Before We Talk About AI, We Need to Talk About Us

“How Are People Supposed to Make Ends Meet?”

“How are people in the industry supposed to make ends meet if nothing starts happening again?” Finni Jóhannsson, managing director of Kukl, asked that question on Icelandic evening news this month, standing in front of a camera‑rental house that has literally had to move its offices into the warehouse to save costs. The largest equipment rental in the country is shrinking to fit inside its own shelves, because both local and international shoots have fallen off a cliff over the last couple of years.

You can feel that question echoing through the whole ecosystem. Crews who built their lives around a mix of Icelandic features, international service work and series are watching the calendar thin out. Suppliers cut square meters and salaries. Younger people who pushed their way into the industry in the good years quietly drift back to other jobs. All this is happening in a country that advertises up to 35% production refunds and has spent decades building a reputation as a place that somehow makes impossible shoots work.

It would be comforting to blame this moment on one thing: the last government, the new government, tax incentives, end‑of‑Covid jitters, or now, conveniently, AI. But that would be dishonest. What is happening in Iceland is part of a longer, wider shift in how film and television are financed, produced and distributed. AI hasn’t replaced those forces; it has arrived on top of them.

This Is Bigger Than Iceland

If you zoom out from our little island, the pattern looks uncomfortably familiar. In North America, the number of films released in theaters every year dropped sharply after 2020 and has not recovered to pre‑pandemic levels. Covid shutdowns wiped out a year of production, the dual Hollywood strikes in 2023 halted most scripted work again, and studios used the disruption to lean harder into a streaming‑first strategy: fewer wide releases, more titles going straight to platforms, and a slate that tilts towards franchises and big‑budget “event” films.

At the same time, box office is being “stabilized” by raising prices and lowering volume. Theaters in the US are now generating roughly similar revenue with hundreds of millions fewer admissions, because tickets are more expensive and the experience is being upsold with recliners, cocktails and premium formats. As one recent analysis put it, this is not a temporary dip waiting to bounce back; it is a permanently smaller theatrical market being reframed as recovery, with studios deliberately releasing fewer films and treating cinemas as one marketing window among many rather than the organizing centre of the business.

On the production side, the hubs that used to feel invincible are wobbling too. Film LA’s latest Scripted Content Study shows that total scripted releases in the US fell by more than 13% between 2023 and 2024, and the number of projects actually filmed in Los Angeles dropped even faster. Scripted television series shot in LA declined by around a quarter in a single year, and LA’s share of all US scripted releases fell from 22% to 18%. Behind those percentages are crew days lost, vendors seeing less work, and people wondering whether to hang on or leave.

When an ecosystem is already shrinking and consolidating like this, every new technology gets absorbed into that logic. It’s in this context—not in a vacuum—that AI is being deployed, sold and argued about.

The Structural Forces Reshaping Film (Before AI)

Long before anyone typed a prompt into a model, the basic economics of film and TV had started to tilt. The first big shift is streaming: instead of commissioning a wide spread of titles and letting some surprise hits carry the misses, major studios are now funneling money into fewer, bigger, safer projects that can justify a global marketing push and a prime slot on a platform’s homepage. Theatrical has become the home of franchises and “event” films, while mid‑budget dramas, comedies and local stories are quietly redirected to streaming, where success is defined by retention curves and algorithmic engagement rather than cultural impact.

Subscription models intensify that pressure. When your revenue comes from monthly fees, the safest bet is content that keeps people habitually watching rather than work that might matter deeply to a smaller audience. Shows and films are greenlit and cancelled based on how they move a few internal metrics, and those metrics rarely reward risk, regional specificity or slow‑burn loyalty. Over time, that squeezes exactly the kind of projects small countries tend to make: culturally specific, mid‑scale, sometimes formally weird.

On top of that sits the incentive arms race. Over the past decade, more and more countries and regions have pushed their production rebates into the 25–40% range to lure shoots, with Iceland now offering refunds of up to 35% of eligible spend. The UK, Canada, Ireland, Italy and others are in similar bands. The unintended consequence is that no single place is uniquely cheap or generous for long; big buyers can play locations off against each other, and work flows to wherever the latest combination of rebate, currency and politics is most convenient that quarter. Service economies built on being “the place to go” suddenly find themselves one tile among many on a global incentive board.

Then came the shocks. Covid shutdowns wiped out production for months at a time; when work resumed, it did so under heavy protocols and with backlogs everywhere. The 2023 strikes stopped most scripted work again, emptied pipelines and forced studios and streamers to confront how much money they had tied up in unreleased or under‑performing content. Consolidation, debt and investor pressure followed. In boardrooms, the mandate coming down to film and TV divisions is not “experiment and grow” but “stabilize, deleverage and hit margin targets” for years, not just one bad quarter.

