Agentic First is a Strategy, just not the one you intended

Last time, we met Bob, a prolifically effective AI replacement for software engineers, and discovered how to reason about him so we don't fall into the misunderstanding trap many businesses are currently digging for themselves.

Of course, a common and expected retort to that is "but AI will get better."

So let's ignore all misgivings about AI and assume it's perfect. Fast, cheap, powerful, error-free, fabulous. In this world, all businesses would be mad not to be AI First and fully Agentic.

Or would they?

Outside of debates about what AI can do, or not do, most of the discourse is about whether it is good. Good for the environment, good for society, good for the creators whose content was absorbed into it in the first place.

Almost nobody is talking about what comes out of these models and who owns it.

We'll get into the implications in a bit, but spoiler alert: you probably don't.

Copyright for AI-generated content is either non-existent, limited, or legally uncertain across most major jurisdictions. In general, where there is little or no meaningful human authorship, copyright protection becomes much harder to establish.

If you want the detail, Norton Rose Fulbright, one of the world's top law firms, published a twelve-jurisdiction guide in May 2025 (and another about the risks of infringing someone else's work using AI).

Anything your Agentic AI First business publishes that was generated wholly by AI is unprotected. Which means your competitors can copy it, and you cannot sue.

Let's run a reductive abstract thought experiment.

Uber was in the news recently because they burned through their 2026 AI token budget in four months, and now the COO is questioning the value AI adds to their business.

But what if that hadn't happened? What if they instead made an announcement about how successful the experiment had been and how AI was now replacing all 5,000 of their developers?

With a handful of skilled business analysts, coordinating swarms of agents, Uber would take first-mover advantage and dominate the on-demand ride share business.

For a while, at least.

Because here comes Nuber. Nuber is also 100% Agentic First. They hire some equally skilled analysts and overtly and outrageously copy everything Uber does. The actual code will differ because language models are non-deterministic. No one is copying code verbatim, so there is nothing to litigate.

Neither company has meaningful copyright protection over the generated output. Neither has trade secrets worth the name.

Nuber can match Uber indefinitely at negligible cost. Competitive differentiation collapses to zero.

The only moat either company has is its brand, its customer relationships, and maybe how well it treats its drivers. For many companies, this might not feel threatening. A great brand with loyal customers is a powerful thing, unless of course you are Kodak, Blockbuster, BlackBerry, Woolworths, Jessops, Encyclopaedia Britannica, Nokia, Myspace, HMV.

Are the CEOs of Uber and Nuber able to sleep soundly at night?

No. Because here comes Huber.

Huber is 100% human. They have a higher wage bill for sure. And in pure output terms, they're slower. But the Huber codebase is human-authored and copyrightable. Any novel technical implementations can be patented. Genuine know-how sits in its people's heads and cannot be extracted by prompting a model.

Anything Huber does cannot be replicated on demand, because Huber's research & development is not encoded in any publicly available model. Worse, for Uber and Nuber, Huber can copy their features freely, but not vice versa.

Over time, the Uber and Nuber share prices will reflect their inability to differentiate. Maybe Huber will buy them both just for the customer base and distribution. Bin their codebases and make the analysts redundant.

OK. It's just a thought experiment. But it's also a standard technology consolidation pattern, with an IP dimension that most boards will not have on their radar.

What can be done?

Well, the standard argument for human oversight of AI is usually about quality, accuracy, and bias-checking. It should also be part of an IP strategy. Though it means no 100% generated code should be going into production unchanged. Meaningful human contribution restores copyright. Human inventors strengthen patentability. Humans who shaped the output create defensible know-how.

Agentic development may, in legal terms, be a decision to produce unprotectable output. Companies approving that decision should understand what they are giving up and not wander into unrecognised asset disposal.

Does all this mean the economic argument for all AI is significantly undermined? Well, the Telegraph certainly thinks so now. We'll get into that next time.

If you'd like to find out more about how we use AI whilst keeping our clients' IP safe, drop us a line.