Three years is a long time on the Internet. 

Long enough for novelty to wear off. Long enough for early panic to cool. Long enough for hot takes to age badly. And long enough, in the case of generative AI, for freelancers to move past the question of whether tools like ChatGPT exist and into the harder question of how they actually fit into real work. 

When ChatGPT first appeared, the conversation split quickly into camps. AI was either a looming threat or a productivity miracle. A replacement for creative labour or a shortcut to better thinking. A friend or a foe. With some distance now, that binary feels increasingly inadequate. At a recent CFG Experts panel, moderator George Butters sat down with Dr. Owen Brierley and Dr. Nadine Robinson to reflect on what’s changed since those early days—not just in the tools themselves, but in how freelancers understand, trust, and use them. 

What emerged wasn’t a verdict so much as a reframing: AI hasn’t resolved into something we can neatly label. Instead, it has settled into something messier, more powerful, and far more dependent on human judgment than many early narratives suggested.

The Myth of Arrival

One of the most persistent misconceptions about AI is the idea that it has a clear endpoint—that we are moving toward some stable, final version of the technology where rules will be settled and expertise fully formed. In reality, the ground still feels unsteady. 

As Dr. Owen Brierley noted during the conversation, even those working closely with AI systems hesitate to call themselves experts. Models evolve quickly. Capabilities shift. Interfaces change. What felt true even a short time ago may no longer apply. 

That instability matters. It means confidence can easily outpace understanding. It also means freelancers—who are often expected to sound authoritative to clients—are navigating a landscape where certainty is difficult to claim. Rather than mastering AI, the panel suggested, freelancers are learning to work with uncertainty. That may be a more honest skill to cultivate.

Prediction Is Not Intelligence

One of the most important clarifications to surface in the discussion was a reminder of what large language models actually do: They predict. They don’t understand in the way humans do. They don’t reason independently. They generate statistically plausible responses based on patterns in data. That distinction is easy to forget when outputs sound fluent, confident, and—at times—uncannily human. 

Dr. Nadine Robinson emphasized how risky that fluency can be when it’s mistaken for accuracy. Hallucinations, bias, and subtle errors don’t announce themselves. They arrive wrapped in confidence, which makes human oversight not just helpful, but essential. This is where the “black box” problem becomes practical rather than theoretical. Freelancers aren’t just users of AI, they are responsible for what leaves their hands. When AI-assisted work is wrong, misleading, or harmful, accountability doesn’t sit with the model. It sits with the person who chose to rely on it.

The Intern Metaphor Still Holds

Early in the AI conversation, a popular metaphor emerged: AI as an intern. Capable, fast, occasionally impressive — but unreliable without supervision. That metaphor still resonates. As Dr. Brierley observed, the intern framing helps recalibrate expectations. You wouldn’t hand an intern full authority without review. You wouldn’t publish their work unchecked. You also wouldn’t dismiss their contributions when they’re useful. For freelancers, this framing is practical. AI can draft, summarize, brainstorm, reorganize, and accelerate. It can also mislead, fabricate, and flatten nuance. The value lies not in delegation, but in collaboration — and collaboration requires judgment. In that sense, freelancers may actually have an advantage. Independent workers are already used to managing tools, workflows, and responsibility without institutional buffers. They understand what it means to be accountable for final output.

Human-in-the-Loop Is Not Optional

Across the panel, one idea surfaced repeatedly: human-in-the-loop isn’t a nice-to-have safeguard. It’s central to ethical AI use. Verification, transparency, and process matter more than the tool itself. That means knowing when AI has been used, how it has been used, and where human decision-making intervenes. 

In educational settings, this has sparked conversations about policy and disclosure. In freelance work, it shows up as process: fact-checking, attribution, version control, and clear boundaries around what AI can and cannot do. The panel resisted blanket prohibitions just as strongly as blind adoption. Banning tools doesn’t teach discernment. Uncritical use doesn’t either. What’s required instead is literacy—not technical mastery, but ethical and contextual understanding.

Freelancers at the Centre of the Question

If institutions are still defining their stance on AI, freelancers are already living with the implications. 

Clients are curious. Expectations are shifting. Some prioritize speed above all else; others want reassurance that human expertise remains central. Freelancers often have to navigate these conversations without clear industry standards to rely on. That’s where judgment becomes a professional differentiator. 

Knowing when AI meaningfully helps, and when it undermines quality, isn’t something a tool can decide for you. It’s developed through experience, reflection, and sometimes missteps. In that sense, AI hasn’t simplified freelance work. It has added a layer of responsibility.

So…Friend or Foe?

By the end of the discussion, the original question felt less urgent. 

AI isn’t a friend in the way a colleague is. It isn’t a foe in the way a competitor is. It’s a system—powerful, limited, and shaped by how humans choose to use it. What matters isn’t allegiance, but agency. Freelancers don’t need to solve AI or predict its trajectory. They need to think clearly about how it fits into their work, their values, and their accountability to others. 

Three years on, the most useful posture may be neither optimism nor fear, but attentiveness. Paying attention to what these tools actually do. Paying attention to where they help and where they fail. And paying attention to the human judgment that remains—quietly but decisively—at the centre of the work. 

For those continuing to wrestle with that tension, the full CFG panel, 3 Years After ChatGPT: Friend or Foe? offers a thoughtful, grounded conversation — not about definitive answers, but about better questions.