9 AI Production Mistakes Creative Agencies Keep Making (and How to Avoid Them)

Every one of these is avoidable. Most of them come from the same root cause: knowing the tools without knowing the craft — or the reverse.

Agencies are past the "should we use AI" debate. The question now is why so many are getting burned by it — blown budgets, embarrassed clients, deliverables that looked great in the deck and fell apart in delivery. Having watched this from the production side for a while, the pattern is clear: the mistakes are consistent, predictable, and almost all preventable.

Here are the nine we see most, and what the agencies who've stopped making them do differently.

1. Showing Clients Concepts Production Can't Deliver

The pitch deck features a generated frame with impossible lighting, a location that doesn't exist, and a camera move no rig can make. The client falls in love. Now you get to choose between an expensive VFX rescue or an awkward walk-back.

The fix: put a camera-literate person between the generation session and the client meeting. Every frame gets one question — can we actually make this, on this budget? We've written honestly about AI's limits on real shoots; internalizing those limits before the pitch is what keeps the pitch honest.

2. Skipping the Craft Pass and Shipping Artifacts

Warped hands. Text that almost says something. A logo that morphs between frames. These artifacts are invisible at 2 a.m. on deadline and glaring in the client's boardroom. Once a client spots one, they start looking for them in everything you deliver — and the audience does too.

Every AI-touched deliverable needs a human craft pass: an editor, retoucher, or compositor whose explicit job is to catch what the model got wrong. Not a glance — a pass, budgeted and scheduled.

3. Using AI for the Trust Moments

Some shots carry the brand's credibility: the founder's face, the customer testimonial, the product hero shot a purchase decision rests on. Generating these is the fastest way to spend a client's trust. Audiences have sharpened their detection instincts, and being caught faking an authenticity moment reads as deception, not efficiency.

• Faces in close-up: shoot them
• Product accuracy shots: shoot them
• Anything labeled or implied to be "real people, real results": shoot them

AI belongs in the environments, variants, and volume around these moments — a distinction that sits at the heart of whether AI content is brand-safe.

4. Ignoring Rights and ToS Differences Between Tools

Midjourney, Runway, Kling, Veo, Sora, ElevenLabs — each has different commercial-use terms, different indemnification postures, different rules by subscription tier, and different answers about training data. Agencies routinely treat them as interchangeable. Client legal departments do not.

Map every tool in your pipeline against every client contract before generated assets ship. Some clients will bar specific tools outright; better to learn that in onboarding than in discovery.

5. Underestimating Cleanup Time

The generation took twenty minutes, so the deliverable was quoted like it takes an hour. Then came the cleanup: frame-by-frame artifact fixes, continuity repairs, color matching, upscaling. The projected savings quietly evaporated, and the project lost money with extra steps.

Rule of thumb: budget cleanup as a first-class line item, not a contingency. Generation is the cheap part; making generated material shippable is skilled labor, and it's the part clients are actually paying for.

6. Building the Whole Pipeline on One Tool

The team that went all-in on a single video model last year has rebuilt its workflow twice since. This market shifts monthly — capabilities leapfrog, pricing changes, terms change, tools disappear. One-tool dependence turns every vendor announcement into an operational emergency.

Stay tool-agnostic: build your pipeline around stages (concept, generation, edit, finish) rather than around products, and keep a working knowledge of the field. Our regularly updated look at the AI tools worth knowing exists because the answer to "which tool" genuinely changes quarter to quarter.

7. Pricing AI Work Like Traditional Work

Some agencies quietly bill AI-accelerated deliverables at legacy day rates and pocket the difference. It works until the client learns what the workflow actually was — and they always learn — at which point the conversation stops being about price and starts being about honesty.

Reprice transparently. Charge for judgment, direction, and the craft pass rather than for hours the machine saved, and share some of the efficiency with the client. If you need a grounding in the real numbers on the traditional side, start with what production really costs — then price the hybrid model on its own honest terms.

8. Treating AI as a Style Instead of a Pipeline

"Make it look AI" is already dated — the surreal dreamscape aesthetic is a trend with an expiration date. Agencies that treat AI as a look produce work that timestamps itself. Agencies that treat it as a pipeline — pre-vis, variant generation, localization, cleanup, finishing — get leverage that's invisible in the final product.

The best AI-assisted work doesn't announce itself. It just ships faster, in more versions, at a margin that works. That's how we use AI: as infrastructure, not as aesthetic.

9. Going AI-Only or AI-Never

The biggest mistake is a fork with two bad ends. The AI-only shop can't deliver the trust moments, can't vet feasibility, and produces work with a ceiling. The AI-never shop protects its craft reputation right up until it loses the pitch to a team that showed moving frames and quoted half the versioning cost.

The real gap in the market isn't tools or cameras — it's the expertise to run both on a single job. Teams fluent in generation and in lighting, lenses, and set logistics make each side better: the craft keeps the AI honest, and the AI multiplies the craft. That's the whole argument for working with a hybrid production partner — one team accountable for both halves of the pipeline, so nothing falls in the seam between them.

The Bottom Line

Read back through the list and notice the pattern: almost every mistake happens in the gap between AI fluency and production craft. Frames nobody vetted for feasibility. Artifacts nobody with an editor's eye caught. Cleanup nobody who'd done it before budgeted. The agencies avoiding these mistakes aren't the ones with the best prompts or the best cameras — they're the ones with both skill sets at the same table, on the same job.