Artificial intelligence changes how companies discover technical possibilities, develop solutions, create visual worlds, search existing knowledge, protect assets and bring intellectual property into commercial use. These developments affect the connections between innovation, legal protection and business decisions. As individual activities become faster, companies need to reconsider how information, responsibilities and decisions move through the entire IP value chain.

This is the central perspective of my presentation at ISAR INTELLECTUALS 2026 in Munich. The conference on innovation, intellectual property and AI takes place on 22 October at the Münchner Künstlerhaus. My keynote contributes to the panel on AI in IP practice, with Uwe Herrmann of the German Patent and Trade Mark Office and Jan Witt of Anaqua participating in the discussion, moderated by Dr Alexander Harguth.

The examples range from materials science and drug development to packaging design, patent search, brand protection and interactive character licensing. Together, they raise a management question: which economic effects should a company seek to achieve when it scales IP work with AI?

Understanding the complete IP value chain

The presentation follows seven connected activities: aligning, creating, checking, protecting, managing, exploiting and enforcing. Aligning means defining where IP should support the business. Creating includes inventions, designs, content and know-how. Checking covers the examination of technical alternatives, existing knowledge and relevant rights. Protecting concerns the choice and implementation of appropriate instruments. Managing connects the resulting assets with changing business needs. Exploiting brings them into products, services, partnerships and licensing. Enforcing helps preserve the intended commercial position.

These activities apply across technical solutions, brands, designs, content and characters. Know-how and data connect many of them. A new product may combine a technical invention, software, a distinctive appearance, brand elements and licensed content. Decisions made in one area can change the options available in another.

Faster creation therefore increases the importance of effective handovers. A team capable of generating hundreds of concepts still needs a way to select promising options, investigate them and decide which deserve investment. A monitoring system that identifies more suspicious offers needs a process for prioritising cases and authorising responses. The relevant measure of progress is the improvement in the complete business process.

This perspective connects directly with the IPBA platform article Why the 360° IP Strategy Matters More Than Ever. Its starting point is the customer benefit and competitive position that a business wants to secure. AI expands the available means of pursuing that objective.

Materials and proteins reveal the scale of the new search space

Google DeepMind’s GNoME research illustrates how AI can expand the range of technical possibilities available for investigation. In 2023, the project reported approximately 2.2 million predicted crystal structures, including around 380,000 candidates predicted to be particularly stable. Crystal structures describe how atoms are arranged in a solid. That arrangement helps determine properties relevant to applications such as batteries, electronics and energy technologies.

The distinction between a prediction and a usable material is essential. A stable structure identified computationally still needs to be examined for synthesis, practical performance and suitability for a particular application. Manufacturing cost, processing conditions and reproducibility can determine whether an interesting candidate has industrial value. Simulation and automated laboratories can help investigate these questions, creating feedback between prediction and experiment.

For IP management, this expands the field in which companies must choose. A large pool of candidates creates opportunities to identify useful compositions, production processes and applications. It also makes prioritisation more demanding. The business needs a view of which performance improvement matters, which technical route is feasible and where a defensible position could support future revenues.

AlphaFold provides a related example in biology. Proteins consist of amino-acid chains whose three-dimensional structures help scientists investigate how they function. The presentation uses a reference figure of more than 241 million predicted protein structures. The live AlphaFold Protein Structure Database now reports more than 260 million predictions, illustrating the continuing expansion of this resource.

A structural prediction can help researchers formulate hypotheses, examine potential interactions and select further experiments. It does not itself establish the biological effectiveness of a treatment. The economic opportunity depends on connecting accessible structural knowledge with experimental evidence and a useful application.

My management interpretation is that wider access to predictions can increase the strategic importance of application knowledge, proprietary experimental results and the ability to validate and implement a solution. The IPBA article Synthetic Inventing for Digital Business Models provides a method for directing a larger solution space towards commercially relevant objectives. It starts with the competitive effect a company wants to achieve and works towards technical solutions capable of supporting that position.

AlphaEvolve connects generation with measurable technical improvement

Google DeepMind’s AlphaEvolve, presented in May 2025, combines AI-generated proposals with automated evaluation. Candidate programs are tested, assessed and improved through repeated iterations. This creates a development loop in which the ability to evaluate a proposal becomes a central part of the invention process.

