Speed killed the pipeline and nobody noticed

Somewhere in the last two or three years the conversation around content production shifted. I noticed it at industry events first, then in pitch meetings, then in the way clients talked about timelines. The question stopped being "how do we make this well" and became "how do we make this fast." Speed became the product. Not the output, not the craft, not even the strategy behind it. Just speed.

And the tools followed. AI generates the script. AI cuts a rough assembly from transcripts. AI does a first-pass colour grade. AI writes the subtitles in fourteen languages. AI reformats for nine platforms. What used to take a team of specialists working through a structured pipeline now collapses into something that feels like dragging files through a series of automated gates. The pipeline didn't get improved. It got compressed until you couldn't see the individual stages any more.

For a specific kind of content, this genuinely works. Performance marketing assets that live for three days. Social cutdowns that exist to be scrolled past. Iterative ad creative where you're testing forty variations and the winner gets another week of life before the next batch replaces it. Speed is the whole point there. The pipeline is thin on purpose because the output is disposable on purpose.

But I keep running into the consequences of applying that same thinking to work that isn't disposable.

The call at 11pm

A producer I've known for years rang me late one evening, properly stressed. She'd been working on a branded documentary series, four episodes, decent budget, international distribution. The kind of job that used to run through a proper pipeline with clear stages and sign-offs at each gate. Her company had been experimenting with a faster workflow, partly AI-assisted, partly just cutting steps they'd decided were overhead.

The problem was a talent release. One of the contributors had signed a release for UK digital distribution. The series had since been picked up for broadcast in three additional territories. Nobody had gone back to check whether the original release covered that, because the step where someone reviews rights clearances before delivery had been removed from the workflow. It was considered admin. Something that slowed things down.

She spent most of that night and the following morning untangling it. The contributor's agent was not pleased. The broadcaster was not patient about the delay. The total cost of that "efficiency" was two days of crisis management, a renegotiated fee, and a relationship with a contributor that went from warm to cautious. All because a step in the pipeline had been deemed too slow.

Which steps are actually overhead

Some traditional production steps genuinely were busywork. Manual metadata tagging where someone types "exterior, daytime, wide shot" into a spreadsheet for six hundred clips. Format conversion where you wait for a progress bar. Spec sheet lookup where you dig through a broadcaster's PDF to find the audio channel mapping. Dailies ingest where raw files get copied to a drive while everyone waits. Those are progress-bar tasks, as someone at NAB put it, and AI should absolutely handle them. Nobody got into production to babysit a transcode.

The problem is that the enthusiasm for automating the genuinely tedious work has blurred into eliminating stages that aren't tedious at all. They're just inconvenient. Brief approval is inconvenient because it means the client might push back and delay the start. Budget sign-off is inconvenient because it forces you to commit to numbers before you've started spending. Rights clearance before delivery is inconvenient because it requires someone to actually read the contracts. These stages feel like friction. They are friction. But they're the kind of friction that exists because the consequences of skipping them are expensive.

AI doesn't make the stages unnecessary

An AI-generated rough cut still needs a director's eye. I watched someone demo an AI assembly tool last month. It pulled selects from interview footage based on transcript analysis, laid them on a timeline in a sequence that made grammatical sense, and added B-roll from tagged footage. It took about four minutes. The result was technically competent and completely lifeless. It had no rhythm. No breathing room. No sense of when to stay on someone's face a beat longer because the pause was where the meaning lived. A junior editor would have made the same mistakes. A senior editor would have taken longer but produced something a human actually wants to watch.

An AI colour pass still needs a colourist who understands what the story is trying to feel like. The AI can match shots and apply a technical baseline, and that's useful, particularly on projects with mixed camera formats where getting everything into the same colour space used to eat half a day. But the creative grade, the part where the look serves the narrative, that's still a human conversation between the colourist and the director. And it probably will be for a long time.

An AI-generated brief still needs a human to look at it and notice what's missing. I saw an AI brief tool that auto-populated project parameters from a client's email. It got the deliverables right, the timeline right, the format specs right. It completely missed that the client's real concern, buried three paragraphs into their email, was that the last agency had gone over budget and they wanted tighter financial controls this time. An AI reads text. A producer reads between the text.

The split I'm seeing

Teams I respect are using AI to compress the administrative friction inside each stage while keeping the stages themselves intact. They still write briefs. They still do pre-production. They still have review rounds with actual humans making actual decisions. They still track rights. They still archive properly. They just do each of those things with less wasted time at the edges. The AI handles the progress-bar work within each stage. The humans handle the judgment calls at each gate.

Teams that worry me are the ones who've confused eliminating friction with eliminating process. They've removed stages entirely because they looked like overhead from the outside. No formal brief, just a Slack message. No budget lock before production starts, just a rough estimate and we'll figure it out. No rights review before delivery, because legal is slow and the client wants it tomorrow. Everything moves faster right up until something goes wrong, and then there's no paper trail, no approval history, no record of what was agreed, and nobody who can point to the moment where the decision was made.

The difference between those two approaches isn't philosophical. It's financial. The first group absorbs the cost of maintaining process and benefits from the protection it provides. The second group saves time on every project until the one project where the absence of process costs more than all the time they saved combined.

Where AI genuinely helps inside the pipeline

I don't want to sound like someone who thinks AI is a threat. I use AI tools regularly. Automated transcription has changed how quickly I can get through interview-heavy projects. AI-assisted captioning is close to human accuracy on most content, and the time savings are real. Automated proxy generation means I'm not waiting around for media to become workable. Intelligent tagging on archival footage has saved me hours that I used to spend scrubbing through drives looking for a shot I could picture but couldn't find.

These are all things that happened inside a stage of the pipeline. Transcription happens during post. Captioning happens during delivery prep. Proxy generation happens during ingest. Tagging happens during archive. The AI made the work inside each stage faster without collapsing the stages together. That's the version of AI adoption that makes production better rather than just faster.

What gets lost when the pipeline disappears

Accountability. That's the short answer. When a project moves through defined stages with sign-offs at each gate, you know who approved what and when. You know when the brief was locked. You know who signed off on the budget. You know which version the client approved and whether anyone contradicted that approval later. You know whether the rights were cleared before delivery. Every stage creates a decision record, and that record is what protects you when something goes sideways six months later.

When the pipeline compresses into a blur, those decision points dissolve. Work flows continuously from concept to delivery without anyone explicitly saying "this is locked, we're moving forward on this basis." Changes happen mid-stream without documentation. Approvals happen verbally or in chat threads that nobody will be able to find in six months. The speed feels good in the moment. The absence of records feels terrifying in the dispute.

I've been building Telova around the idea that the pipeline matters, that the six stages from intake to archive aren't bureaucracy but protection, and that making each stage faster is a fundamentally different thing from making them disappear.