The industry took the best part of 15 years to move from tape to file-based workflows. Not because the technology wasn't ready -- the early tools existed long before the shift was complete. But because the real work of a transition like that isn't the technology. It's everything around it.

Anyone who lived through it remembers the pattern. Vendor demos that looked magical and deployments that didn't. File formats that broke when you moved them between systems. Codecs that claimed to be compatible and weren't. Playout chains that worked in the lab and fell apart at 2am during a live programme. Archive systems where the metadata drifted quietly for years until someone needed a specific asset and couldn't find it.

Then there was the people part, which was usually bigger than the tech. Operators who had learned a craft over decades being told the craft was changing. Training programmes that underestimated how much of the knowledge was tacit -- the thing you just knew from doing it 10,000 times. Teams that built shadow workflows around the new systems because the new systems didn't fit how they actually worked.

And the timeline. Every vendor in 2002 would have told you the industry would be fully file-based by 2007. It wasn't. Not even close. The last tape operation I visited was still running in 2019.

That's not a cautionary tale about slow adoption. It's the normal shape of a genuine operational shift in broadcast. The timeline stretches because the system under change isn't the technology -- it's an interlocking set of people, processes, vendors, standards, budget cycles, and risk appetites. All of those adjust at their own speed, and the slowest one sets the pace.

Agentic AI is in the same position now that file-based workflows were in the early 2000s.

The underlying technology works. There are real deployments delivering real value. The demos are impressive. The vendor claims are ahead of the reality. Integration is harder than it looks. The operational impact is bigger than most people realise. And there's a healthy ecosystem of companies promising transformation that will deliver incremental improvement at best.

This is not a criticism of the technology. It's a description of where we are.

The broadcast operators who did well during the tape-to-file transition weren't the ones who moved fastest or slowest. They were the ones who were honest about the shape of the transition. They invested in architecture that could evolve rather than betting on specific vendors or specific formats. They built teams with the skills to run hybrid environments for as long as needed. They wrote off the parts of the first wave that didn't work and kept investing in the parts that did.

That's the playbook for agentic AI too. The technology will embed itself into your operation over the next 5 to 10 years. The question isn't whether, it's how you position yourself to ride the S-curve without getting burned by the hype, the early-vendor failures, and the gap between promise and delivery.

Treat it like tape-to-file. You've seen this movie before.

What I wish someone had told me about tape-to-file in 2005

I was working in the early file-based workflow projects when the big promises were landing. Here's what I wish someone had told me then, which applies directly now.

The business case you wrote in 2005 was wrong. Not because the logic was bad, but because the real benefits came from second-order effects nobody modelled. The ability to run parallel workflows. The ability to automate quality control. The ability to repurpose content faster. The savings you wrote down as headcount reduction actually materialised as capacity, you did more with the same team, rather than the same with fewer.

The vendor you bet on probably didn't survive. Or got acquired. Or pivoted. The technology choices that mattered weren't the vendor choices, they were the standards choices. Open formats. Documented interfaces. Exit strategies.

The training budget you set was too small. Always. Every time. By a factor of about two.

The operators who fought the hardest at the start often became the best advocates once they understood the tools. The ones who adopted eagerly and uncritically often built the most fragile workflows. Scepticism is a feature.

Most of the project plans from that era assumed a clean cutover. Almost none of the real operations cutover cleanly. They ran hybrid for years. Build for the hybrid period. It's longer than you think.

None of this is specific to tape-to-file. It's the shape of any real operational transition in broadcast. The people who went through it and paid attention have a genuine advantage when reading the current wave of agentic AI hype. The patterns repeat.

Worth paying attention to: Any vendor who talks about the hybrid period rather than the end state. They've seen this before.

Overhyped right now: "AI-first" broadcasters. Nobody is AI-first yet, and the ones claiming to be are either misusing the term or doing something risky.

Worth reading: Anything published by people who were in broadcast engineering during the HD transition or the file-based transition. The patterns map directly onto what's happening now.

The Clean Feed is published every Thursday. Forward this to someone who builds broadcast systems.