Dear Readers, Strategists and Game-Changers,
Hope you are doing great!
There is a natural reflex in innovative companies. When something valuable is created, someone eventually asks: “Can we patent it?”
It is an important question. But it is rarely the first question that should be asked.
The better question is: “What exactly creates the advantage — and what is the best way to preserve it?”
Because not every competitive advantage should become a patent.
Some should. Others should remain secret.
Some should be embedded into contracts, processes, software architectures or data structures. Others should deliberately be published. And some are not adequately described by any traditional intellectual property category at all.
They live in the way people work. In the sequence of operations. In the small decisions an experienced engineer makes without writing them down.
In the way a team calibrates a machine, qualifies a supplier, trains a model, combines ingredients, recruits a collaborator, manages a client or solves a recurring technical problem.
I call this the gesture, or the Cognitive Assets!
And sometimes the gesture is worth more than the patent.
The patent instinct
Patents are powerful strategic instruments. They can create exclusivity, support licensing, strengthen negotiations, deter competitors, attract investors and contribute to enterprise value.
But a patent is based on a bargain.
The owner receives a potentially enforceable exclusive right in exchange for disclosing the invention to the public. WIPO describes that exchange explicitly, and the EPO similarly refers to disclosure as being at the heart of the “patent bargain.”
That trade-off is perfectly rational when the invention can be observed, reverse-engineered or independently recreated. If a competitor can buy your product, analyse it and understand the invention, secrecy may provide very very little durable protection.
A patent can then transform something visible into something legally exclusive.
But imagine the opposite situation.
Suppose the important part of your advantage happens inside your factory. Or inside your algorithmic workflow. Or during calibration. Or in a sequence of parameters that customers never see. Or in hundreds of small operational decisions accumulated over ten years.
Publishing that knowledge in a patent application may sometimes give competitors information they would otherwise have struggled to obtain.
The strategic question therefore changes. You are no longer asking simply: Can it be patented?
You are asking: Should the world be allowed to see it?
What exactly is “the gesture”?
Consider two companies using apparently similar technology. They may buy comparable equipment. Employ engineers with similar degrees. Use the same programming languages. Purchase components from the same suppliers.
And yet one consistently produces better results.
Why?
Perhaps its engineers have discovered that changing three parameters in a particular sequence reduces failure rates.
Perhaps its manufacturing team has developed an undocumented calibration method.
Perhaps its sales organisation knows exactly how to identify customers with a particular hidden need.
Perhaps its researchers know which approaches not to try.
Perhaps years of failed experiments have produced an internal map of dead ends that competitors must still discover themselves.
None of these advantages necessarily looks spectacular in isolation.
But competitive advantage often does not reside in one spectacular insight.
It resides in accumulated asymmetry.
One company knows something another does not. One organisation can execute something another cannot.
And that asymmetry compounds.
That is where know-how becomes strategic.
Trade secret protection is not “doing nothing”
One of the most persistent misconceptions in IP strategy is that there are two choices: patent something, or leave it unprotected.
That is wrong.
Trade secrets represent a deliberate protection mechanism in their own right.
Internationally, the core principles of trade secret protection are set out in Article 39 of the WTO TRIPS Agreement and reflected in legal frameworks such as the EU Trade Secrets Directive. Broadly speaking, information qualifies for protection when it is not generally known or readily accessible within the relevant professional circles, derives commercial value from its secrecy, and has been subject to reasonable steps to keep it confidential. Such protection can extend to technological knowledge, business information and know-how, provided these conditions are met.
The third criterion deserves particular attention: reasonable steps.
A company cannot credibly treat everything as secret while allowing everything to circulate freely.
Trade secret strategy therefore requires architecture.
Who can access the information? Where is it stored?
How is it classified? Who can download it?
What happens when an employee leaves?
What do supplier agreements say?
What does the company disclose to investors?
Are critical processes documented?
Can the company later demonstrate that particular information was confidential at a particular date?
Secrecy is not the absence of IP strategy. Properly implemented, secrecy is IP strategy.
The four questions before filing a patent
Before deciding between patenting and secrecy, I would ask four questions.
First: can competitors observe the advantage?
If the innovation becomes apparent as soon as the product reaches the market, secrecy becomes fragile. If competitors can reverse-engineer it relatively easily, patents may provide much stronger protection.
But if the valuable mechanism remains inaccessible from the final product, secrecy becomes more interesting.