Meanwhile, there has been a quiet efficiency drive running through every department. The shift from film to digital, from on‑prem to cloud, from local to remote has already allowed productions to trim shooting days, reduce duplication and run leaner crews. Editing, sound, grading and VFX have been progressively reorganized around global pipelines and remote collaboration tools, often moving work to lower‑cost markets or larger multi‑client vendors. None of this required generative models; it was standard process and technology optimization.

If we don’t acknowledge these forces—streaming economics, incentive competition, post‑pandemic balance sheets and a decade of efficiency gains—then talking about AI in isolation becomes a distraction. We end up blaming or worshipping the latest tool instead of seeing the system it is being dropped into.

Then AI Walks Into the Room

Into this environment walks generative AI. In film and TV today, it is not a single monolithic thing but a scatter of tools and experiments. Studios, vendors and independents are already using models for concept art, storyboard passes, background design, character and environment modelling, roto and cleanup, upscaling, dubbing and automated dialogue replacement, sound design elements and temp scores. In the offices around the work, AI drafts casting breakdowns, meeting notes, marketing copy, audience‑segmentation reports and internal decks. Writers and producers quietly use assistants to summarize research, spin variants on loglines and outline scenes.

The industry is starting to count the potential impact in jobs, not just features. A 2024 “Future Unscripted” report commissioned by the Animation Guild estimated that by 2026, generative AI could disrupt around 203,000 jobs in US entertainment, broadly defined. In their survey of studios and vendors, roughly three‑quarters of respondents said AI had already enabled them to eliminate, consolidate or redefine roles. This is not a theoretical conversation about 2035; it is an acceleration of a restructuring that was already underway.

Some tasks are clearly more exposed than others. Analyses from unions and newsrooms point to 3D modelling, concept and environment design, rotoscoping, compositing, certain strands of sound design, dubbing and translation, marketing artwork and basic editorial as especially vulnerable, because they involve repeatable pattern work that current tools handle surprisingly well. Many of those roles have historically been entry points into the industry—the places where people learned pipelines, language and taste before moving up.

Put that alongside the macro pressures and the likely pattern emerges. In a world of fewer greenlit projects, a smaller theatrical slate, intense streaming cost control and a global incentive auction, it is naive to assume that AI will first be used to create more work or gentler schedules. As several commentators on Hollywood’s “AI dilemma” have pointed out, the immediate temptation is to use AI to hit lower cost targets on the work that still exists: fewer concept artists per show, fewer assistant editors per series, smaller in‑house marketing teams, more work moved to vendors who promise AI‑boosted efficiencies.

That doesn’t mean AI cannot be used differently. It does mean that unless we look honestly at the forces already reshaping film, we will misread what AI is doing now: not descending from the sky as a neutral productivity tool, but being steered by a system that has been in contraction mode for nearly a decade.

What This Looks Like From a Small Market

If you live in a small production market, all of this feels less like an abstract trend and more like a weather report you can’t escape. Through the 2010s, tax incentives and a reputation for reliability pulled a steady stream of international projects into Iceland; the country now offers refunds of up to 35% of eligible spend and promotes itself as a one‑stop shop for complex shoots. Similar stories played out in Ireland, parts of Canada and Eastern Europe: governments invested in rebates and infrastructure, and crews learned to deliver world‑class work in difficult conditions.

The underlying model, though, is structurally fragile. A service‑heavy ecosystem depends on international orders to keep people busy; domestic volume is rarely enough to fill the calendar. When global production slows, or streamers rebalance their slates, there is no large home market to soak up the slack. Every cancelled series or delayed feature hits crews and vendors directly, because there is nothing “down the road” to replace it. Offices are folded into warehouses, rental houses sublet their own space, and experienced freelancers quietly leave the business.

In that context, AI is already being sold as a way to “stay competitive”. The promise is familiar: with the right tools, a small market can match the speed and scope of bigger players, turn around bids faster and deliver more work with fewer people. But competitive against what? Other countries are cutting just as hard, offering similar or better rebates, and deploying the same tools. Buyers—the studios, streamers and big independents—are consolidating their vendor lists and looking for partners who can absorb more work with less friction. The danger is that small markets embrace AI primarily as another round of self‑imposed cost‑cutting in a game where they don’t control the rules.

This is where my own vantage point colours the way I see the hype. I’ve spent more than two decades watching Iceland be “the extreme location that somehow works”: building shoots on glaciers and in winter storms, helping foreign productions navigate permits and politics, and eventually founding Polarama and Polarama Greenland to make that work less chaotic. I didn’t start Massif or AIMgr because I wanted fewer people on set; I started them because I was tired of watching good crews waste days chasing information that should have been at their fingertips. The goal was to digitise the boring parts, not to pretend the work itself could disappear.