One application concerns the scheduling of work across Google’s data centres. According to Google, an AlphaEvolve-discovered scheduling heuristic continuously recovered an average of 0.7 percent of its worldwide compute resources. The percentage refers to computing capacity made available through more efficient scheduling. It is not a reported increase in corporate profit or a general productivity benchmark for AI.

At the scale of a large computing infrastructure, a small percentage can still have substantial operational significance. Google also reported that a circuit-design proposal was incorporated into a forthcoming Tensor Processing Unit. This is a separate example of AI contributing to a technical implementation.

The lesson for IP management concerns the relationship between proposal, evaluation and implementation. Companies need to identify the technical contribution, preserve the evidence supporting it and decide how it fits their protection and commercialisation strategy. Access to useful test environments, evaluation criteria and implementation expertise can become important sources of advantage. These examples support examining the complete development loop when assessing what should be protected and controlled.

Rentosertib shows why validation remains central

Rentosertib, developed by Insilico Medicine, illustrates AI involvement in both the selection of a biological target and the design of a molecule intended to act on it. The candidate was investigated for idiopathic pulmonary fibrosis, a disease involving progressive scarring of lung tissue.

The randomised phase 2a study published in Nature Medicine in 2025 included 71 patients treated over 12 weeks. Participants received one of three rentosertib dosing regimens or placebo. The study examined safety and tolerability and explored effects including changes in lung function.

The significance of this example is the progression from AI-supported discovery into a controlled clinical investigation. A phase 2a study provides evidence about a candidate under defined conditions. Its size and duration do not establish long-term clinical benefit or regulatory approval.

From a management perspective, AI can contribute at several points in one development process, while laboratory work, clinical evidence and regulatory development remain major investment requirements. A company’s IP position must be considered alongside the capabilities and partnerships needed to advance the candidate. As generation becomes more productive, the ability to select and validate valuable candidates deserves corresponding attention.

Mattel and Getty Images connect creative production with rights and identity

In June 2024, Adobe described Mattel’s use of Firefly to develop packaging concepts for Barbie. Designers use text prompts to generate imagery, environments and colour ideas, then edit and refine the results. More developed visual concepts can be shared earlier with colleagues, helping teams assess the intended direction before final production.

The example shows how AI changes creative coordination as well as image generation. Marketing and design teams can discuss something closer to the intended result earlier in the process. Human selection, refinement and approval determine which variants become part of the brand’s public presence.

For IP management, a growing supply of visual alternatives raises questions about recognisability and consistent use. Which elements help customers identify the brand? Which variations strengthen that identity? Which materials are suitable for the planned channels and markets? The number of available images alone says little about the strength of the resulting brand assets.

The Getty Images and NVIDIA collaboration, launched in 2023, illustrates another aspect of the same development: image generation built around a curated creative library. NVIDIA also described this approach in its January 2024 announcement concerning Generative AI by iStock. The business proposition combines generation technology with controlled source material and a framework for commercial use.

The broader implication is that provenance and usage conditions become part of the product offered to creative teams. Management needs to connect the origin of inputs, the selected output and its intended use. A rights framework makes particular uses assessable; it should not be interpreted as a universal clearance of every possible depiction, brand reference or application.

A sketch can become the beginning of a patent search

One of the presentation’s examples starts with a technical sketch on a napkin and an explanation spoken into a phone. The illustration makes a practical point: an early idea often exists as a mixture of drawings, conversation and incomplete notes. Requiring a polished technical description before consulting patent information can delay useful feedback.

IPRally Smart Search, introduced in December 2025, accepts materials including images, documents, presentations and free text to help construct a patent search. In the illustrated workflow, spoken explanation is first transcribed into text. The sketch and text can then support a structured description of the technical concept.

This matters because information about existing solutions can influence an idea while it is still easy to change. A team may discover that a mechanism is well known, find a different way to perform the same function or identify an application beyond its original field. Patent information becomes an input to development at an earlier stage.

The interpretation of the input must remain visible and correctable. An AI system may misunderstand a movement, assign the wrong function to a component or overlook a relationship that the engineer considers essential. Checking the extracted technical features is therefore part of the search process. An initial search also needs to be distinguished from a completed patentability assessment or freedom-to-operate opinion.

Research can accompany the development workshop

The next step is an integrated workflow in which search accompanies a team’s discussion. As engineers compare alternative solutions, AI could help retrieve relevant documents, summarise comparable mechanisms and follow new search questions. The workshop image in the presentation illustrates a possible workflow; it does not document a specific customer installation.

