What D&D Can Teach Personal Agent Labs": https://www.linkedin.com/pulse/what-dd-can-teach-personal-agent-labs-seth-cronin-d8sje/

Sources, transcript and captions: https://youtu.be/OJx3MW0HImY

The blue-light results are from lab cultures, not tests on meat. The plasma antenna has transmitted over a short range; reception has not been shown. Rho-1 is a research preview, and its robot demo ran in simulation. The VR study is one experiment in which participants split a payment; it is not a salary survey. A patent does not establish that a product is shipping. Illustrative live-action scenes were generated with AI.

Music: "The Crate Desk," original for Invent Anything. Piano recordings by Versilian Studios (tonejs-instruments), CC BY 3.0. Additional sounds: Karoryfer Black and Blue Basses and VCSL (CC0); GeneralUser GS by S. Christian Collins.

If you and your competitors can buy roughly the same AI, where does the difference come from? Seth Cronin looks at three personal agents launched in about seven weeks (OpenAI's Dots, Meta's Muse and Grok Bot) and how to keep a divergent outlook when everyone has the same model: bring AI a problem you've actually observed, widen the approaches before choosing one, run small tests, and ask what you have specifically figured out. Plus the IP side: a patent covers its claims, not a product category.

Also: Thomson Reuters v. ROSS, the first U.S. appeals ruling on fair use in AI training; Microsoft's Quine and Anthropic's Model Hardware Standard; Northwestern's protective-clothing particles; and Patent of the Week, Toyota's smart vehicle-to-home patent.

Chapters

0:00 Same AI. How do you stand out?

0:45 Thomson Reuters v. ROSS: the first appeals ruling on AI training

1:37 Microsoft’s Quine and Anthropic’s Model Hardware Standard

2:50 Clothing that could break down toxic compounds

3:23 Dots, Muse and Grok Bot: the personal agents converge

4:12 Main story: same tools, different questions

5:03 When everyone uses the same algorithm

5:47 A worked example: the irrigation controller

7:41 The IP side: claims, backlog and know-how

8:52 Patent of the Week: Toyota’s smart V2H

10:12 Your examples

Sources, transcript and captions: https://youtu.be/i6UgQwGRzak

The MIT paper is a philosophical analysis of hiring and similar decisions; applying it to invention is Seth's interpretation. The irrigation controller is a hypothetical example. Illustrative live-action scenes were generated with AI.

Music: "The Crate Desk," original for Invent Anything. Piano recordings by Versilian Studios (tonejs-instruments), CC BY 3.0. Additional sounds: Karoryfer Black and Blue Basses and VCSL (CC0); GeneralUser GS by S. Christian Collins.

AI can give you a hundred invention ideas. Which one solves a problem somebody actually cares about—and what would you build to find out?

Seth Cronin looks at where AI belongs in invention: new Atlassian research on brainstorming, a Procore patent for smart safety glasses, and a simple travel-mug problem that turns a list of features into something you can test.

Also: MIT’s dissolving batteries, Microchip’s completed Hailo acquisition, a robotic wine server, eleven Purdue patents, and Subaru’s exterior-airbag design for protecting people outside the vehicle.

Where has AI helped you get from a plausible idea to a useful test? Share the specific example.

This Week in IP, from Invent Anything. Hosted by Seth Cronin | ipCapital Group. Recorded September 23, 2026; sources dated September 21–22.

CHAPTERS
0:00 100 ideas. Now what?
0:36 MIT’s dissolving batteries
1:19 Microchip buys Hailo
2:04 A robotic wine server
2:54 Purdue’s eleven patents
3:42 AI and the first spark of invention
5:42 Procore’s smart safety glasses patent
6:24 A better travel mug: problem, constraints, test
8:22 What is different? The prior-art question
9:37 Subaru’s exterior-airbag patent

Sources, links and the full video: https://youtu.be/vw_ruDzFTD0

The Atlassian result concerns brainstorming tasks, not patentability. The MIT work includes animal testing. Patent disclosures do not establish a product launch, performance or safety. Generated illustrative scenes are not footage of the companies’ products.

MUSIC: “The Crate Desk” — original composition for Invent Anything. Piano recordings by Versilian Studios, packaged by Nicholaus P. Brosowsky (tonejs-instruments), CC BY 3.0; pitch-shifted, arranged and processed. No endorsement implied. Additional sounds: Karoryfer Black and Blue Basses and VCSL (CC0); GeneralUser GS by S. Christian Collins. Jev LevelDrive saturation used under its MIT license.