Second: how long might the advantage remain useful?
Patent protection is time-limited.
Certain processes, recipes, datasets, models or operational methods may potentially retain value for much longer if secrecy can realistically be preserved.
Other technologies will be obsolete long before a patent expires. The relevant horizon therefore matters.
Third: could infringement realistically be detected?
A legal right that cannot realistically be enforced may have limited practical value.
If the invention is used entirely inside a competitor's inaccessible manufacturing process, proving infringement may be extremely difficult.
Sometimes the ability to keep your own process secret is strategically stronger than trying to monitor everyone else's.
Fourth: can your organisation actually keep a secret?
This is the question companies often avoid. Trade secret protection is not only a legal question.
It is an organisational capability.
A business whose confidential knowledge is scattered across personal laptops, Slack messages, uncontrolled AI systems, suppliers, consultants and departing employees does not really possess a trade-secret strategy.
It possesses a hope.
The human problem
There is another important distinction.
Companies do not own human beings. And they cannot simply declare everything an employee has learned to be proprietary.
For example, the EU Trade Secrets Directive explicitly distinguishes protectable trade secrets from the experience and skills employees honestly acquire in the normal course of their employment.
That distinction makes the management of know-how more interesting, not less.
A company should ask: What exists only inside someone's head that should become institutional knowledge?
Not because the company should confiscate that person's general expertise.
But because critical company-specific procedures, experiment histories, technical decisions, customer knowledge and proprietary methods should not disappear simply because one person changes jobs.
This is where IP strategy intersects with knowledge management, HR, cybersecurity and corporate governance.
From IP portfolio to knowledge architecture
The most sophisticated companies will increasingly stop thinking of IP merely as a portfolio of registered rights.
They will build a knowledge architecture.
Inside that architecture, some inventions become patents. Some algorithms remain confidential. Some datasets receive access controls and contractual protection.
Some methods are documented as proprietary processes. Some discoveries are deliberately published to prevent competitors from patenting them.
Some brand elements are registered. Some knowledge is distributed across teams so that no single employee becomes a point of failure.
And some sensitive information is intentionally compartmentalised.
$The objective is not to patent as much as possible. Nor is it to keep everything secret.
The objective is to decide, consciously, which form of control maximises the business value of each strategic asset.
A simple exercise
Walk through your organisation and ask your teams one question: “What do we know how to do significantly better than most competitors?”
Do not start with registered patents. Do not start with legal documents.
Start with reality.
Ask engineering. Ask operations.
Ask sales. Ask customer support.
Ask the founders. Ask the people who have been there for ten years.
The answers may surprise you.
You may discover that some of your most valuable intellectual assets have never been formally identified.
And once you identify them, ask the second question: “What would happen if our strongest competitor knew exactly how we do this?”
That question often reveals more about protection priorities than a spreadsheet containing hundreds of IP rights.
When the gesture becomes an AI skill
Artificial intelligence adds another dimension to this question.
Because some of the knowledge that makes a company difficult to replicate can now be organised into instructions and resources that guide an AI agent.
An agent skill, for example, can combine task-specific instructions, reference materials and executable scripts. Rather than retraining the underlying model, it provides a structured way to approach particular work.
The strategic interest lies in what a company puts inside that structure.
Consider a manufacturing business investigating a recurring defect.
One team asks a general-purpose AI model to suggest possible causes.
Another equips an agent with its internal diagnostic sequence, proprietary calibration parameters, records of previous failures and criteria for escalating the problem to an experienced engineer.
Both may use the same model.
But they are not giving it the same knowledge.
The model may be widely available. The method need not be.
In this example, part of the advantage lies in the accumulated experience that the company has translated into a usable workflow.
The sequence matters. The exceptions matter.
Knowing which apparently reasonable approach has already failed matters. And so does knowing when the agent should stop and ask a human.
This is where I see a deeper connection between AI strategy and intangible asset strategy.
A carefully developed skill library can be more than a collection of productivity tools. It can become a structured expression of company-specific know-how.
The gesture has not disappeared. Part of it has become executable.
But an important qualification follows.
Writing down a method does not prove that an agent will apply it correctly. A convincing demonstration is not the same as a reliable capability. Skills need evaluation on representative tasks, attention to failure cases and revision based on observed behaviour.
The strategic asset is therefore not simply an instruction file.
It is the method, the supporting knowledge and the evidence that the resulting workflow performs as intended.