Consider a team discussing a clamping mechanism. An initial search may identify familiar lever arrangements. A discussion about reducing force or improving access could then prompt investigation of other mechanical principles. New information feeds into the next design choice, which produces another question for investigation. The process becomes a sequence of learning and adjustment.

The IPBA platform article From Patent Monitoring to IP Intelligence develops the broader management logic: information gains value when it is connected to technology roadmaps, business priorities and decisions. A larger stream of results requires relevance criteria and people able to interpret what it means for the company.

A complementary example is patentbutler.ai, which supports feature-based patent searching and systematic white spot analysis. Users define the technical features of a proposed solution, compare them with existing patent documents and iteratively modify individual features or combinations to investigate potential development opportunities. In Hannes Burger’s CEIPI MIPLM workshop, students applied this approach to smart parking, combining patent analysis with structured invention methods such as TRIZ to develop alternative solutions and preliminary claim structures. This illustrates how searching and creating can become closely connected: findings from patent literature inform the next technical proposal, which becomes the subject of further investigation. Potential white spots remain hypotheses requiring technical and legal assessment. Our platform article Finding the Gaps: How AI is Transforming IP Design and Invention explains the approach and the workshop in detail.

Our lecture AI Supported IP Management in Industrial Engineering provides a complementary operational perspective. Project knowledge needs a usable path from identification through evaluation and protection to reuse. Bringing AI into that path can help teams find and structure knowledge while they are working. It also requires appropriate access to internal information and enough time for substantive review.

Drafting and protection require decisions across IP categories

The July 2026 announcement of the Solve Intelligence and HGF collaboration illustrates specialised AI-assisted patent drafting. The described approach uses customised workflows and templates to help attorneys move from an initial claim set to a reviewable first draft.

A draft is one stage within a larger decision. The company still needs to determine which technical solution matters commercially, which aspects should be disclosed and how the intended protection relates to the product and potential competitive alternatives. Faster document preparation can change the capacity available for this work, but the business objective must guide the process.

The same product can raise several protection questions at once. Its operating principle may call for a patent assessment. Manufacturing parameters may require controlled secrecy. Its appearance and name may need design and brand decisions. Software and visual materials may bring additional ownership and usage conditions.

These questions need coordinated answers. Publishing a technical explanation, providing a partner with data or approving a generated product image can affect several parts of the IP position. The platform’s Know-how Management Deep Dive explains why valuable knowledge needs to be identified, documented, made usable and shared under appropriate conditions. AI increases the practical importance of knowing which information may enter a system and which users may receive the resulting output.

Brand monitoring connects IP with ongoing sales activity

AI-supported visual analysis also changes how companies search for relevant signs and monitor commercial use. WIPO’s Global Brand Database provides image-similarity search for trademark information. Corsearch Logo Detect addresses market monitoring, including detecting altered or disguised logos in listings that may omit the brand name.

These are related capabilities serving different tasks. A database similarity search helps investigate registered signs. Monitoring market listings helps identify uses that may deserve examination. Visual resemblance is an input to assessment; the applicable rights, products, territories and circumstances of use determine the relevance of a finding.

Amazon’s Brand Protection Report for 2024 reports that its proactive controls blocked more than 99 percent of suspected infringing listings before a brand had to find and report them. This is Amazon’s own reported measure of its combined protection system. It does not measure AI performance in isolation or demonstrate that more than 99 percent of all infringement everywhere is detected.

The management opportunity lies in connecting detection with an effective response. Teams need to distinguish authorised partners from suspicious sellers, preserve useful evidence and prioritise cases by likely commercial harm. Feedback about incorrect findings should improve subsequent decisions. Marketers, sales teams and IP specialists consequently need a shared view of the marketplace and clear authority to act.

Character licensing can become a continuing interactive experience

The presentation’s Peppa Pig example shows how AI can change the object being licensed. Hey Peppa Pig, developed with Hasbro by Personality AI, launched on Amazon Kids+ in December 2025. It enables children to interact with the character through voice-based games and activities. WildBrain’s announcement of its acquisition of Personality AI on 24 August 2026 describes interactive character experiences as an extension of its licensing and franchise activities.

