App. No. 19/629,591

compare the input signature to the accessed agent signatures to select one or more AI agents from the plurality of AI agents ... and cause transmission of the input to at least one AI agent of the selected AI agent set

The office action dismisses the recited AI agents, input signature, and hash functions as merely 'generally linking the use of the judicial exception to a particular ... field of use' (MPEP 2106.05(h)) and treats obtaining/transmitting the input as insignificant extra-solution activity. Counsel can weigh whether the claims, taken as an ordered combination, instead recite an improvement to computer-implemented AI-agent routing — converting an input into a signature/fingerprint via hash functions and matching it against precomputed agent signatures to select which agent(s) receive the input. Under MPEP § 2106.05(a) and Enfish/McRO, an improvement to the way a computer performs a task integrates an exception into a practical application at Step 2A Prong 2, ending the analysis. The office action does not appear to have evaluated the ordered combination for a technological improvement, having instead dissected the elements individually — a point counsel may press against the Prong 2 conclusion.

• —Office action (claim 6, Prong 2): 'The recitation of AI agents, input signature, and hash function amounts to generally linking the use of the judicial exception to a particular environment of field of use (See MPEP 2106.05(h)).'
• —Office action (claim 6, Prong 2): 'obtain an input ... cause transmission of the input ... amounts to extra-solution activity of gathering data'
• —Pending claim 18: 'the plurality of AI agents are organized in a hierarchical architecture comprising a general-purpose agent at a first level ... and at least one domain-specific agent at a second level'

⚠ Risk The examiner may respond that the specification does not describe a specific technical improvement over prior routing methods (i.e., the alleged improvement is to the abstract idea itself, not to computer technology), and that using hashing for matching is a generic application. Counsel should confirm the as-filed specification actually supports a technological-improvement narrative before advancing it, and be mindful that characterizing the improvement narrowly may create estoppel.

Strategy check: re-ranked from #4 — The claim-7-specific practical-application theory (rank 4) is dispositive if won, while the generic Berkheimer argument at #1 only forces documentation; the record itself flags the distributed-ledger/attestation element as the 'least easily cured,' so its Prong 2 treatment deserves the lead for this claim.

generate a cryptographic attestation of a routing decision that identifies the at least one AI agent of the selected AI agent set; and maintain the cryptographic attestation in a distributed ledger

The office action treats claim 7's generation of a cryptographic attestation of the routing decision as 'apply it' plus 'mere data output,' and treats maintaining that attestation in a distributed ledger as extra-solution data-gathering. Counsel can weigh whether a cryptographic attestation tied to a distributed ledger is properly analyzed as insignificant extra-solution activity or generic 'apply it,' or whether it is an additional element that constrains how the routing decision is recorded and verified. Because the office action's Step 2B treatment of this element again rests on an unsupported conventionality assertion (see rank 1), the distributed-ledger element may warrant a distinct Berkheimer challenge and a Prong 2 practical-application inquiry rather than the generic labels applied.

• —Office action (claim 7, Prong 2): 'generate a cryptographic attestation of a routing decision ... these elements ... amounts to no more than adding the words "apply it" to the judicial exception ... mere data output (see MPEP 2106.05(g))'
• —Office action (claim 7): 'The claim further recites a distributed ledger which is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).'

Risk The examiner may respond that distributed ledgers and cryptographic attestations are well-known generic tools recited without technical detail, and that recording a decision is classic post-solution output. This argument reaches only claim 7 and does not disturb the independent-claim rejection.

Strategy check: re-ranked from #5 — The claim-14 practical-application/computational-transformation point (rank 6), tied to parent-claim integration, is a dispositive route to withdrawal, whereas the #1 Berkheimer gap is curable and the examiner's truncated 'apply it' basis is itself contestable on completeness.

determining the feature set from the input comprises applying an embedding model to a representation of the input

The office action treats claim 14's application of an embedding model as recited 'at a high-level of generality with no detail of the applying an embedding model process' and amounting to no more than 'apply it,' and the analysis then cuts off mid-sentence. Counsel can weigh whether applying an embedding model to a representation of the input is properly dismissed as generic 'apply it,' or whether — like the hashing operations at rank 2 — it recites a computational transformation that is not practically performed in the human mind and that may contribute to a practical application under Step 2A Prong 2. Because the office action's stated basis is truncated, the completeness of the Prong 2/Step 2B rationale for claim 14 is itself contestable.