There is also a distinction worth preserving: An open format does not mean the knowledge inside it must be public.
The Agent Skills specification, for example, expressly accommodates a proprietary licence designation. A company can use a shared technical format without choosing to publish its own methods.
That distinction matters because capturing know-how creates a tension.
Knowledge that was once scattered across conversations, individual experience and undocumented practices may become easier to preserve and reuse.
It may also become easier to copy.
A skill containing diagnostic rules, internal examples and the lessons from years of failed experiments could reveal far more than the final report it helps produce.
The protection question therefore moves with the knowledge.
Who can read, modify or export these skills? Which external systems receive their contents? Who approves changes, and who can establish which version was used?
These are not merely administrative questions. Skills can include executable code and instructions that connect to external services; untrusted skills can also introduce vulnerabilities or cause data to leave the intended environment. Their contents, dependencies and connections therefore deserve scrutiny.
Nor does calling something a “skill” automatically make it a trade secret. The underlying information must still meet the applicable legal requirements. The TRIPS framework centres on secrecy, commercial value arising from that secrecy and reasonable steps to preserve it.
Local AI: bringing intelligence to the knowledge, not the other way around
There is another strategic choice behind all of this.
Where should that intelligence run?
When an agent’s skills contain proprietary methods, confidential examples and the lessons from years of experimentation, choosing where those skills are processed becomes part of the knowledge architecture.
A properly configured local AI deployment allows a model to run on the organisation’s own computers or on-premises servers. Local inference and document analysis can operate without sending their contents to an external AI provider, and some implementations can work entirely offline once the necessary models are installed.
For the manufacturing business described above, imagine an agent consulting its proprietary diagnostic skills, calibration procedures and internal failure records entirely within that environment.
The ambition would be straightforward:
Make the company’s knowledge more useful without making external disclosure a prerequisite for using it.
But an important distinction follows.
A local application is not necessarily local AI.
Software installed on a company laptop can still send requests to a remotely hosted model. Ollama, for example, supports both local execution and cloud-hosted models. The interface alone does not establish where processing happens.
I would therefore examine the whole workflow: the model, the skill library, document retrieval, connected tools, logs and backups. Which components remain internal? Which communicate externally? What information can they transmit? Security guidance similarly stresses controlling information sent to external services and protecting prompts, documentation and logs as assets in their own right.
There is also a distinction on the cloud side.
Some commercial AI services commit not to use customer inputs and outputs for model training by default. But those commitments are separate from the terms governing processing and retention.
“Not used for training” does not mean “never leaves our infrastructure.”
That does not make cloud AI inherently unsuitable.
It means the decision should be deliberate, based on the information involved, the applicable service terms and the organisation’s requirements.
Nor does local deployment automatically make a system secure.
Access controls, secure storage, scrutiny of third-party components, monitoring and controlled updates remain necessary. Keeping the model inside the organisation does not remove the need to govern the people, software and tools around it.
For particularly sensitive know-how, I would therefore assess a local-first architecture (like our DELSOL AI solutions: Eve, Athena, Hank or Alfred): keep proprietary skills and their supporting knowledge internal, and permit external processing only through explicitly approved workflows.
That choice should still be tested against task performance and the resources required to operate the system properly.
The objective is not local AI at any cost. It is control by design.
Because when a skill captures something that makes the company difficult to replicate, the question is no longer only: “Which model gives us the best answer?”
It is also: “How can we bring intelligence to our know-how without unnecessarily sending our know-how elsewhere?”
For me, this adds another question to the exercise above: “Which parts of our distinctive know-how are we turning into capabilities for our agents — and how are we preserving control over them?”
Because the objective should not be merely to automate what makes the company different.
It should be to make that difference more useful, more durable and more consistently available to the organisation.
Without making it equally available to everyone else.
The final thought
We tend to imagine intellectual property as something visible.
A patent number. A trademark. A registered design. A copyright notice.
But some of the most powerful intellectual assets inside an organisation are almost invisible.
They exist in routines. In judgement. In sequences. In accumulated learning.
In things that experienced people do almost automatically.
Sometimes the strategic task is not to expose those things to obtain a right. Sometimes it is to recognise them, capture them, govern them and keep them invisible.
Because the question is not: “How much IP do we own?”
The question is: “How much of what makes us difficult to replicate do we actually control?”
And that difference can be enormous.