My interpretation is that an interactive character requires management of behaviour over time. A static image has a defined appearance. An interactive service also needs parameters for responses, permitted subject matter, updates and the maintenance of the character’s identity. Voice contributions, platform conditions and the intended audience become part of the operating arrangement.

This creates potential for licences involving continuing service delivery and recurring revenues. It also creates continuing work: approval, supervision, technical maintenance and coordination between rights holders and service providers. These responsibilities need to be reflected in the commercial model.

The IPBA research article Licensing of Storyline and Characters in the Video Game Industry offers useful background on structuring rights around characters and fictional worlds. Interactive AI adds another mode of use to questions already familiar from games, publishing and entertainment licensing.

Rightsline AI illustrates the accompanying administrative change. Its tools support extraction of contract terms, rights-availability checks and natural-language questions about rights data. A business can investigate whether content is available for a particular territory, period or use. The usefulness of the answer depends on the underlying contracts and data capturing the relevant permissions and restrictions.

Portfolio management needs a connection to actual use

AI can also help organisations interrogate their existing IP portfolios. LexisNexis’s 2 October 2026 Protégé release for PatentSight+ describes analysis of tagged patent sets, summaries of patent families and comparisons across groups of patents. Users can move between conversational analysis and the underlying result lists.

Such capabilities make it easier to ask questions about a technology field or selected portfolio and inspect the records behind an answer. The next management step is to connect that analysis with the company’s products, future applications and commercial priorities. An apparent gap in patent coverage becomes strategically meaningful only when it relates to something the business needs to control.

Comparable questions arise for other assets. Brand and design portfolios need to reflect the markets and product variants actually used. Content portfolios need a view of available rights, contractual commitments and potential licence opportunities. These are complementary management tasks; a patent-analysis product should not be assumed to perform all of them.

The common requirement is a reliable connection between the asset, the relevant permission or protection and the activity that creates revenue or supports a competitive position.

Economic value can shift as generation becomes more accessible

The final part of the presentation draws management implications from the examples. These are possible changes in the sources of advantage, rather than a quantitative forecast. Their relevance depends on the market and the business model.

For technical solutions, easier access to search, design and simulation can increase the importance of experimental evidence, integration and industrial scale-up. In brands and designs, recognisability, reputation and consistent use can become more significant when generating additional visual variants becomes inexpensive. In content and characters, an established audience, a distinctive fictional world and defined usage permissions may help turn plentiful creative output into a sustainable business. In know-how and data, exclusive evidence, access conditions and controlled sharing can influence who captures value from learning.

AI can improve these activities too, and competing firms may gain access to similar capabilities. Each company therefore needs to examine where its own advantage is created and what remains difficult for others to reproduce. The relevant asset may be a combination of technology, knowledge, relationships and rights.

Bringing IP knowledge into business decisions requires an operating model

As IP knowledge becomes available earlier, the location of IP work can change. Engineers may consult patent information during a workshop. Brand teams may compare marketplace images with their rights while campaigns are running. Studios and licensing teams may check usage conditions while developing an interactive experience.

These possibilities require resources and clear responsibilities. Earlier involvement needs access to plans, technical information and relevant contracts. More findings need a funded review process. Teams need to know which decisions they can make and which require specialist assessment or management approval. Otherwise, additional output can increase the amount of unresolved work.

The platform article The New IP Culture explains how IP decisions arise throughout an organisation. The practical implication is to give people recognisable triggers and usable routes to expertise. Product changes, partner requests, public disclosures and new uses of content should lead to the appropriate decision process.

For the IP profession, this creates opportunities to shape search questions, evaluation criteria, protection choices and commercial arrangements. Delivering that contribution requires a working model that connects expertise with the teams and decisions that need it.

Three questions for management

The presentation closes with three questions. First, where should IP knowledge have an earlier effect: in development workshops, brand management, content production or licensing? The answer identifies the business process in which better information could improve decisions.

Second, which assets should the company build: technical solutions, brands, designs, content, characters, know-how or data? This directs investment towards the resources that support the intended customer benefit and competitive position.

Third, how should those assets create economic value: through the company’s own products and services, licence income, market access, partnerships or controlled sharing? This connects the chosen assets with an explicit commercial purpose.

Together, these questions provide a starting point for a 360° IP strategy in an AI-enabled business. Companies can then decide which connection between creation, checking, protection and exploitation deserves investment, and which roles and capabilities are needed to make it work.