• —Office action (claim 14, Prong 2): 'determining the feature set from the input comprises applying an embedding model to a representation of the input ... amounts to no more than adding the words "apply it" ... or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application ... (See'
• —Rejection map (OA1): analysis of claim 14 is where 'The office action text cuts off'

⚠ Risk The examiner may respond that claim 14 recites the embedding model only functionally, without an algorithm, so it remains a generic tool applied to the abstract idea. This argument reaches only claim 14. Characterizing embedding-model application as inherently non-mental may create estoppel and should be weighed against the specification's actual disclosure.

cause transmission of the input to at least one AI agent of the selected AI agent set / obtain an input for generation of an output using one or more artificial intelligence (AI) agents; and for claim 7 'maintain the cryptographic attestation in a distributed ledger'

At Step 2B the office action concludes the additional elements (obtaining an input, causing transmission, storing/retrieving signatures, maintaining a cryptographic attestation in a distributed ledger) are 'well-understood, routine, and conventional,' resting solely on MPEP 2106.05(d)(II) categories and courts' generic labels. Under Berkheimer v. HP and the USPCourt guidance now codified at MPEP § 2106.05(d)(I), a conclusion that an additional element is well-understood, routine, and conventional must be supported by one of four evidentiary showings — an express applicant admission, a citation to a court decision noting the element's conventionality, a citation to a publication, or an official notice. The office action supplies none of these for the specific claimed elements; it cites only generic MPEP category descriptions, which is a factual gap counsel can press for each claim. Counsel may weigh whether a distributed-ledger cryptographic attestation (claim 7) in particular was ever shown to be conventional on this record.

• —Office action Step 2B (claim 6): 'The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II)...)'
• —Office action Step 2B (claim 7): additional element 'maintain the cryptographic attestation in a distributed ledger' treated as extra-solution activity and 'well-understood, routine, and conventional'
• —Rejection map (OA1): 'the examiner finds the extra-solution activity elements are well-understood, routine, and conventional (citing MPEP 2106.05(d)(II)...)'

⚠ Risk The examiner may respond by adding one of the four Berkheimer support categories (e.g., citing a case or taking official notice) in the next action, curing the gap; the argument is procedural and may only delay rather than defeat the Step 2B analysis. No prosecution-history-estoppel concern for §101, but avoid characterizing the routing/hashing steps as conventional in the response.

determining an input signature by applying one or more hash functions to the feature set (claim 6) / determine an input fingerprint by applying one or more sets of hash functions to the feature set (claim 13); wherein the one or more hash functions comprise locality-sensitive hash functions (claim 12); perform a bitwise operation between the input signature and an agent signature to determine a distance value (claim 9)

The office action characterizes the signature/fingerprint step as a mental process, offering 'a human can determine a distinct identifier for each and every AI agent' as its example. Counsel can weigh whether that example actually captures the recited step of applying one or more hash functions to a feature set — a computational transformation — rather than a bare human labeling exercise. Under MPEP § 2106.04(a)(2)(III), a limitation is a mental process only if it can PRACTICALLY be performed in the human mind; claim 12 narrows the hash functions to locality-sensitive hash functions and claim 9 recites a bitwise operation producing a distance value, operations that OA4 notes presuppose specialized computational primitives. The Step 2A Prong 1 characterization may therefore be contestable as an oversimplification of what the claims actually require.

• —Office action (claim 6, Prong 1): 'determining an input signature by applying one or more hash functions to the feature set; ○ For example, a human can determine a distinct identifier for each and every AI agent.'
• —Pending claim 9: 'perform a bitwise operation between the input signature and an agent signature to determine a distance value'
• —Pending claim 12: 'wherein the one or more hash functions comprise locality-sensitive hash functions'
• —OA4 analysis note: certain steps may 'presuppose specialized computational operations' relevant to whether they are 'truly performable in the human mind'

⚠ Risk The examiner may respond that the broadest reasonable interpretation of 'hash function' and 'signature' still reads on a simple human-assignable identifier, and that claims 9 and 12 are dependent (leaving independent claims 6/13 characterized as mental processes). Framing the hash operations as inherently computational could narrow claim scope in the file wrapper — counsel should weigh prosecution-history-estoppel exposure before committing to that characterization.

each dependent limitation of claims 15-20 (e.g., claim 16 'modifying one or more parameters of the one or more hash functions based on the feedback data'; claim 18 hierarchical general-purpose/domain-specific agent architecture; claim 20 'generating a routing path ... and storing the routing path in a persistent data structure')

The office action summary lists claims 6-20 as rejected under §101, but the detailed analysis, as supplied, cuts off during claim 14 and contains no individualized Step 2A/2B treatment of claims 15-20. Under MPEP § 2106.07 and § 707.07, each claim's additional elements must be evaluated, and a rejection must give the applicant fair notice of the basis for rejecting each dependent claim. Counsel can weigh raising the absence of any articulated analysis for claims 15-20 — several of which recite additional elements (feedback-based parameter modification in claim 16, a hierarchical multi-agent architecture in claim 18, routing-path storage in claim 20) that were never addressed for practical-application or inventive-concept significance. This is a notice/completeness point for counsel to press alongside the substantive arguments.

• —Rejection map (OA1): 'The office action text cuts off during the analysis of claim 14 and does not contain individualized analyses for claims 15-20, though the summary page lists claims 6-20 as rejected.'
• —Office action summary: 'Claim(s) 6-20 is/are rejected.'
• —Pending claim 16: 'modifying one or more parameters of the one or more hash functions based on the feedback data'

Risk The examiner may cure this by supplying full individualized analyses for claims 15-20 in the next action, and may contend the cutoff is an artifact of the copy provided rather than the mailed action. Counsel should confirm the complete mailed office action does in fact omit claims 15-20 before raising the point, to avoid a mischaracterization of the record.