How to Build an IP Portfolio With AI Research Agents

Building an IP portfolio with AI research agents comes down to a repeatable loop: verify a premise against primary literature, screen candidates against every real-world constraint, falsify the survivors, map the patent white space, and file provisional patents on the novel claims before you publish anything. Run that loop across many domains and the portfolio compounds.

This isn't theory. The AIMS methodology has run exactly this loop to 78 USPTO-filed provisional patents, roughly 1,275 claims, across 40 scientific domains — with the methodology itself patented (Patent 14) and documented in its own paper. Here's how the loop works, stage by stage, and where agents versus humans carry the weight.

Why is IP the right output for AI research?

Because a defensible claim is the smallest unit of research value you can own. Papers build credibility; patents build assets. The AIMS business model makes the sequence explicit — Validate → Protect → License → Spin Out. Use AI to discover at scale, file on everything novel, then monetize through licensing agreements with manufacturers and R&D teams or by spinning ventures out around the biggest breakthroughs. Without the "Protect" step, everything upstream is a donation to whoever commercializes it first.

Step 1: Verify the premise before spending anything

Every project starts with an assumption — this sorbent chemistry could capture CO2 at ambient temperature, this dopant could extend electrode life. Agents check that assumption against primary literature first. Most premises die here, which is the point: a premise killed in hours costs nothing, and every one that survives is load-bearing for real filings later. The mechanics are covered in how AI agents accelerate literature review.

Step 2: Map the landscape — and the white space

An exhaustive literature review does double duty in an IP workflow. The same sweep that tells you what's known tells you what's claimed. Agents map published research and existing patent claims together, so by the end of this stage you know precisely where the boundaries of the prior art sit. Novelty isn't a vibe — it's a gap in a map you've actually drawn.

Step 3: Screen by constraints, not curiosity

Compile the testable physical, chemical, and economic constraints the material must satisfy, then let agents search the bounded space and score every candidate against all constraints simultaneously — including manufacturability, cost, and destructive byproducts. This matters for IP specifically: claims on a material that can't be manufactured at price are paper trophies. Constraint-satisfaction screening produces candidates that are both novel and commercially plausible, which is what a licensee actually pays for.

Step 4: Falsify before you file

Stage 5 of the AIMS pipeline is a structured adversarial attack on the top candidates: hunt for the disqualifying paper, audit every citation, stress-test every number. This is where the IP-gap analysis is finalized too — confirming that surviving claims genuinely clear the prior art. Filing on unfalsified AI output is how you end up paying attorney fees to protect a hallucination. The full protocol is in how to falsify AI-generated scientific claims.

Step 5: File provisionals — then publish

The ordering is non-negotiable: protect before you disclose. A U.S. provisional patent application secures a twelve-month priority date without examination, at a small fraction of the cost of full utility prosecution. That cost profile is what makes portfolio-scale filing viable for a small operation — you can protect dozens of novel claim sets quickly, then let the market (licensing interest, partner validation) tell you which ones deserve conversion to full applications within the year.

Only after filing does the work go public. AIMS publishes full papers to peer-review standards, with published work carrying permanent DOIs on Zenodo. Open publication after filing is a feature, not a risk: it establishes the work, invites expert falsification, and markets the IP to potential licensees simultaneously.

Where do humans stay essential?

Three places, and they're structural:

• Inventorship. Under current U.S. law, an AI system cannot be named as an inventor — a human must make the significant inventive contributions. In practice that means humans own the constraint definitions, the candidate selections, and the claim strategy. (More in can AI actually generate patentable inventions?)
• Kill decisions. Agents surface evidence; a human architect decides what dies and what proceeds. Sunk-cost immunity is a human discipline.
• Portfolio strategy. Which domains to enter, which claims to convert, which to license versus spin out — these are capital-allocation calls, not research tasks.

What makes the loop compound

The methodology is domain-agnostic. Once the engine works, each new domain reuses it: the same ten stages that produced PFAS-removal membranes also produced iron-air battery electrodes, non-toxic fire retardants, and topological quantum computing architectures. Breadth is the moat — 40 domains means 40 uncorrelated bets riding one process. And because agents do the heavy reading and screening, the marginal cost of entering domain 41 keeps falling. That cost curve is the subject of what materials R&D actually costs. The agent layer that powers all of it is the FAST framework.

FAQ

Why file provisional patents instead of full utility patents?

A U.S. provisional application secures a priority date for twelve months at a fraction of the cost of a full utility filing, and it isn't examined. For a high-throughput research operation, provisionals let you protect many novel claims quickly, then decide within the year which ones justify the expense of full prosecution — after licensing interest or experimental validation clarifies their value.

Should I publish research before or after filing?

File first, always. Public disclosure before filing can destroy patentability in most jurisdictions outside the U.S. grace period. The AIMS sequence is deliberate: patent filing is stage 9 of the pipeline and publication follows it — protect the IP, then publish openly.

How many patents has the AIMS methodology produced?

78 USPTO-filed provisional patents carrying roughly 1,275 claims across 40 scientific domains, alongside 41 research papers. The methodology itself is also patented (Patent 14) and documented in its own paper.

Can AI agents do the patent drafting?

Agents can draft claims, map prior art, and structure specifications, but a human inventor must make the significant inventive contributions and a human should own final claim decisions. Under current U.S. law an AI system cannot be named as an inventor, so the human role isn't a formality — it's a legal requirement.