Full analysis

Decision Council — Briefing Mode

Step 1 — Frame

The story: A creative-agency CEO makes the practitioner's case that generative AI is a productivity tool, not an idea engine — it can't originate concepts, can't carry emotion, and (the sharpest operational point) can't export editable files you can rework in post. She predicts experiential and brand work grow as AI commoditizes lower-funnel content.

What's actually being decided for the reader: Nothing urgent. There's no platform change, no deal, no regulation. The real question for an ad-tech operator is narrower: should the "AI can't do real creative yet" thesis change how I price, position, or roadmap anything in the next 12 months?

• Reversibility: N/A — this is a read, not a move. Any bet it informs is Type 2 (easily reversed).
• Forcing function: None. The only clock is model improvement, which moves on its own schedule.

Verdict up front: low direct impact. This is a creative-craft conversation, not ecosystem news. But two threads are worth a working executive's attention — the editable-output gap (a genuine product wedge) and the funnel-shift prediction. Four personas move that analysis.

Step 2 — The Council

The Engineer The "consistency errors" complaint — clean-shaven man grows a beard by second seven, a third hand appears, Coke's truck gains and loses wheels — is real but dating fast. Temporal consistency is the single most-attacked problem in video models right now; the half-life of these specific failures is months, not years. Plain version: the exact glitches she's mocking are the ones every lab is racing to fix. But her file-format point is different and durable: these tools output flat video, not layered, editable project files. That's not a model-quality problem, it's a pipeline problem — and it's where the real friction lives.

The Operator The editable-output gap is the only line in this episode that survives contact with a real workflow. An editor can drop an AI clip into a timeline but can't re-grade, re-time, or fix one element without regenerating the whole thing. Plain version: you can't open the hood and tweak — you can only hit "redo" and hope. That breaks how post-production actually works. Whoever ships AI video that exports clean layers (PSD, project files, alpha channels) wins the agency pipeline, not whoever has the prettiest demo reel.

The Customer (Brands / CMOs) The Coca-Cola wheel-count story is the one that travels. A flagship brand shipped a flawed AI spot and got publicly mocked — that's the cautionary tale every CMO now cites in a roomful of vendors. Plain version: one bad AI ad from a giant brand made every marketer nervous about being next. For agencies, this is permission to keep charging for human craft. For AI-creative vendors, it means the buyer's bar isn't "good enough" — it's "won't embarrass us." Reputational risk, not cost, is gating adoption at the high end.

The Market Analyst Treat the predictions as weak signals, not theses. The funnel-shift call — AI commoditizes lower-funnel/demand-gen content, so experiential and brand awareness become more valuable — is plausible and directionally aligned with where money already drifts when production gets cheap (scarcity moves to what can't be automated). Plain version: when everyone can crank out cheap ads, the expensive, can't-fake-it stuff gets more valuable. The Claude-over-ChatGPT aside is a tiny but real data point: model-switching among creative practitioners isn't sticky to OpenAI. Don't overweight any of it — sample size is one CEO.

Step 3 — Tensions

1. Engineer vs. Customer on the glitches. The Engineer says the consistency errors are temporary and nearly solved; the Customer says they've already done lasting reputational damage at the brand-safety level. Both can be true — the tech catches up while the buyer's trust lags behind it.

2. "AI can't originate ideas" vs. the durable moat. Michnowicz stakes her position on ideation ("if AI can get to that, I'm scared"). The Operator and Engineer both think that's the wrong hill — the defensible gap isn't the idea, it's the editable production pipeline. One is a comforting belief; the other is a buildable wedge.

Step 4 — Synthesis

What this hinges on: two of her claims are fragile, one is solid.

• "AI can't originate ideas" and "AI can't do emotion" are current-state observations dressed as permanent truths. They're the kind of claim that ages badly — bet against them as a long-term moat.
• "AI output isn't editable" is a structural workflow gap, not a quality gap. That one is real today and is the most actionable thing in the episode.

Which way the council leans: This is a skip for ecosystem news. But for two specific readers it's worth a beat:

• If you build or sell AI-creative tooling: the editable-export problem is your roadmap priority and your differentiation story. "We output layered, editable files" beats "our model looks better" with the agency buyer. Solve the pipeline, not just the pixels.
• If you run an agency or a publisher's branded-content studio: the funnel-shift prediction is a reasonable hypothesis to position against, not bet the P&L on. Lean into the experiential/cinematic/authentic high end as the part AI commoditizes slowest — but don't assume the ideation moat holds past the next model cycle.

What to verify before acting on any of it: sample size here is one agency CEO with an obvious incentive to argue human craft is irreplaceable. Pressure-test the editable-output claim against the current versions of the major video tools (it's improving) and watch whether the "experiential grows" call shows up in actual 2025 budget allocations before treating it as a trend.

My view: Low-impact episode, one durable nugget. The editable-file gap is the only thing here a product or strategy leader should write down. Everything else is a practitioner's confidence that will erode on the model-release schedule.

What did we miss? Is there a persona we should add for this specific decision? A Top Rep could sharpen the GTM angle — how an AI-creative vendor actually closes an agency skeptic in the room — if that's the reader you care about.