In the last twelve months, 65.3% of agencies had a client pull work in-house (Basis, 2026). 32% of brands now expect to handle nearly all their creative in-house within the year, and another 23% expect to bring at least half of it in (Adweek, 2026). Read one way, that is a decisive move in-house. Read with any history, it is a pendulum that has swung before, and always swung back.
This time it does not swing back. The mechanism that forced every previous reversal is gone. Here is why, and what it means for anyone hiring in creative and media.
Key Takeaways
- In-housing is not new. Creative and production work has cycled between in-house and outsourced for decades, and every previous return reversed
- It reversed for one reason: in-house capacity was fixed, so volume spikes forced work back out to agencies
- AI removes that reason. Output no longer scales with headcount, so the spike stays in-house
- The brand-trained model is an asset that compounds inside the building, which makes going back out uneconomic
- The durable hiring demand is brand-side, for the specialists who can run the model and the pipeline
The cycle is old, and it always turned back
Production has moved in-house and back out for as long as brands have made work. Advertising ran in-house early, then the agency model owned the craft for decades, then in-housing resurged after 2008 on cost control, data ownership and programmatic transparency. The same pattern runs well beyond marketing. Manufacturing and IT offshored aggressively, then reshored once the hidden total cost of ownership, quality gaps and IP risk caught up with the labour saving (CBS reshoring research).
Every turn was a bet on the same trade. Outsourcing bought cost arbitrage and the room to flex to volume. In-housing bought control, speed, IP and institutional memory. The balance tipped, the hidden costs surfaced, and the pendulum came back. Cyclicality has been the norm, not permanence. “It is happening again” is not, on its own, interesting.
What always forced it back out
One mechanism drove nearly every reversal. In-house capacity was fixed and scaled with headcount. Bring production inside and you took on the people, the overhead and the idle time between peaks. When volume spiked or a specialist need appeared, the owned team could not flex, so the work went back out. Agencies existed, in large part, to absorb the spikes an owned team could not. Remove that single constraint and the cycle changes shape.
Why the reversal isn’t coming this time
Two things changed the underlying economics, and together they remove the reason the pendulum came back.
The first: output no longer scales with headcount. A small senior team working with generative tools now produces volume that used to need an agency roster, and the tool is already in nearly every hand, with 76% of marketers using at least one form of AI (Salesforce, 2026). The demand it feeds is climbing just as fast: digital video now takes more than 60% of all TV and video spend for the first time, growing faster than the ad market overall (IAB, 2026). The historical trigger for outsourcing was volume outrunning fixed capacity. That is exactly what AI now absorbs inside the building. The spike no longer forces the work out.
The second has no precedent. The brand-trained model compounds in-house. Every previous swing out bled institutional memory, and the brand rebuilt from scratch. This time the model is the memory, and brands are building their own rather than buying it: 85% of companies expect to customise AI agents to fit their own business (Deloitte, 2026). In June 2026, Adobe launched Firefly Foundry, letting enterprises build private AI models trained on their own intellectual property (Adobe, 2026). The institutional memory used to walk out the door with the agency. Now it stays, and it grows.
The physical layer is moving in too. Real-time engines, virtual production and generative video that once demanded a specialist facility now sit inside brands, and the market for that kit is growing around 16% a year (Mordor Intelligence, 2026). We covered the early signs in how creative tech is quietly building its AI capability.
The brands are already pulling creative in
The talent is moving with the work. 54% of agency leaders say they are likely to start their own firm within two years (Adweek, 2026), and roughly 40% of agencies ran layoffs in the past year as clients pulled work in (Basis, 2026). Much of that capability is landing inside brands rather than back in the holding companies. A brand studio is no longer a design team with a nicer name. It is a production house with a shooting style, a video pipeline, and increasingly a model of its own.
The specialist tail still goes out
Permanent does not mean total. Specialist, spiky, prestige craft still goes out, and should. Early, high-volume production moves inside; post-production, VFX and one-off specialist work stays with the studios that live and breathe it. The pendulum has not been abolished. Its resting point has moved, and it has moved toward in-house for core, high-volume production. The one counter-risk worth naming is that the next thing outsourced becomes the model operation itself, run as a managed service. That does not undo the core shift, for the same reason as before. The compounding brand memory sits in the owned model, and that is the expensive thing to give back.
The roles move brand-side, and get harder to fill
When production moves in-house for good, the roles do not disappear. They move brand-side, to the people who can run the model and the pipeline. The pull is measurable. Jobs that require AI skills are growing roughly eight times faster than the market overall, up 69% against 9%, and workers with those skills now command a 62% wage premium, up from 57% a year ago (PwC 2026 Global AI Jobs Barometer, from more than a billion job ads across 27 countries). Owning that talent now beats renting it.
The scarcity is sharpest where media production and AI meet. The people who can run a brand-trained model, direct real-time production and make the judgment calls sit at an intersection no single training pathway produces. It is the role most in demand and slowest to fill, and many of these titles did not exist five years ago, a gap we mapped in six emerging AI creative roles. The brands getting it right hire for judgment and craft first, then AI fluency, a point we made in what the best AI talent sees.
Own the core, flex the spikes
Treat build versus buy as a capability decision, not a cost one. Own the work that is core and repeatable, the always-on production that benefits from an owned model and institutional memory. Flex the specialist, one-off craft to a partner.
| Approach | Best for | The trade-off |
|---|---|---|
| Own it (in-house studio plus a brand-trained model) | Core, always-on, high-volume production | Upfront investment in the model and the specialists to run it |
| Buy it (external partner) | Specialist, spiky, one-off craft such as post and VFX | More expensive to replicate internally, plus handoff time |
| Generic AI alone | Fast first drafts on low-stakes work | No brand consistency, and still needs human judgment on top |
The sequence is simple. Audit what recurs, because the same category of asset every week is a candidate to own. Hire the pipeline lead before the volume, so your first hire stands up the model and the workflow rather than feeding it. Test for judgment, not tool lists, because what someone has shipped beats what they have on paper. Keep a partner for the spikes, because the winning model is hybrid, and always was.
The shift worth staffing for
Creative work has cycled in and out of the building for decades, and every previous return reversed. This one does not, because the reason it always reversed is gone. Output no longer scales with headcount, the brand-trained model compounds inside the studio, and production no longer needs a specialist facility outside it. The firms that treat the shift as permanent will staff for it before the market catches up. The ones still briefing creative out will bid for the same scarce specialists everyone else wants, and pay a premium for the wait.
