For most of the last decade, “in-housing” meant one thing: brands pulling headcount and infrastructure away from agencies and building their own production capability. It was a story about people and processes – hiring editors, standing up studios, and building media desks.
In-Housing 2.0 is a different animal. It’s an internal team, working through an orchestrated stack of AI tools, doing what used to require an agency, a production house, or a freelance network. The output looks the same campaigns, ads, content at scale but the mechanics behind it have changed completely. AI hasn’t just sped up production it’s changed what in-house teams are actually expected to own, and with that shift comes a question the first wave of in-housing never had to answer with any urgency: when the work is produced with AI, who actually owns it?
At Brandtech+ we work through this question daily as we support our group’s in-house teams. And in practice, ownership doesn’t start with a legal clause after the work is delivered. It starts before a single asset is generated, with how the tools themselves are governed.
When drafts can be produced in seconds, production capability is no longer the limitation. Judgment is. Accountability is. Craft is. The question shifts from “can we make this ourselves” to “who’s responsible for what gets made, and can we own it” a governance question disguised as a production challenge.
By the time a piece of content exists, several ownership-relevant decisions have already been made which tools were used, whose data trained them, what commercial terms govern the output, and who had oversight of the process. Get that layer wrong, and no amount of legal argument after the fact fixes it.
The law itself is still catching up. In the US, the Copyright Office has been consistent that purely AI-generated content without meaningful human authorship doesn’t qualify for copyright protection. India’s position is murkier still: the Copyright Act, 1957 has no provision for AI-generated content at all, leaving authorship largely undefined Ankit Sahni’s “Suryast” being a case in point. Regulators on both sides are working to catch up, but for now, brands operating across markets are building on ground that hasn’t finished settling.
Given that uncertainty, we take the position that governance has to come first. Our teams work only with tools that have been vetted and greenlit group-wide and signed off by the client not because every unapproved tool is unsafe, but because “approved for creative experimentation” and “approved for client-facing production” are two very different bars, and conflating them is where most avoidable risk creeps in.
That’s also why we work through Pencil, the platform The Brandtech Group acquired in 2023 and that all our teams are trained on. Rather than tying production to a single model or vendor, Pencil aggregates the best available models in the market into one environment so we’re never locked into one provider’s roadmap, and we can always use the right model for the job.
When Gartner predicts that by 2027, more than half of the GenAI models enterprises use will be tailored to their specific industry or business function, it raises the question: how will it be tailored to a specific brand? That’s the type of question being addressed in how Pencil is built today: each brand works from its own dedicated workspace, trained on that brand’s guidelines and documentation, so output isn’t generic it’s shaped by material the brand already owns, which is part of what makes it ownable in the first place. Just as importantly, using Pencil means none of our client outputs are fed back to train the broader underlying models.
Typical copyright protection still applies automatically to original work created by employees in the course of their employment. But that default doesn’t necessarily hold the same way for AI-generated output. Even with clean governance and appropriate tool usage, the authorship question doesn’t disappear it just becomes something you can actively manage. That ambiguity matters more, not less, for in-house teams like ours, where production is often distributed across a global team rather than sitting with a single named creator.
Practically, this means a human must stay meaningfully in the loop throughout the process. What strengthens the position is real, demonstrable modification: rewriting AI-generated text in a human voice, or folding AI output into a larger piece of work that clearly required human creative decisions. There’s no bright line here. But the principle holds: the more substantive the human hand, the stronger the claim.
That also reframes what craft means in this era. For most of this profession’s history, creative leadership meant shaping the work directly writing the line, framing the shot, sitting in the edit suite. Increasingly, it’s leaning closer to curation: setting the taste and judgment a system executes against and knowing what not to generate in the first place arguably the harder skill of the two.
You’ve heard the line before: it’s not AI that takes your job; it’s someone who knows how to use AI. Cliché, sure. But strip away the LinkedIn-post version of that idea, and there’s something real underneath it: craft is what turns a generated draft into something a brand can call its own. It’s also why we hire the way we do leaning on executional capability and strong creative foundations over tool fluency.
That standard doesn’t stay contained to a hiring philosophy it’s the same one that will separate the brands built to last from the ones chasing the current moment. The real work of In-Housing 2.0 is not in showing how much content they can produce and how fast. It’s about building an operating system with humans at the core, one that lets them produce responsibly, indefinitely. Content is abundant now. Trust isn’t. Any brand can generate more of the former; few will be able to answer, asset by asset, why it’s theirs to keep.
(Views are personal)