The Real Question: Who Is AI For in a Shrinking Industry?

All of this leads to a simple but uncomfortable question: in an industry that is already shrinking, who is AI really for?

One path—call it Path A—is to use these tools to give the same people better conditions. That means applying AI to remove friction from schedules and communication, reduce overtime, automate pure paperwork, and make it easier to keep continuity, safety and creative intent intact. In this scenario, a location manager gets a cleaner briefing pack, a production office sends fewer contradictory call sheets, and an editor spends less time hunting for shots and more time shaping the story. AI becomes another layer of infrastructure that supports human decisions instead of replacing them.

The other path—Path B—is to use AI to deliver the same output with fewer people. In this version of the future, concept artists, assistant editors, junior coordinators, marketing teams and even parts of VFX and sound are thinned out, while the remaining people inherit more responsibility, longer to‑do lists and the stress of supervising systems they didn’t design. Reports on the early “AI era” in Hollywood already describe this dynamic: value created by automation being captured at the corporate level, while crews feel only precarity and speed‑up.

The available data suggests we are not automatically walking down Path A. The Animation Guild’s 2024 “Future Unscripted” study found that around three‑quarters of surveyed entertainment companies said generative AI had already enabled them to eliminate, consolidate or significantly redefine roles. Combined with projections of more than 200,000 US entertainment jobs being affected by AI‑driven changes by 2026, it is hard to argue that the default use case is “same teams, less burnout”.animationguild+1

That is why I keep coming back to what I’ve been calling the Budget Test. In a pressured industry, every AI initiative should have to answer at least two questions in plain language:

Does this materially improve the schedule or the budget in a way we can document? If you can’t show shorter shoots, fewer overages, fewer emergency days or lower all‑in costs for the same quality, then you are dealing in anecdotes, not transformation.
Does this keep more sustainable work and learning inside the ecosystem, or drain it? If a tool saves money mostly by removing junior roles, offshoring experience or pushing more unpaid labour onto freelancers, then whatever it does for the spreadsheet, it is weakening the long‑term health of the industry that spreadsheet belongs to.

AI is not neutral on these questions; it will flow wherever the pressure pushes it. The point of this series is not to decide whether AI is good or bad in the abstract, but to be explicit about which path we’re choosing—and who we’re choosing it for.

What Needs to Be Measured Before We Buy Any More Tools

Before anyone in this business talks about “AI transformation”, we need something much more boring: a clear picture of where our time and money actually go right now. On most shows, that picture barely exists. We have line items and cost reports, but we don’t systematically track how many days are lost to approvals, how many evenings vanish into last‑minute rewrites, or how often pickups and reshoots quietly eat the contingency. Without that, any claim that a tool “saves 20%” is just a guess with nicer branding.

The same blind spot exists around learning. Every production has tasks that are, in practice, training grounds: assistant work in editorial, basic location research, script formatting, continuity notes, simple VFX fixes. Those jobs are how people learn the language and pressure of the work. If we don’t identify them, we risk automating them away first, then wondering why there are no mid‑level editors, coordinators or supervisors five years from now. The short‑term saving comes directly out of the next generation’s apprenticeship.

We also need to know which functions are already at breaking point. Ask location teams that have been doing three scouts’ worth of work with two people, or post supervisors trying to wrangle ever‑more complex deliveries with the same headcount. In those places, the right automation might genuinely prevent burnout. In others, piling “AI” on top of an already fragile process just adds another layer of confusion and unpaid learning.

So the checklist, before we buy any more tools, is painfully simple:

Map delays and overages: where do days and euros actually leak out of the schedule and budget?
- Mark the training ground: which tasks are where juniors really learn the job?
- Flag the red zones: which teams are already stretched to the point where one more experiment could break them?

Without that baseline, every AI efficiency claim is free to target the most vulnerable work, not the least valuable work. We may end up accelerating the wrong parts of the process and cutting away exactly the places where people enter and grow.

The rest of this series is an attempt to make that baseline more concrete. In Post 4, we’ll go into pre‑production as the cleanest testing ground, because that’s where the math should be easiest to track. In Post 5, I’ll share honest tool tests against real scripts and constraints. Later on, we’ll come back to jobs and training: what happens to apprenticeships, to crews, to the long‑term health of the craft.

If we pretend everything was fine until AI arrived, we’ll choose the wrong tools and blame the wrong technology. If we face the cracks that were already there, we might use AI to fix a few of them instead of widening all of them at once.